T03 3-4

Contents
Telektronikk Network Planning

Volume 99 No. 3/4 2003 1 Guest Editorial
ISSN 0085-7130 Terje Jensen
Editors:
3 A Tale of a Technical Officers Day
Ola Espvik
Tel: (+47) 913 14 507
Terje Jensen
ola.espvik@telenor.com
9 Network Planning Introductory Issues
Terje Jensen
Per Hjalmar Lehne
Tel: (+47) 916 94 909
47 Optimization-Based Network Planning Tools in Telenor During
per-hjalmar.lehne@telenor.com
the Last 15 Years A Survey
Editorial assistant: Ralph Lorentzen
Gunhild Luke
Tel: (+47) 415 14 125 68 Network Strategy Studies
gunhild.luke@telenor.com Terje Jensen
Editorial office: 99 Portfolio Evaluation Under Uncertainty

Telenor ASA Ralph Lorentzen
Telenor R&D
NO-1331 Fornebu 110 Forecasting An Important Factor for Network Planning
Norway Kjell Stordahl
Tel: (+47) 810 77 000
telektronikk@telenor.com 122 Traffic Forecasting Models for the Incumbent Based on
www.telenor.com/rd/telektronikk New Drivers in the Market
Editorial board:
Kjell Stordahl
Berit Svendsen, CTO Telenor
128 Planning Dependable Network for IP/MPLS Over Optics
Ole P. Hkonsen, Professor
Terje Jensen
Oddvar Hesjedal, Director
Bjrn Lken, Director
Graphic design:
Design Consult AS (Odd Andersen), Oslo
163 Terms and Acronyms Network Planning
Layout and illustrations:

Gunhild Luke and se Aardal,
Telenor R&D 177 Telektronikk Index 2001 2003
Prepress and printing:

Gan Optimal as, Oslo
Circulation:
3,100
Guest Editorial
TERJE JENSEN
Chance favours a prepared mind! role of network planning as a basis for manage-
reflect on it ... ment decision support.
Spending some effort on systemising ones situ- Network planning has a range of scopes, as
ation allows for a revelation of ones strengths described later in this issue. Several types of
and weaknesses. Moreover, it leads to steps to input and aspects have to be considered. In sepa-
take in order to gain from chances by effi- rating the time frames it is important that strate-
ciently incorporating means into ones portfolio. gic, tactical and operational means are har-
Spotting opportunities and smoothly turning monised, which implies a need for good coordi-
Terje Jensen these into profitable operations is a key issue for nation between the commonly different groups
terje.jensen1@telenor.com any actor. As the market matures and the safe/ managing the different scopes. A consequence
traditional services and systems are deployed, of this is that the strategic plans have to be con-
new products and systems are required to main- nected to current operations in order to show
tain more profitable operation. In particular as effects. This is one example that actions result-
the number of competing actors grows it is ing from planning would be organised along a
important not to lose out on the chances that turn time axis and be reflected in the organisation.
up. Not every chance is to be seized though;
only the ones that are expected to strengthen Considering the operators set of interdependent
ones situation. systems, products, traffic flows and customers, it
is essential to have a methodical support of the
So how can network planning assist? First, in planning exercises. Covering all combinations of
order to carry out the planning the actual situa- options, the tasks frequently become too tedious
tion must be characterised, hence describing the for mere manual analyses. This also assists in
overall situation to the operator. This also out- assessing consequences and selecting better
lines strengths and weaknesses. Second, strate- actions for urgent questions, like responding to
gic planning commonly applies scenario work competitors product rollouts, sudden traffic
that captures the possible future an operator may increases, offers from vendors, and so forth.
experience. The scenarios are characterised by
a number of factors attached with uncertainties. More than 125 years after Alexander Graham
These factors should therefore be particularly Bells invention in 1876, few people oppose the
monitored as time evolves. Third, relating poten- notion that the electronic communication means
tial target states with the current situation reveals is one of the major indicators of a communitys
Front cover: a roadmap describing specific events that may welfare. This shows the essence of having ade-
Network Planning
happen. These events may well reflect points quately operational telecommunication networks
The overall basis for implementing where decisions are to be made. Naturally, sev- and corresponding systems in a nation. The
a network plan is a global techno-
economical optimal network solu-
eral factors have to be evaluated when preparing number of people who are aware of the opportu-
tion created through an optimisa- for the decision, and these are also part of the nities and are able to put the practical solutions
tion process. Such a global pro- network planning. Fourth, simply discussing and into operation grows in importance as more can-
cess involves a number of local
disciplines and problems.
elaborating the plans is likely to reveal further didate solutions and arrangements are faced and
opportunities. That is, gathering a number of competition levels fluctuate. Hence, the speedy
The artist Odd Andersen visu-
alises the iterative optimisation individuals from different departments helps to technological, market and service changes make
process as a spiral reaching inform several profitable ideas. it more challenging for a network planner to
wards to a desired target. In the keep ahead of efficient network solutions. The
same manner the local processes
have to find their optimal solutions The ultimate outcome of a planning exercise is convergence of telecommunications, information
given global constraints. The local quantified numbers related to specific network systems, broadcasting, user devices partially
processes are visualised as minor solutions. However, a mixture of qualitative and fuelled by transition from complexity in hard-
spirals interacting with and
adjusting the path of the global quantitative evaluations is included in the full- ware to software requires that the network
process. This myriad of adjust- blown planning. One example of a fruitful mix- planning methodology is flexible and able to
ments goes on until an acceptable ture is to apply qualitative approaches for the consider the holistic view and take into account
global optimum is achieved. To
illustrate the new plans depen- wide set of options that might appear, while a future optional migration tracks.
dency on logical and physical net- more limited set is selected for the calculation
work structures already present, exercise. The cases will also be taken into The drivers for network planning include tech-
Odd Andersen locates the optimi-
sation process on top of the exist- account when conducting a number of what-if nology, markets, business and customer service.
ing band of possibilities. evaluations. This also leads to the fundamental It is the financial turmoil in the telecom industry
Ola Espvik, Editor in Chief that has produced the latest changes. There is a
Telektronikk 3/4.2003 1
renewed interest in cost reduction and maximis- The final article returns to one of the central
ing use of resources which increases the need for questions for backbone networks; that is, how to
network planning. However, this also comes at a efficiently utilise the capacity of transport net-
time of great technological change with IP-based works.
architectures and wireless access producing a
great potential for new applications, and there- As stated above, a full coverage is beyond the
fore additional planning challenges to be solved scope of any single issue. The main objective of
and there is no sign that the pace of change is this issue is to provide insight into systematics
slowing. and methodology applied for network planning.
This also points to the need for awareness
The network planning area is too wide to be cov- regarding adequate planning competence, defin-
ered by a single issue. However, a number of ing relevant planning tasks and continuing
aspects are treated in the following articles. The assessment of input data and surveying the criti-
first article provides a survey of network plan- cal events revealed through the planning work.
ning, relating network planning to management/ The wider scope of planning is taken on describ-
financial topics and the technical questions to be ing the need for on-going activities and links
treated. Going directly to the core of the plan- between different planning tasks. In particular,
ning task a survey of between one and two linkage between long and short term planning, as
decades of implementing planning tools is well as relations between different systems must
described in the subsequent article. be obeyed.
The next area considered places more weight on In order to provide the material in this issue of
the strategic perspective. This is described by a Telektronikk, interesting discussions with a wide
survey article as well as a more in-depth descrip- range of individuals have been appreciated. The
tion of risk considerations. Gone are the days interest for the questions addressed during pro-
with never-failure as the sole objective with jects in later years clearly shows that network
less focus on delivery time-scales and solution planning activities engage most people involved
cost. Risk management is now a key as technical in electronic communications.
and commercial risks are balanced against the
need to meet market opportunities on time and in Enjoy your reading!
accordance with budget.
The essentials of markets must be understood as

it profoundly affects network planning objec-
tives. This is also documented in articles on
forecasting, describing methodologies and
results for obtaining estimates of number of
customers and amount of traffic.
2 Telektronikk 3/4.2003
A Tale of a Technical Officers Day
TERJE JENSEN
Imagine being a technical officer or maybe you are? A long list of questions are raised about how to
get the highest return by smart spending of the companys money. In a broader perspective everyone
involved in network evolution is faced with similar questions; how to organise the systems and system
management and how to allocate resources to achieve company goals. An intuitive statement, however,
is that the choices made have more dramatic effects when made at a higher level in the company. On
the other hand, every function must work in a coherent way to allow for flexible, rapid and efficent
service provisioning.
The following imaginary story tells about a workday of a technical officer, called TO. So, let us follow
this imaginary technical officer during a day to see how various aspects of network planning could be
Dr. Terje Jensen (41) is utilised to ensure effective company operation in the shorter and longer term.
Research Manager at Telenor
Research and Development. In
recent years he has mostly been
engaged in network strategy Introduction however, traffic can mostly be directed more
studies addressing the overall
Bleep bleep, a message arrives on TOs freely than what is accepted on the road.
network portfolio of an operator.
Besides these activities he has mobile. The market survey group has spotted a
been involved in internal and competitor launching a new product. Immediate Approaching this in a systematic manner, the
international projects on network
question: How do we respond to that? Well, questions can be arranged along a time axis. Then
planning, performance model-
ling/analyses and dimensioning. our technical officer, we will call him TO for the number of options will likely grow with time.
terje.jensen1@telenor.com short, receives the news in a relaxed manner. One pictorial model for this is to think of the
Having gone through a set of possible scenarios, future options as constructing a cone a wider
the one emerging has already been examined. set of outcomes and actions are available in the
And it turned out that the launched product does longer time horizon (see Figure 1). On the other
not allow sufficient profit margins compared to hand, one may also have a clearer picture of the
the ones already offered. All calculation results, major trends dominating the picture. Hence, by
internal as well as jointly with others, showed shifting the scope in the longer term, the choices
that this product will disappear from the market to be made may be clearer on a general level
again or the competitor will experience when considering the technical areas. Besides,
increasing loss. So a quick messaging reply was several other areas have to be taken into account
issued: Dont worry stay happy and have a even though technical decisions have to be made.
look at the strategy document on scenarios point Such areas include financial means, customer
6.4. Well, being prepared is rarely a disad- requests, competitors, regulatory issues, etc.
vantage, TO thinks as he continues his break- Factors from these areas may well inspire addi-
fast. Soon he receives a reply, Thanks corre- tional actions relating to technological choices.
sponding media action is under preparation.
Well, a nice start to a sunny day, TO thinks, Still, although there are a lot of uncertainties a
pouring another glass of milk. number of decisions have to be made. In order to
ensure a robust company portfolio, several strat-
On the way to his office, TO happened to ob- egies could be pursued, although this must be a
serve the cars changing lanes without signalling, clear company choice and not simply something
causing sudden breaking and changing speeds. that happens because of lack of coordination.
Like most systems involving humans or uncoor- Hm, this is almost on a philosophical level,
dinated instances efficiency measured on the TO reflects.
individual level differs from that measured on
the overall system level. In this case a single Before Morning Coffee
driver might believe that he is reaching his desti- Parking his car around 07.30 our technical offi-
nation very fast, while the total effect on all the cer TO notices that the car parking places are
other cars on the same road may be that they are about to be re-arranged into a common pool. So
all delayed. This could well be applied to tele- far, department-specific places have been allo-
com systems. In fact, the inherent stochastic cated. To TO, with his planning background, this
nature of systems, user behaviour and technical seems like a natural step to take to increase the
capabilities, competitors, governmental bodies, efficiency of the parking area; yet an example of
etc. further adds to the complexity of finding the scale effect loosely described as the big-
which steps to take to achieve company goals. ger the better. Again, some similarities between
Commonly these days, such goals are given as controlling vehicular traffic and telecom traffic
maximising growth profit. In a telecom system, could be recognised.
Figure 1 Depending on Well, you never know where inspiration may
options selected available come from ...
situations in the future may
become wider (top). Going through his e-mails, a few news bulletins
Commonly diverting on mergers and network contracts attract his
recommendations and interest. Keeping an overview of the companies
observations are seen in the dispositions is always a necessity in this fluctu-
shorter term, while longer ating industry. Particularly interesting is the
term observations could well NOW time notice on a telecom manufacturer buying a video
be more harmonised (bottom) generator/storage company. Could this be an-
other sign that content for broadband access will
be provided in the near future? Or is it simply
that the company wants to be involved in more
links of the service delivery chain? So, all this
goes into the memory bank for the on-going
analysis of industry trends and possible use in
discussions like the lunch meeting.
NOW time Before Lunch

Well, prior to earning his lunch rights some
Leaving his car and approaching his office, TO effort should be placed on the following ques-
reads todays lunch offer: As many pancakes as tion: Should one be engaged in UMTS opera-
you can eat for 5 . As he is interested in inspi- tions in Awayland? The current status is that
ration for something to discuss during a lunch operations are running in two of Awaylands
meeting with the secretary to the Ministry of three neighbouring countries (see Figure 2). One
Industry, this is something to reflect on, he reason for the question is that the interactions
thinks. In fact, tariffing models offered by the between the countries in that region seem to be
range of start-ups within the broadband area steadily increasing.
have reached the agenda of the National Assem-
bly as part of the e-society discussions. Flat So, the principal choices are: i) no engagements
rates, their simplicity and options for cross sub- in Awayland, ii) application of UMTS licence for
sidising between customer groups might be one network operation, iii) establishment of a virtual
of the subjects to bring up. Too many pancakes operator simply offering the service to users with-
gaining weight too few carrots to provide a out deploying a network infrastructure. These are
balanced diet, and less alert for the introduction the options available today. However, future
of more services and meeting multinational com- options will become available depending on the
petitors could be another subject to bring up. choice made, see Figure 3. One of todays options
is not to engage in any service offering. However,
this is a decision that can be revised in the future.
Figure 2 Awayland, indicating
neighbouring countries where When the UMTS network track is followed a
operations are in place or are Mobile presence second network deployment phase may be real-
being planned ised after two years and interactions with other
Mobile wireless systems assumed to be intensified after
presence three years. It is also an option to leave the UMTS
Awayland
UMTS network operation track after the second year by
operation? selling ownership of the company.
The virtual operator choice is also accompanied

by future options; after two years of service offer
Merge Virtual
Virtual operator Wir operator
e
Sell/close acc less
ess
Wireless
s access
UMTS UMTS phase 2 Wireles
No access
engagement
UMTS
Opt-out
Figure 3 Options available

in Awayland NOW 1 2 3
Figure 4 Illustrations of
estimated cost and revenue
0 0 0 0 distributions associated with
M M M M each branch of the
0
M Merge Virtual decision map
Wir operator Note: distributions not in scale
Virtual operator e
Sell/close acc less blue = revenue; red = cost
ess
Wireless
ss access
UMTS UMTS phase 2 Wirele
No c c e ss
a
engagement
UMTS
Opt-out
NOW 1 2 3
0 0 0 0 0
M M M M M
one would consider merging with others or sell Facing this fairly simple decision map, all points
the ownership. The same options are also seen and data can be inserted into the same calcula-
after three years, although an additional option to tion. The problem size also allows for using dis-
enhance the operation with wireless access sys- tributions for main variables such as demands,
tems (similar to the UMTS track) is also present. price levels, cost and timing of technical sys-
tems. Naturally, in several cases exact distribu-
As we can see, four different combinations may tions are not known. However, running several
happen: UMTS operation with or without wire- combinations gives a broader foundation for
less access, or virtual operator with or without making the decision.
wireless access. It is also believed that the track
to be followed should be given a period of about Well, starting with the UMTS re-planning
two years before further steps are chosen. How- assisted by tools such as those described in
ever, the actual criteria for making these deci- [Lore03b], TO reflects on the times when no
sions are rather based on market demand, com- planning tools were available every step
petition level, price of equipment, and so forth. required a lot of manual work, much longer
The current estimation shows that the chosen time, and whether or not an efficient solution
tactic can be followed in the next two years. was identified depended a lot on the skills of the
persons involved. Today it is more or less suffi-
Preparing the basis for the current decision to be cient to insert the adjusted data on equipment
made, numbers should be estimated for each of costs. However, as always, some reflections of
the tracks cost levels, income levels and proba- the results are needed to see whether a skilled
bility estimates. Thanks to the harmonised way eye could come up with improvements on the
of preparing and describing the decision basis proposed solutions. After obtaining the new cost
within the corporation, outlined in [Jens03a], TO structure the results are inserted into the overall
does not have to spend any time explaining the picture of cost/revenue for branches in the deci-
questions to be answered in order to prepare this sion map, see Figure 4.
case to the corporate management. So basically,
the task can be separated into problems to be For each branch distributions of the cost and
addressed by a number of groups. However, even revenue are given the amount of money is
though the overall problem could look like a indicated along the x-axis and the probability
tedious assignment, efficient tools have been level along the y-axis. These distributions are
implemented to make most of the calculations. estimated by utilising different means such as
Moreover, the input data is readily obtained from studies on variations in demand, level of compe-
survey activities in the company. Todays duty is tition placing pressure on the tariffs, cost of hir-
to assess some more information for the UMTS ing transmission capacity, and so forth. Other
track. The virtual operator track has already been means include more subjective means as
covered. Meetings have also been held to esti- described in [Stor03a].
mate the likelihood levels. So most of the basis
is complete, and an offer for UMTS equipment Completing these numbers and distributions,
at significantly reduced prices has just been re- an evaluation is executed similar to the ones
ceived, which has to be inserted into the calcula- described for option theory in [Lore03a]. This
tions. returns that the UMTS track would give the
highest return almost three times the level
obtained if the virtual operator track were to be slide show close to the topic to be brought up.
followed. Basically two issues are expected; firstly mis-
conceptions regarding industry trends and glob-
Although this could be run more or less mechan- alisation, secondly the social effect of flat rates.
ically, it is essential to have some additional
thoughts on the case. TOs experience in this Without going into detail in the conversations,
field motivates for further contemplation. Even the pancake example comes in handy for the lat-
though it is an extensive model, there are always ter topic just as a few visual illustrations flash on
some factors that are not fully captured. In this the nearby screen. The documentation prepared
case there is a bill underway to the National providing examples of TOs view on these mat-
Assembly of Awayland proposing to lift the ters is also well received. Nice to have collected
requirement that data on the countrys residents references from industry magazines and various
cannot be stored in a base abroad. In case this is statements you never know when these can be
lifted common bases can be utilised for several applied.
countries, further reducing the cost level. An-
other factor is that a general agreement with an After Lunch
equipment manufacturer may be finalised, also One of the challenges to attack after lunch is
allowing for lower equipment and operation how to further deploy broadband services in the
costs for the network operator. Both these fac- home market? Broadband services have been
tors are to be settled within the coming month. offered for some time, but the overall corporate
Settling these may eliminate some uncertainties, push in this area has not been completely har-
hence TO forwards a recommendation to the monised. An example of the market adoption
management board to postpone the decision. explanation comes to TOs mind, see Figure 5.
On the other hand, the decision to go for a UMTS So, it is essential to have an understanding of
roll-out could be made based on the available where in the life-cycle the different product
results. However, there is no need to push through areas reside, as described in [Stor03a]. A major
such a decision as there is enough lead-time in the challenge, however, is estimating demands for
plans to wait for about three months. products that have not yet been introduced or are
in the initial phase. Prolonging the adoption rate
The complete documentation set together with illustration in Figure 5, phasing out products is
a short note explaining the recommendation is an essential skill. As the presence of an operator
finalised and submitted for the boards meeting continues, there is a tendency that more effort is
next Monday. Well, better be prepared for placed on steadily developing new products and
what may be decided next month, TO thinks corresponding systems than cleansing the exist-
while swiftly going through the e-mails that ing set of products and systems. TO reminds
have appeared during this mornings exercise. himself of the climate a few years ago when
slang expressions such as cash in hand and
Then, its time for lunch burn rate were uttered by several start-ups and
proposed as measures of success. It seems that
Lunch Meeting the companies staying on that bandwagon for
The small delegation from the Ministry appears too long ended up spending money on disperse
in the lobby. As reservations have been made at activities, eventually facing quite an unsound
the restaurant TO guides the group to the portfolio of products and systems. TO almost
reserved seats. Well, perhaps one should avoid laughs out loud when he remembers the situation
the pancakes for lunch although this mornings in a neighbouring country where a company
point could be applied. One could say that the actually had six different products more or less
lunch table is rigged in such a way that a screen addressing the same customer segment and all
is placed close by accidentally running a these products were actually promoted by six
Number of
customers
Figure 5 Illustration of
adoption rate for a Technology Early Pragmatists Conservatives Skeptics Enthusiasts Customer Nostalgies
enthusiasts adopters replaces upgrading
telecom service (cost/benefit)
SDH Figure 6 Imaginary example
Ethernet IP/MPLS SDH
of transport platform life
Ethernet phase locations
IP/MPLS ATM
Optical ATM
Year 2004 Year 2010
different sales representatives holding meetings multipoint configurations, self-management and

with the same customers. Well, some diversifi- so forth, but these requirements are topics for
cation is often good, but too much is surely not another day. During previous exercises on port-
healthy in the long run. folio for the transport platform area, the candi-
dates were organised in terms of which life cycle
So, returning to todays task TOs company phase they were assumed to be in. Moreover,
has been in the telecom market for some decades there was an overall plan describing which of
and a necessary product and system washout these candidate systems to phase out and which
was carried out last year. The main results were belong to a core portfolio in the medium and
plans to phase out products and systems. Besides longer terms. These were also backed by deci-
the direct cost savings obtained when fewer sys- sions of dominating vendors in this area.
tems are in operation, a dramatic increase in Assume the picture is like the one shown in Fig-
product rollout speed is achieved as the develop- ure 6, but besides such illustrations the sizes of
ment and deployment procedures are harmon- these systems must be considered. On the raising
ised. The effect on the sales side is also begin- edge, a systems size grows. Two systems at the
ning to pay back as clear market communication peak of their size, however, may be quite differ-
strengthens reputation. ent in absolute size, e.g. measured by number of
customers, traffic carried, number of nodes, etc.
In order to make decisions on which product On the falling edge it must be decided whether
and system to remove from the list, a portfolio systems should be reconfigured or simply kept
management perspective was introduced, see without changes. This decision is strongly influ-
[Jens03a]. This allows for including several fac- enced by the total cost of the different options
tors in the decision such as profit level and risks. that is both operational expenses and capital
Moreover, both shorter and longer terms were expenses have to be considered. Again, a risk
taken into account. evaluation must be present.
Still, while the overall plan is to reduce the num- As Xrem is a new product we are faced with sev-
ber of systems and products, new ones are also eral challenges when making forecasts as
considered. So today the analysis model is to be described in [Stor03b]. Still, as elaborated in
prepared for a new product group. The main [Stor03a], it is simply a must to have numbers
question is whether or not to introduce this prod- regarding future demands for different product
uct group and in case the product is to be groups. Naturally, different scenarios can be
offered, how to do that. invented accompanied by different demand mix-
tures, see [Jens03b]. To some extent Xrem is
In the overall mature market a new product expected to capture customers from SDH leased
would cannibalise existing products in some lines, ATM connections, Ethernet lines and IP-
sense. Hence, the motivation is to avoid com- VPNs. A nice feature of Xrem is the very low
petitors capturing customers, improving the cost (investments and operation/administration)
production efficiency (reduced costs allowing that a customer has to cover. Therefore, the Xrem
higher margins), appearing as an actor in the demand is expected to grow as the customers
market forefront in accordance with trends, and want to upgrade their current telecommunication
so forth. Then, the evaluations must always con- configuration. Different sources are considered in
sider the alternatives as before. order to estimate demand; interviews with main
customers, discussions with vendors of user
The product groups, called Xrem, can be sup- equipment, examination of traffic growth, finan-
ported by several candidate transport plat- cial situation in different industries, as well as
forms such as SDH, IP/MPLS, Ethernet and background information from analysis agencies
ATM. On this level the functionality require- and trend descriptions.
ments are basically to be able to connect a num-
ber of points together allowing for a range of Applying procedures as described in [Stor03b],
throughput, delay, jitter and dependability a set of forecasts of the Xrem product as well as
requirements. More advanced service features the other competing products is estimated, tak-
are also involved supporting custom-tailoring, ing into account both the number of customers
and the traffic loads. So todays task is to pre- his coffee cup to the dishwasher, TO thinks that
pare some reasoning around the forecasts. Hav- better insight should be incorporated in phasing
ing established a corporate approach for fore- out products and networks/systems. In fact,
casts, TO prepares input for tomorrows meeting clearer decision criteria for how and when sys-
on these issues. Then the forecasts are to be pre- tems should be switched off would greatly
sented and consequences on other product improve the evaluation procedure. As time goes,
groups assessed. This is a significant factor for it is a lot easier to introduce new systems than to
estimating overall profit improvements of intro- get rid of current systems. As TO takes on the
ducing the Xrem product. Other factors are those overall portfolio perspective he thinks that having
expressing the costs both investments and all the systems of various levels of success may in
operational expenses. Having all the various some cases be justified. But in most cases it is
product groups within the corporation, decisions likely that too much attention from the operation
are to be made on the overall level, which is and administration side is spent on less fruitful
whether the overall profit will increase or solutions. Again, it is a question of devoting more
decrease. attention to the real challenges and allowing
progress to be made in these matters without
More challenges are faced because Xrem is a tedious managerial bother. For example, should
new product and limited information is avail- one have a steady argument on when to start the
able. So nearby products are looked at and the dishwasher or which powder to use, the office
diffusion for the demand of Xrem may be esti- would likely run out of clean cups. Moreover, if
mated. Still, the price levels have to be taken the dishwasher should be restarted every time
into account arguing for assessing the product someone returned with a dirty cup, the result of
(cross) elasticities. The result is a product pene- clean cups would also be delayed.
tration ratio for Xrem. Then, factors providing
the effective traffic loads must also be estimated, On his way out TO reflects on the advantages of
as described in [Stor03b]. For each area the having clear lines of duties and result handovers.
number of customer sites, average and peak Efficient network planning tasks from strategy
packet loads, etc. must be found in order to operational matters fit very well into this
to decide on the effective load and hence the picture. Similar principles also apply to other
needed capacity of network nodes and links. areas of business and lessons could be learned
from those areas. Keeping ones eyes and mind
Finding an optimal network design given the open there is always more to learn, TO thinks;
effective load, methods and tools as described in perhaps examples from the food industry could
[Lore03b] can be applied. The result is typically be applied in the telecom industry as well
locations, dimensions and handling of traffic. something to bring up during tomorrows free
Several time periods could be considered, possi- style discussions?
bly running a multi-period optimisation algo-
rithm. However, in several cases, much uncer- References
tainty is attached, and a simpler techno-eco- [Jens03a] Jensen, T. Network planning intro-
nomic approach could be applied (see e.g. ductory issues. Telektronikk, 99 (3/4), 946,
[Stor03a], [Jens03a]) in order to provide a first 2003 (this issue).
assessment of the financial aspects. Treating the
uncertainties more systematically, risk manage- [Jens03b] Jensen, T. Network strategy studies.
ment procedures can be applied as described in Telektronikk, 99 (3/4), 6898, 2003 (this issue).
[Lore03a] and [Jens03b]. It also needs verifying
that realising the Xrem product still fits with the [Lore03a] Lorentzen, R. Portfolio evaluation
overall network strategy, complying with the set under uncertainty. Telektronikk, 99 (3/4),
of decisions to be made. However, in case it 99109, 2003 (this issue).
turns out to be a strong case for realising Xrem,
the overall network strategy should be adapted. [Stor03a] Stordahl, K. Forecasting an impor-
tant factor for network planning. Telektronikk,
Well, well, TO thinks Having all this pre- 99 (3/4), 110121, 2003 (this issue).
defined and an organisation tuned into the same
way of working greatly simplifies the evaluations [Stor03b] Stordahl, K. Traffic forecasting mod-
and planning activities that need to be under- els for the incumbent based on new drivers in
taken. Having demonstrated the swift work pro- the market. Telektronikk, 99 (3/4), 122127,
cedure, the top management have gained an 2003 (this issue).
interest in exporting this to other subsidiaries.
So, there seems to be several options open in the

future, allowing for more features and concerns to
be incorporated in the work procedure. Returning
Network Planning Introductory Issues
TERJE JENSEN
This article gives a brief introduction to a number of network planning issues. Besides several core
technical areas, relations with telecom management and financial topics are also included. However,
the list of issues that could be treated is quite long, and so is the vast amount of material that might be
captured by the network planning title. A key message in this presentation is that network planning is
multi-faceted and has relations with (almost) all other types of activities going on in a network operator.
Hence, it is important to balance the system portfolio insight with skills in planning techniques as well
as the financial means for comparing results.
Considering the required efficiency improvements most network operators are faced with, having an
adequate network provides a crucial contribution. In fact, the systematic descriptions and treatment
Dr. Terje Jensen (41) is involved can allow for immediate improvements and reveal business opportunities.
recent years he has mostly been
engaged in network strategy 1 Introduction threshold to the service-offering arena such that
In its widest interpretation network planning has the power of newcomers may be overwhelming.
Besides these activities he has relations to every activity going on in a network
been involved in internal and operator. However, in a practical organisation, Relations between network planning and product
a number of clear interfaces are defined where development should be clear; the portfolio of
ling/analyses and dimensioning. units within the operator collaborate. This allows networks is used as production means for deliv-
terje.jensen1@telenor.com network planners to be concentrated on network ering a set of products. This implies that features
dispositions both in the short and long term. in the network portfolio give significant guid-
In order to ensure efficient networks, however, ance on the products that are possible. It may
input and results have to be conveyed to other also decide when a product group should be sup-
units. ported or perhaps the product group should not
be supported after all as the market is considered
As found in dictionaries a plan is an arrange- too small, other products will rapidly take over
ment for doing or using something, considered their positions, etc.
or worked out in advance. Typically a plan is
shown by a drawing/scheme. Then, to make a A network planner should take on the holistic
plan is to make preparations and hence consider perspective, considering how a portfolio of sys-
something in detail and arrange it in advance. tems fulfils its overall purpose. This includes
Referring to networks, a general interpretation questions like:
could interact with most other tasks, although
the planning task itself is focussed on how to Which usage patterns will be seen?
evolve the network portfolio managed by an Which services should be offered?
operator. How will the competitors act?
Will any regulatory changes happen?
A fundamental gain from planning is certified Which technologies should be introduced
resource utilisation. An effect of this is lower and which should be phased out?
cost of deploying network elements, for example What personnel competence is needed?
found by running a network optimisation pro- Which tools should be applied to assist in the
gram deciding where nodes should be located, planning process?
nodes should be interconnected and traffic
routed in order to minimise the overall cost. In The level of detail is strongly correlated with the
the longer time scope a number of options will time horizon; a shorter time horizon requires
be looked at, including different trends of what more concrete details and limited scope. This is
the industry will look like, how competitors will illustrated in Figure 1. With a longer time hori-
behave, what customers will ask for, what ven- zon, less attention is given to individual tasks
dors will offer, and so forth. Having a systematic and more is assigned to having an integrating
description of these issues also allows for detect- and holistic view.
ing new business opportunities for a network
operator. That is, chances are revealed, and trig- On a technical level network planning includes
gers/factors for when to go for these chances are designing, optimising and operating telecommu-
described. Besides these upsides, other chal- nication networks. This will also assist decision-
lenges can also be prepared for, such as presence making on migration of network portfolio to
of a disruptive system reducing the entrance maximise benefit for an operator using a com-
Network design Network strategy
Specific implementation knowledge knowledge
knowledge
Network portfolio
strategy
New system/network
installation
Larger network
re-structurering
Specific
work orders
Figure 1 Different time bination of quantitative and qualitative evalua- self, drawing on his experience to choose the
horizons covered by network tion. Commonly, quantitative evaluations are proper set of tools and the proper candidate
planning, relating to different associated with the core of network planning, in descriptions. Having a capable network planner
level of system detail insights, particular for the short and medium term hori- allows for i) strong arguments for future tasks,
design and strategies zons. Key methods included in network planning ii) contribution to development of suitable tools
are: for future tasks, and iii) a presenter of recom-
mended and optional evolution plans to the man-
Market and demand forecasting agement for decision.
Teletraffic-related methods
Economics engineering Relations between network planning and tele-
Operational research and optimisation com management is discussed in the next sec-
Architecture and technology know-how tion, with particular emphasis on financial
issues. This is to acknowledge that most deci-
All these methods are combined in a complete sions are based on financial measures. Various
planning exercise incorporating shorter to planning scopes are handled in Section 3. Then,
longer terms (ref. Figure 1). Considering the Section 4 gives an example related to Internet
planning scopes, in principle the complete sys- traffic engineering. A number of technical meth-
tem portfolio should be taken into account. ods are treated in Section 5 Section 15.
However, some separation into segments be it Finally, a network planning tool from ITU-D
geographical, functional layer or system-depen- is outlined.
dent is mostly carried out in order to limit the
task. Still, it is important to bear in mind that the 2 Network Planning as Part of
solutions recommended must fit into the overall Telecom Management
portfolio managed by the operator. Besides find- Considering different time perspectives on net-
ing efficient architectures, interconnection work planning, there will also likely be interac-
schemes, performance levels, SLA conditions, tions between the planning activities and man-
operational expenses and processes for support agement levels. A possible model is depicted in
systems should be considered. Figure 2.
Having a full-fledged set of tools is essential to A network operator, as an instance of an enter-

be able to fulfil all the planning objectives. The prise, would have a business plan. The business
main resource, however, is the individual him- plan gives the financial indicators enabling the
Technical Forecasts Competition Regulatory
trends Prognoses on
income/costs
Actor
alliances
Long/medium Strategy
term planning Long-term
Operation/ budgets
measurements
Merging and Politicial

Mechanisms/ Short acquistation activities
parameter settings term planning
Budget/year
Yearly
Revised budget
plans
Cash-flow
and finance
manager to evaluate the financial performance of Figure 2 Illustration of relations between network planning and
the enterprise in order to make further decisions. telecom management
In brief, a business plan summarizes the results
of the planning process for projects involved,
including:
Objectives to reach (future revenue, number of

customers, etc.)
Strategic objectives, strategic guidelines
Effort requested (human resources, equipment (on enterprise level)
investments, etc.)
Influence on performance indicators, in partic- Marketing

planning
ular the financial indicators.
Network
planning
In a similar manner as network planning hori-
zons, business plans can be divided into I) strate- Operation
planning
gic business plan (decisions with a longer hori- Human
zon, affecting all enterprise), II) tactical business resources
planning
plan (decisions for particular units/projects, modifications
affecting market segments, system choices, etc.),
III) short-term business plans for management
Business plan Evaluate
control (assisting monitoring of performance,
follow-up of budgets, etc.).
The relations between network planning, plan- Figure 3 Relations between business plans, enterprise strategic objectives and area
ning for other areas and business plans, can be planning improvements through iterations
illustrated as an iterative process, see Figure 3.
Statement Relations Purposes (examples)
Income Net income = revenues expenses to analyze potential profit

is the profit high enough?
Balance sheet Enterprise capital = assets liabilities to analyze the financial structure
how to finance the migration
enough/too much equity?
enough/too much debt?
Cash flow Cash balance = inflows outflows to make payments at every due date Table 1 Elements of
to have the right cash at the right time fundamental financial
indicators
Besides the outer iteration, there may also be
Operational Revenues
expenses (top-line) interactions between the different planning
groups. However, a fundamental message is that
Capital + these have to be coherent in order to efficiently
expenses support the enterprises objectives.
EBITDA The elements in the business plan contain finan-

(Earning before income taxes,
depreciation and amortisation)
cial indicators, models for costs and revenues as
Depreciation well as sensitivity studies. These may be updated
during planning exercises, both parameter ranges
and modelling considerations; through evalua-
EBIT
(Earning before interests tion of effects from activities, change of techni-
and income taxes) cal/organization/etc. solutions, portfolio adjust-
Taxes
ments, strength-weakness-opportunities-threats
analyses, and so forth. As shown in the figure,
iterations allows for steady improvements.
Net income
Figure 4 Income statement It is hard to describe business plans and business
modelling without entering an arena of financial
indicators. All fundamental financial indicators
are carried out with elements as given in Table 1.
Figure 4 illustrates relations between key mea-

Cost of sures related to income. As seen the basic net-
sales work-related costs can be divided into invest-
ment-related (capital cost) and operational
Operating
Total expenses
revenues
(sales/
top-line) Gross Depre-
margin ciation
Income
from
operations Taxes,
interest
EBITDA
EBIT Dividends
Other Net
income income Retained
earnings
Figure 5 Income statement relations between measures
Customers Shareholders Banks, lenders
Sales of services Increase of equity Increase of debt

(operating income) (capital increase) (credit, loans)
Inflows
Cash flow (inflow outflow)
Outflows
Technical Debt
Labour Network
and repayment
costs, and Tax Dividends
administrative with
Training equipment
expenses interest
Figure 6 Cash flow statement Staff Suppliers Service Government Banks, Shareholders
(inflows and outflows) providers lenders
related (rights of way, renting space, power, Depreciation Figure 7 Cash flow calculations
Capital
etc.). Similar illustrations for relations between expenses
central measures are given in Figures 5, 6 and 7.
The indicators mostly used for economics/finan-
cial issues are: Change in
working
capital
Internal rate of return (IRR) Net income
Net present value (NPV) Repayment
Discounted payback period (DPP) of debt
Operating
Net cash flow (NCF) cash flow
Retained
Discounted cash flow (DCF)
earnings
Operating income
Revenue per product group Dividends
New equity,
Using the cash flow calculations for decisions is New debt Net change
illustrated in Figure 8. From that example it is in cash
seen that introducing more services on the com-
mon infrastructure improves the cash flow as
seen for the upgrade project. However, the total
picture for the operation (including effects on
other networks) must also be considered when 3
an investment activity is decided. Similar cash
flow calculations can also be made for other
competition scenarios, e.g. with severe competi- 2 ce
+ voi
tion putting pressure on the price level that can video
be chosen for the different services. rnet +
1 Inte ideo
et + v
Intern
Actions to consider to improve the cash flow for
a project are to 0
Y1 Y2 Y2 Y4 Y5
Increase the number of new customers, sales et

-1 Intern
of connections and product bundles
Limit the churn rate of customers by keeping -2

the loyalty of the present customers
Introduce new services and service bundles in also includes radio base stations. The top layer Figure 8 Cash flows for three
the portfolio in this example is the service logic nodes, for network upgrade options
example service control nodes in the intelligent
Minimize expenses (capital and operational) network architecture, home location registers,
and so forth.
3 Planning Scope and
Life-cycle In ITU-T three planes have been described refer-
Avoiding initiating a too tedious task, the plan- ring to various aspects of the network capabili-
ning scope should be properly defined. Tractable ties. Briefly, the main purpose of each of the
models state that the broader the scope the less planes are:
detail should be considered. The scope can be
limited in several directions, such as the geo- User plane to convey the user information
graphic area, the network layers, the time period, (information from higher layers)
etc. A number of directions are illustrated in Fig-
ure 9. Control plane to control traffic flows and
resource configurations
An example on network layering is shown in
Figures 10 and 11. A finer or coarser separation Management plane to manage network
into layers might be needed depending on the resources, including fault management, con-
objective of the planning exercise. In the exam- figuration management accounting manage-
ple shown in Figure 11 the lowest layer refers to ment, performance management, security
buildings (including power, cooling, etc.) and management.
cables. The next level refers to transport net-
works, such as SDH possibly including wave- The actual distinction between user data, control
length division systems. On top of this are and management might be rather blurred in
placed the networks of switches and routers; this some cases. Any larger network would typically
Figure 9 Directions for By time Long, medium, short term
limiting the network
planning task
By network (horizontal) layer Fibre/cable, wavelength, packet transport, etc.
By network segment Access, aggregation, region, core, end-to-end
By geographical area City X, Region Y, Country Z, etc.
By network technology ISDN, GSM, UMTS, IP, ATM, SDH, etc.
By service group Voice, leased capacity, broadcasting channels, etc.
Strategic, business modelling, design,

By level of detail
parameter tuning, etc.
Combinations of these
for a specific activity
have functions belonging to all these planes, addresses and routing functions must be imple-
although the way functions are implemented mented for all three planes.
varies. An objective is to find efficient complete
solutions and combinations of functions that In DSL forum a three-layer logical/functional
incorporate all needed functionality. architecture has been elaborated as shown in
Figure 10. This architecture is intended as a ref-
In order to locate where to direct the informa- erence for delivery of differentiated and ensured
tion, addresses and routing functionality have to services. Figure 10 shows an architecture for
be present. So, addresses are used to identify a support of media-on-demand and conversational
Figure 10 The three-layer unit/interface, while routing is used to find how services. Three layers are defined: i) network
logical/functional architecture, to direct the information towards the address layer, ii) common enabling services layer, and
(from [DSL-058]) (and the unit/interface it represents). Note that iii) application layer.
A possible set of interactions between the layers Planning results and information are trans- Figure 11 One way of defining
as well as questions to be raised are given in Fig- ferred between management and operational network (horizontal) layers
ure 12. However, the idea behind separation into support systems and network elements/archi- (from [Soto02])
layers is that the layers are examined in series; tectures, e.g. based on measurements or num-
one-by-one. This would naturally not deal with ber of nodes.
the dependencies at the same time as solutions
are obtained for each of the layers and might Description of basic infrastructure information
therefore imply that sub-optimal solutions are considered (location of customer sites, hous- Figure 12 Illustration of
obtained. ing, power, cooling). questions and relations
between network (horizontal)
As an ultimate goal, all the questions should be layers (from [Soto02])
addressed in a single go for network planning.
However, this is not a realistic case beyond the
very simple networks. However, relations
between current state, financial aspects, product Network Layers Planning process
forecasts and technical solutions must be consid-
ered, as exemplified in Figure 13. Note the set of Data flows
procedures and corresponding tools to assist in Service and Control level NE dimensioning
Delay evaluation
conducting the various activities. A variation of
this is also depicted in Figure 14.
Functional Traffic Demand
level Circuit location design
A number of information elements are required Routing optimization
as input and flowing between the different steps
and scopes involved in network planning. Some
BW grooming
samples are: Transport/SDH Route design/optim.
Path protection
Description of network architectures, node

locations, product mapping, etc. resulting Physical layout
from longer term planning is related to solu- Infrastructure and Cable level Cable routing grouping
Nodeduct sharing
tions on shorter terms (commonly carried out
iteratively between the activities).
Figure 13 Integrating plan- Assessment of current situation
ning activities (technical, business)
Product forecasting
(# customers, traffic revenue) Technical
Scenarios
and
for
financial
future
recommend-
events Configuration of network architecture ations
(network portfolio)
Financial and business planning

(incl. organisation)
Coherent procedures and tools
Compliance:
- Services
- Capacities Techno-economic evaluation:
- Availability - Architecture
- Operational constraints - Density
- Capital expenditure
- NPV of full costs Qualifiers:
- Entry costs
- Flexibility
Activities, - Speed to service
areas Selected - Evolution
scenarios - Experiences
Network
Input models
criteria
Mapping Selected
technical network
solutions solutions
Results
Mapping
economic
solutions Recommendations
Actions Mapping
qualified
solutions
Figure 14 Illustration in steps Description of future scenarios regarding Regulatory restrictions on service offerings
towards recommending product forecasting, technical issues, regula- and interfaces/unbundling, interconnection
network architectures tory issues, competition, etc. points.
(from [Soto02a])
Geographical and demographical information Description of existing network (nodes, links,
on households and other customer groups. cables) on locations, capacities, traffic, re-
usability.
Customer segmentation; characteristics, prod-
uct bundles, usage patterns. Financial calculation data including interest
rates, amortization periods, etc.
Product portfolio, covering access-related and
other service types, as well as their require- Available labor force of relevant types.
ments/SLA-conditions.
A number of factors influence the choice of the
Network resource loads from service usage, better network configuration, ref. Figure 15. A
covering all relevant resource types (nodes, general objective is to describe the better net-
links). work solution considering the often opposing
sets of requirements. It is important to note that
Description of technical components available the better solution for one period may not be
for network deployment; price, capacity/per- the preferred option for another period. Nor may
formance levels, functionality. a better solution at all be wanted when both
Demand Cost
Service General Figure 15 Sample of factors
level policy/ and activities influencing the
forecasts estimates
demands strategy
better network solution(s)
Search for
Collecting
network
data
solutions
Recommended
Existing
network
networks
solution(s)
Data Evaluating
processing network
solutions
Limitations Historic
Technical
(financial, views,
capabilities
personnel) trends, etc.
the (historic/) current configuration and a longer Related to applications and customer demands
term recommended solution are taken into are also the management of services and the for-
account. mulation of adequate SLAs. What is needed for
efficient service management? It is essential that
Lead times when ordering new equipment, lim- the service management system works well with
ited work force for installation and testing and so the network and element management systems,
forth can also influence the speed of introducing supporting the following tasks:
new network solutions or upgrade network ele-
ments. Dynamic network representation to handle
new and out-of-service managed resources.
As seen from Figure 15 data handling is needed
at different stages such as: Proactive data and event monitoring to inte-
grate data from a set of diverse vendor and
Collecting data from various sources protocol sources and identify potential prob-
lems.
Adaptation of data to the purpose at hand, e.g.
for input to a calculation program Automated topological and model-based rea-
soning to understand the impacts of events
Use of data when analysing the models as a based on connectivity and configuration.
simplification of the real world case
Cause effect inference to map between the
Evaluation of results/output data from the cal- effects of anomalies and their service impacts.
culations.
Root cause analysis to differentiate between
During all these stages, the data quality must be symptoms of a problem and the true problem.
assessed, including the uncertainty attached and
which values are the most likely ones. In order Operator guidance to assist to prioritising
to do this, insight must be gained into the origin operator actions.
of the data as well as any processing on the way
to model input. Automated testing to verify diagnoses while
minimizing network impacts.
Arriving at an understanding of the applications
and services during the course of planning is im- Automated fault correction to implement
portant. However, a number of sensitivity analy- network reconfiguration to optimise service
ses is typically run for different demand patterns delivery.
to see how robust the network solution is with
respect to these patterns. Some examples on These capabilities are essential for truly effective
applications related to TV and PC are shown in service level management. That is, not meeting
Figure 16. the requirements, most of the event management
TV Focused Services Typical Typical Delay Packet On
bandwidth bandwidth bound loss demand
(upstream)
Broadcast TV e.g. MPEG2 2 6 Mbit/s1) ~1 s 10-5 Yes

High definition TV HDTV 12 19 Mbit/s1) ~1 s 10-5 Yes
Pay Per View and NVOD e.g. MPEG2 26 Mbit/s1) ~1 s 10-5 Yes
VOD e.g. MPEG2 26 Mbit/s1) ~1 s 10-5 Yes
Navigator and EPG (can be locally < 0.5 Mbit/s N/a N/a No
launched and updated in non real time)
Picture in Picture two MPEG2 channels Up to 12 Mbit/s1,2) ~1 s ~1 % Yes
Picture in Browser one MPEG2 Up to 9 Mbit/s1,2) ~1 s 10-5 Yes
Personal Video Recorder PVR replay 2 6 Mbit/s local1) N/a N/a Yes
MPEG2 file off hard disk
ITV TV telephone features < 64 kbit/s < 64 kbit/s < 400 ms ~1 % Yes
(RTT)
TV browser Up to 3 Mbit/s N/a N/a Yes/No
TV e-mail 128 640 kbit/s Up to 3 Mbit/s N/a N/a No
TV Instant Messaging 128 640 kbit/s Up to 3 Mbit/s N/a N/a No
TV Chat 128 640 kbit/s Up to 3 Mbit/s N/a N/a No
TV on-screen notification < 64 kbit/s N/a N/a No
TV interactive games 128 640 kbit/s Up to 3 Mbit/s ~10 ms 10-5 Yes
TV Audio Juke Box < 128 kbit/s ~1 s <1% Yes
Notes:
1) video compression advancements will enable more efficient encoding (1.5 to 3 Mbit/s)
2) more efficient solutions could be available
PC Focused Services Typical Typical Delay Packet On

bandwidth bandwidth bound loss demand
(upstream) (downstream)
High Speed Internet Access (browsing,

IM, Chat, FTP, VPN, access, etc.)
Residential (typically asymmetric) 128 640 kbit/s Up to 3 Mbit/s N/a N/a Yes/No
SME/SOHO (typically symmetric) Up to 6 Mbit/s Up to 6 Mbit/s N/a N/a Yes/No
Server based e-mail as above as above N/a N/a No
Live TV on PC 300 750 kbit/s ~1 s ~1 % Yes
Video on Demand 300 750 kbit/s ~1 s ~1 % Yes/No
Video Conferencing 300 750 kbit/s 300 750 kbit/s < 400 ms (RTT) ~1 % Yes/No
Voice/Video telephony 64 750 kbit/s 64 750 kbit/s < 400 ms (RTT) ~1 % Yes
Interactive Games 10 750 kbit/s 10 750 kbit/s ~10 ms 10-5 Yes
Remote Education 300 750 kbit/s ~1 s ~1 % Yes/No
Figure 16 Samples responsibilities add to the already loaded human include integrated object models of the network
of applications operators. devices and their connection, the services, SLAs
(from [DSL-058]) and customers. It should be able to perform
Mapping network events onto services and what-if analysis for possible customers affected
related SLAs allows for real-time management by a potential failure and be able to generate a
of the business. Key performance indicator can list of all devices used by a customer or a group.
be polled periodically, and trend analysis of the
historical values can be used for proactive SLA A network lifecycle could possibly also be
management, predicting limit violations. As a thought of as several phases interacting. This is
consequence of a violation of an SLA condition, illustrated in Figure 17. Note, however, that the
any actions not just reports can be per- different time scales must also be considered in
formed, allowing corrective measures to be auto- coherence.
mated. The service management system should
4 Internet Traffic Engineering
Exemplifying Network Long term/strategic
planning
The text in this section is based on [RFC3272],
giving an example of how network planning
on the shorter/medium term can be presented.
Short term
As stated there, the main goals of traffic engi-
neering (TE) are to improve performance for
IP-based traffic while still utilising the network Procurement and delivery
resources efficiently. Internet traffic engineering
is defined as Installation and
Re-design commission
Architecture
Services
that aspect of Internet network engineering
dealing with the issue of performance evalua- Customer groups
tions and performance optimisation of opera- Infrastructure
tional IP networks. Hence, measurement, Inventory Network
characterisation, modelling and control of database
traffic are included. Management
Capacity management and traffic management

can be used for the optimisation done as part of Operation
TE. Capacity arrangement includes capacity Service managemen
t
planning, routing control and resource manage-
ment. The resources include link bandwidth,
buffer space and computational power. Traffic
management includes nodal traffic control (e.g. looking at different instances of time to imple- Figure 17 Example of
traffic conditioning, queue management, sched- ment the solutions into the network. network life-cycle
uling) and other functions that regulate traffic
through the network or control access to network The settings described are:
resources. These activities can be carried out Network context; describing the situations
continuously and in an iterative manner. The where traffic engineering challenges are
activities are commonly divided into proactive found. Such situations include network struc-
and reactive. In the former preventive and per- ture, network policy, network characteristics,
fective actions can be found, while corrective network constraints, network quality attrib-
actions would be part of the latter. utes, network optimisation criteria, etc. A net-
work can be represented as a system consist-
The TE actions would operate on various time ing of, i) a set of interconnected resources,
scales; coarse (days, years like for capacity ii) a demand representing the offered load,
management), intermediate (ms, days like and iii) a response consisting of network pro-
for routing control), and fine (ps, ms like for cesses, protocols and mechanisms that carry
packet level processing). the offered load through the network. Several
types of demand classes may be present, simi-
The TE subsystems include capacity augmenta- lar to traffic classes although also different
tion, routing control, traffic control and resource customer types should be taken into account.
control. Input to the TE system would be net- This results in a request for differentiated ser-
work state variables, policy variables and deci- vice. The network resource and the traffic
sion variables etc. The challenge is to introduce handling-related mechanisms do also have
automated capabilities that adapt fast and effi- their characteristics. Figure 18 Settings for
ciently to changes in the network state, while exercising TE activities
stability is maintained. Performance evaluation
is then a critical part of this, to assess the effec-
tiveness of a TE method, to monitor a network
state and to verify compliance with performance
levels. Network context: Problem context:
TE challenges issues addressed
4.1 Settings
A number of settings for exercising TE activities
are identified in [RFC3272]. To some extent
Implementation and
these can also be considered as steps, see Figure Solution context:
operational context:
how to solve the
18. However, considering that TE activities are implementing and
TE problems
maintaining the solutions
carried out continuously the different steps may
time
be active at the same time, although possibly
Problem context; defining the issues that TE methods for abstracting, formulating and solv-
addresses, like identification, abstraction, rep- ing TE problems, v) a set of administrative
resentation, formulation, requirement specifi- control parameters that may be managed by
cation, solution space specification, etc. One a configuration management system, vi) a set
class of problems is how to formulate the of guidelines for network performance evalua-
questions that traffic engineering should tion, optimisation/improvement. Traffic esti-
solve; how to describe requirements on the mates can be derived from customer subscrip-
solution space, how to describe desirable fea- tion information, traffic projections, traffic
tures of good solutions, how to solve the prob- models, and from empirical measurements.
lems and how to characterise and measure the Polices for handling the congestion problem
effectiveness of the solutions. Another prob- can be categorised according to the criteria
lem is how to measure and assess the network i) response time scale (long weeks to
state parameters, including the network topol- months, e.g. capacity planning; medium
ogy. A third class of problems is how to char- minutes to days, e.g. setting routing parame-
acterise and evaluate network states under a ters, adjusting Label Switched Path (LSP)
variety of scenarios. This can be addressed design; short ps to minutes, e.g. packet pro-
both on system level (macro states macro cessing of marking, queue management),
TE) and resource level (micro state micro ii) reactive versus preventive, and iii) supply
TE). Solving congestion is an essential part side (increase available capacity, redistribute
of performance improvement. Handling con- traffic flows) versus demand side (control the
gestion can be divided into demand side poli- offered traffic).
cies (restrictive) and supply side policies
(expansive). Implementation and operational context;
implementing the actual solutions, involving
Solution context; elaborating how to solve planning (e.g. a priori determined actions
the TE problems. This typically means evalua- based on triggers), organisation (e.g. assigning
tion of alternatives, which requires estimating responsibilities to different units and co-ordi-
traffic load, characterising network state, elab- nating activities), and execution (e.g. measure-
orating solutions on TE problems and setting ment and application of corrective and perfec-
up a set of control actions. The instruments tive actions).
relevant include i) a set of policies, objectives
and requirements for network performance These context descriptions may also be looked
evaluation and optimisation, ii) a set of tools upon as gradually getting more precise and
and mechanisms for measurement, characteri- closer to the implementation.
sation, modelling and controlling traffic and
allocation to network resources, iii) a set of 4.2 TE Process Model
constraints on the operating environment, net- A TE process model is presented in [RFC3272]
work protocols and TE system, iv) a set of and depicted in Figure 19 as an iterative proce-
Figure 19 TE process model quantitative and qualitative techniques and dure consisting of four main steps.
The first phase refers to definition of control

policies. These would typically depend on a set
of inputs, like business model, network cost
Business model Operating constraints Optimisation criterion
structure, operating constraints, utility model
Network cost Utility and optimisation criterion.
structure model
The second phase involves measurements in
order to assess the conditions in the network;
network state and traffic load.
Define relevant control policies
The third phase consists of analysing the net-
work state and characterising the traffic load. A
number of potential models and analysing tech-
Measurement from operational network
niques may be relevant, for instance also looking
at the timely and spatial distribution of the traffic
load.
Analyse network state and characterise traffic load
In the fourth phase, performance optimisation is
done. This includes a decision process selecting
Performance optimisation of the network and implementing a set of actions. Actions may
work on the load demand, distribution of load
and network resource configuration and capac-
ity. This may also initiate a network planning in Routing is a central component in efficient han-
order to improve network design, capacity, tech- dling of traffic flows in an IP-based network.
nology, and element configuration. When introducing a number of service classes,
some additional constraints can also be consid-
4.3 TE Key Components ered when deciding upon the possible routing.
The key components of the TE process model Examples of such constraints are available band-
are: width, hop count, and delay. This implies that
possible paths as seen from a router must have
Measurement subsystem: Carrying out mea- the corresponding attributes attached.
surement is essential to provide feedback on
the system state and performance. It is also 4.4 Requirements on TE Systems
critical in order to assess the service level pro- [RFC3272] describes a number of requirements
vided (and QoS) and effect of TE actions. A that a TE system should meet. Here a require-
basic distinction between monitoring and ment is understood as a capability needed to
evaluation is to be observed; monitoring refers solve a TE problem or to achieve a TE objective.
to provision of raw data, while evaluation The requirements are either non-functional or
refers to use of the raw data for inferring on functional. A non-functional requirement relates
the system state and performance. Measure- to the quality attributes of state characteristics of
ments can be carried out at different levels of a TE system. A functional requirement gives the
aggregation, e.g. packet level, flow level, user function a TE system should perform in order to
level, traffic aggregate level, component level, reach an objective or address a problem.
network-wide level, and so forth. In order to
perform measurements systematically, several 4.4.1 Non-functional Requirements
questions have to be answered, like The generic non-functional requirements given
[RFC3272]: Which parameters are to be in [RFC3272] are:
measured? How should the measurements
be accomplished? Where should the measure- Usability. This is a human factor aspect refer-
ment be performed? When should the mea- ring to the ease of deployment and operation
surement be performed? How frequently of a TE system.
should the monitored variables be measured?
What level of measurement accuracy and reli- Automation. Usually, as many functions as
ability is desirable and realistic? To what possible should be automated, reducing the
extent can the measurement system permissi- human effort to control and analyse the infor-
bly interfere with the operational network con- mation and network state. This is even
ditions? What is the acceptable cost of mea- stronger for a larger network.
surements?
Scalability. The TE system should scale well
Modelling and analysis subsystem: A central when the number of routers, links, traffic
part of the modelling is to elaborate a represen- flows, subscribers, etc. grows. This may imply
tation of the relevant traffic characteristics and that a scalable TE architecture is applied.
network behaviour. In case a structural model
is used, the organisation of the network and its Stability. This is an essential requirement for
components are the main emphasis. When an operational system avoiding adverse results
behavioural models are used, the dynamics of for certain combinations of input and state
the network and traffic are the key issues. The information.
latter model is particularly relevant when
doing performance studies. Then adequate Flexibility. A TE system should be flexible
models of the traffic sources are also needed. both in terms of the optimisation policy and
the scope. An example of scope is that addi-
Optimisation subsystem: Optimisation can be tional classes should be considered in case
categorised as real-time and non-real-time. these are introduced into the network. Another
The former operates on short to medium time aspect of flexibility is that some subsystems of
scales (e.g. ms to hours) and works to adjust the TE system could be enabled/disabled.
parameters in mechanisms in order to relieve
congestion and improve performance. Exam- Visibility. Mechanisms to collect information
ples of means are tuning of routing parame- from the network elements and analyse the
ters, tuning of buffer management mecha- data have to be present in a TE system. These
nisms and changing Label Switched Paths would then allow for presenting the opera-
(LSPs). Non-real-time is also seen as network tional conditions of the network.
planning, typically working on a longer scale.
For both of these, stability and robustness are Simplicity. A TE system should be as simple
essential concerns. as possible; that is, considered from the out-
side, not necessarily using simple algorithms. paths for packets and may work well with
Simplicity is particularly important for the path-oriented solutions, that is LSPs.
human interface.
Traffic mapping. This refers to assigning traf-
Efficiency. As little demanding as possible, in fic flows onto paths to meet certain require-
terms of processing and memory resources, is ments considering the set of relevant policies.
requested. However, this also refers to the fact A central issue arises when several paths are
that a result from the TE system should be present between the same pair of originating
obtained in a timely manner. and destination router. Appropriate measures
should be taken to distribute the traffic acc-
Reliability. A TE system should be available ording to some defined ratios, while keeping
in the operational state when needed. the ordering of packets belonging to the same
application (or micro-flow).
Survivability. Recovering from a failure and
maintaining the operation is requested, in par- Measurement. Mechanisms for monitoring,
ticular for the more critical functions of a TE collecting and analysing statistical data have
system. Commonly, this requires that some to be in place. Such data may relate to perfor-
redundancy is introduced. mance and traffic. In particular, being able to
construct traffic matrices per service class is
Correctness. A correct response (according an essential part of a TE system.
to the algorithms implemented) has to be ob-
tained from a TE system. Network survivability. Survivability refers to
the capability to maintain service continuity
Maintainability. It should be simple to main- in the presence of faults. This can be realised
tain a TE system. by rapid recovering or by redundancy. Co-
ordinating of protection and restoration capa-
Extensibility. It should be easy to extend a TE bilities across multiple layers is a challenging
system, e.g. when introducing new functions task. At the different layers protection and
and when the underlying network is extended. restoration would typically occur at different
temporal and bandwidth granularity. At the
Interoperability. Open standards should be bottom layer, an optical network layer may
used for the interfaces in order to simplify be present, e.g. utilising WDM. Then, SDH
interoperation with other systems. and/or ATM could be present below the IP
layer. Restoration at the IP layer is commonly
Security. Means supporting integrity, informa- done by the routing protocols, which may
tion concealment, etc. have to be imple- require some minutes to complete. Some
mented. means being proposed relate to MPLS allow-
ing for faster recovery. A common suite of
As mentioned, some of these requirements may control plane protocols has been proposed
be mandatory while others are optional for a par- for the MPLS and optical transport networks.
ticular TE system. This may support more sophisticated restora-
tion capabilities. When multiple service
4.4.2 Functional Requirements classes are present, their requirements on
Some functional requirements are also described restoration may differ introducing further
in [RFC3272], such as those related to: challenges on the mechanisms to be used.
Resilience attributes can be attached to an
Routing. An efficient routing system should LSP telling how traffic on that LSP can be
take both traffic characteristics and network restored in case of failure. A basic attribute
constraints into account when deriving the may indicate if all traffic trunks in the LSP are
better routing schemes. A load splitting ratio transferred on a backup LSP or some of the
among alternative paths should be config- traffic is to be routed outside, e.g. following
urable allowing for more flexibility in the traf- the routing protocols. Extended attributes may
fic distribution. Some routes of subsets of traf- be introduced giving indications such as
fic should also be controllable without affect- backup LSP is to be pre-established, con-
ing routes of other traffic flows. This is partic- straints for routing the backup LSP, priorities
ularly relevant when several classes are pre- when routing backup LSP, and so forth.
sent in the network. In order to convey infor-
mation on topology, link characteristics and Servers and content distribution. Location and
traffic load, several of the relevant routing allocation of content on servers have signifi-
protocols have to be enhanced. An example cant impact on the traffic distribution, in par-
is constraint-based routing, which is gaining ticular as long as much of the traffic is similar
more interest. This addresses the selection of to clientserver interactions. Hence, load bal-
ancing directing traffic on the different servers 4.5 TE Taxonomy
may improve the overall performance. This is A taxonomy of TE systems is given in
sometimes called traffic directing, operating [RFC3272] according to the following criteria:
on the application layer.
Time-dependent vs. state-dependent vs. event-
DiffServ issues. As DiffServ is more widely dependent. A static TE system implies that no
deployed, adequate TE systems become more TE methods are applied on the time scale con-
critical to ensure that conditions in Service sidered. Therefore, it is commonly assumed
Level Agreements (SLAs) are met. Service that all TE schemes are dynamic (on the time
classes can be offered by defining Per-Hop scale looked at). A time-dependent scheme is
Behaviours (PHBs) along the path, exercising based on timely variations in traffic patterns
DiffServ in the nodes, in particular by configur- and used to pre-program changes in the traffic
ing mechanisms like traffic classification, handling. A state-dependent scheme adapts
marking, policing and shaping (mainly in edge the traffic handling based on state of the net-
routers). A PHB is a forwarding treatment work, allowing for taking actual variations in
including buffer management and scheduling. the traffic patterns into account. The state of a
In addition the amount of service capacity, e.g. network may be based on resource utilisation,
bandwidth, allocated to the different service delay measures, etc. Accurate information
classes has to be decided upon. The following available is crucial for adaptive TE schemes.
issues, from [ID_tepri], give some requirements This information has to be gathered and dis-
on TE in a DiffServ/MPLS environment: tributed to the relevant routers. A challenge is
to limit the amount of information that must
- An LSP should provide configurable maxi- be exchanged between routers, still allowing
mum reservable bandwidth and/or buffer for sufficiently updated data in each of the
for each supported service class. routers to take the traffic handling decisions.
Event-dependent schemes may lead to less
- An LSR may provide configurable mini- information exchanges compared to state-
mum available bandwidth and/or buffer for dependent schemes. Then, certain events are
each class on each of its links. used as input when updating the traffic han-
dling, like traffic load crossing a threshold,
- In order to perform constraint-based routing unsuccessful establishment of an LSP, etc.
on a per-class basis for LSPs, the routing
protocols should support extensions to Offline vs. online. In case changes in traffic
propagate per-class resource information. handling do not need to be done in real time,
When delay bounds is an issue, path selec- the computations can be done offline, e.g.
tion algorithms for traffic trunks with allowing for more thorough searches over the
bounded delay requirement should take feasible solutions finding the best one to
delay constraints into account. apply. On the other hand, when traffic han-
dling is to adapt to changing traffic patterns,
- When an LSR dynamically adjusts resource it is to be done online. For online calculations,
allocation based on per-class LSP resource relatively simple algorithms are applied lead-
requests, adjusting weights for the scheduling to short response times until the updated
ing algorithms should not adversely impact traffic handling can be activated. Still the
delay and jitter characteristics. algorithm should present a solution that is
close to the optimal one.
- An LSR should provide configurable maxi-
mum allocation multiplier on a per-class Centralised vs. distributed. In a centralised
basis. scheme a central function decides upon the
traffic handling in each of the routers. Then,
- Measurement-based admission control may the central function has to collect and return
be used to improve resource usage, espe- the information. In order to limit the overhead,
cially for classes not having strict loss or infrequent information exchanges are sought;
delay/jitter requirements. however, more frequent exchanges are asked
for to keep an accurate picture of the network
Controlling the network. In order to see the state in the central function. This results in a
effect of having a TE system, the relevant classic trade-off problem, finding the time
decisions must be introduced into the network. interval for collecting and returning the infor-
Control mechanisms may be manual or automation. A similar trade-off is also seen for the
matic, the latter being a goal for most. Net- distributed scheme, although then the deci-
work control functions must be secure, reli- sions are made by each router. A drawback of
able and stable, in particular during failure sit- a centralised scheme is often that a single
uations. point of failure is introduced, implying that
the central function is available and has suffi- 5 Key Technical Activities
cient processing capacity for the scheme to Overview
work efficiently. A number of basic methods are seen for struc-
tured handling of network planning tasks. These
Local vs. global information. Local informa- could be described as:
tion refers to a portion of the region/domain
considered by the TE system. An example is Mathematical, analytical methods:
delay for a particular LSP. Global information - Optimisation: linear programming, integer
refers to the whole region/domain considered. programming, dynamic programming.
Applied to find a better solution (opti-
Prescriptive vs. descriptive. When a prescrip- mum) given a number of criteria/con-
tive approach is used, a set of actions would straints, objective functions, parameters
be suggested by the TE system. Such an to be found.
approach can be either corrective (an action
to solve an existing or predicted anomaly) or - Complexity analysis, e.g. for scalability
perfective (an action suggested without identi- studies
fying any particular anomaly). A descriptive
approach characterises the network state and - Queueing theory, including Markovian pro-
assesses the impact from exercising various cesses. Applied for dimensioning, depend-
policies without suggesting any specific ability, etc.
action.
- Network theory
Open loop vs. closed loop. In an open loop
approach, the control actions do not use feed- Simulations: Commonly used for complex,
back information from the network. Such inter-dependent subsystems as they may
feedback information is used when a closed become too involved for analytical treatment
loop approach is followed. (or not solvable at all).
Tactical vs. strategic. A tactical approach con- Measurements/testing: Applied for actual
siders a specific problem, without taking into implementations of systems in operation or
account the overall solutions, tending to be ad configurations mimicking the system to be
hoc in nature. A strategic approach considers placed into operation.
the TE problem from a more organised and
systematic perspective, including immediate A characteristic of modelling teletraffic phenom-
and longer-term consequences. ena is that fairly simple models can be treated by
analytical means. Approximations are com-
Intradomain vs. interdomain. Interdomain monly introduced implying that the results
traffic engineering is primarily concerned with achieved strictly are valid within certain regions
performance of traffic and networks when the of the system loads. Common simplifications are
traffic flows crosses a domain, e.g. between Poisonian distributed arrival rates, exponentially
two operators. Both technical and administra- serving times, independence between arrivals,
tive/business concerns make such a TE activ- service periods, servers. These are seldom com-
ity more complicated. One example is based pletely met, but usually allow observations with
on the fact that Border Gateway Protocol ver- sufficient accuracy. On the other hand, it is
sion 4 being the (default) standard routing important to be aware of the limitations of the
protocol does not carry full information like applied methods, avoiding conclusions to be
an interior gateway protocol (e.g. no topology invalidly made.
and link state information). In a business sense
it would not be likely that two parties, being Analytic algorithms may be used to identify
potential competitors, would reveal all that principle behaviour or theoretic phenomena.
data of its network. Another aspect is the pres- Some of these observations may be fed into sim-
ence of relevant SLAs that govern the inter- ulation models also incorporating more depen-
connection, including description of traffic dencies and time-varying behaviour. Measure-
patterns, QoS, measurements and reactions. ments, either by a laboratory or in the field, are
An SLA may explicitly or implicitly specify a however from the outset closer to the real sys-
Traffic Conditioning Agreement, which tem. The measurement system must then mimic
defines classifier rules as well as metering, the actual behaviour.
marking, discarding and shaping rules (from
TE framework). Then, commonly, observations from measure-
ments can be abstracted to add to the knowledge
A specific TE system can then be categorised by as well as potentially be fed back into both ana-
applying the criteria listed above. lytic and simulation models. The three cate-
gories are mutually supportive and indispensable Observation Figure 20 Teletraffic theory
during the network planning and following the as an inductive discipline
network operation. (from [ITU-D_TE])
Model
Computer systems are regularly needed for net-
work planning tasks of realistic cases because of
number of candidates, number of calculations,
Deduction
etc. Another aspect is relations with databases
containing information on the networks and
systems to be planned. Relevant data must be
fetched, sorted and stored for these bases. Data
Teletraffic theory can be defined as (ref. [ITU-

D_TE]): the application of probability theory to Verification
the solution of problems concerning planning,
performance evaluation, operation and mainte-
nance of telecommunication systems. In a more
general setting, teletraffic theory can be consid- traffic theory as an inductive discipline is de-
ered as a discipline of planning where the tools picted in Figure 20. The models are established
(stochastic processes, queueing theory and after observing the real systems. A number of
numerical simulation) are taken from the disci- parameters are derived, which may be compared
plines of operations research. to corresponding observations (or predictions)
from the real system. In case of (stochastic con-
The objective of teletraffic theory is formulated fidence) correspondence, the model is said to be
as: to make the traffic measurable in well-de- validated. If not, the model should be elaborated
fined units through mathematical models and to further. In general, this is called the research
derive the relationship between grade-of-service spiral.
and system capacity in such a way that the
theory becomes a tool by which investments The tasks involved in traffic engineering, as seen
can be planned. from ITU-D, are illustrated in Figure 21.
Hence, the task is to assist when designing sys- In the following sections a number of key areas
tems given pre-defined grade-of-service levels, and mechanisms are described. For an opera-
traffic demand and capacity of system compo- tional network, all of these areas should be con-
nents. The overall procedure for applying tele- sidered.
Traffic demand characterisation Grade of Service objectives
Traffic Traffic QoS requirements

modelling measurement
End-to-end GoS objectives
Traffic
forecasting Allocation to network components
Traffic controls and dimensioning
Traffic
Dimensioning
controls
Performance monitoring
Performance monitoring
Figure 21 Traffic engineering tasks (from [ITU-D_TE])
6 Basic Traffic Terms That is, Ar = A Ac = A . (1 b), where b is the
One commonly meets the term traffic when dis- probability of a request being blocked/rejected.
cussion network planning. Traffic is used to
denote traffic intensity; that is, traffic per time In the general case, carried traffic is the only
unit. The definition is: Traffic intensity measurable dimension (as one does not actually
The instantaneous traffic intensity on a pool of know service times for requests being rejected).
resources is the number of busy resources at
a given instant of time. The pool of resources Assuming that a system (pool of servers) has
may be a group of servers. Note that one may capacity C, the normalised load is given by
get association of a discrete number of servers
s
although the term is generally applicable also for = .
C
a continuum of resources (e.g. for link band-
width). The observed utilisation (i.e. referring to carried
traffic) is in the interval [0,1].
Statistical moment of the traffic intensity can be
calculated for a time interval, say from 0 to T, In a multiservice system different request types
e.g. for the mean value: may occupy different amounts of server unit
capacity. This must be considered when the traf-
1 T
Ac (T ) = n(t)dt fic is calculated. Assuming that traffic type i
T 0
occupies a server capacity di, the offered traffic
where n(t) expresses the (number of) occupied for K request types is found as:
servers at time t.
K

A= i si di
Often the unit used for traffic intensity is erlang i=1
(referring to the Danish mathematician E.K.
Erlang, 1878 1929). This unit is actually Traffic characterisation is often done by estab-
dimensionless. lishing models that approximate the stochastic
behaviour of the traffic flows. As few traffic
Carried traffic, Ac The amount of traffic car- parameters as possible is strived for (e.g. mean
ried by the group of servers during a time inter- value, variance, etc.) while still being able to
val. This can be measured by observing the capture the pertinent characteristics of the traffic
amount of servers being busy in that interval. flow. In fact, the key trick in traffic modelling is
to identify the simplifying assumptions to be
The total traffic carried during a time interval is made and the set of parameters to use.
named the traffic volume, e.g.
When possible, measurements are conducted to
T
n(t)dt, validate the traffic models. On the other hand,
0 measurements could only be done in order to
assess the parameter values to be used, for exam-
being equal to the sum of all occupation times ple when the traffic models as such are fixed or
inside the time interval. The carried traffic can applied due to previous experience.
never exceed the capacity of the server pool (i.e.
a single server unit can never carry more than For an expected future situation forecasting tech-
one erlang). niques are required in order to estimate the traf-
fic demand. A list of ITU-T recommendations in
Offered traffic, A The traffic that would be car- this area is given in Box A.
ried if no requests were to be rejected due to lack
of server capacity. Two types of parameters are Examples of characteristics of the traffic are per-
often seen to describe traffic: sistence (on-demand or permanent), information
transfer mode (circuit switched or packet
The request intensity, , which is the mean switched), configuration (point-to-point, multi-
number of requests offered per time unit. point-to-multipoint, etc.), transfer bit rate, QoS
parameter requirements, symmetry (unidirec-
The mean service time, s, for a request (note tional, bi-directional asymmetric, bi-directional
that this must be given with the same time unit symmetric), and so forth. These are all related to
as the request intensity). the session and connections being parts of the
session.
The offered traffic is then equal to A = . s.
Again, it is recognized that traffic is dimensionless. A session can be initiated according to a process
such as the request initiation process for
Rejected traffic, Ar The difference between example given by the arrival rate or duration.
offered traffic and carried traffic.
Going further into the time scale, phenomena
relating to individual information units (e.g. Box A Selected ITU-T Recommendations in the Area of
Traffic Demand Characterisation
packets) are considered. These would be packet
interarrival times or packet lengths. Traffic Modelling
E.711 User demand modelling
A further dimension is needed for mobile sys- E.712 User plane traffic modelling
tems where the users whereabouts also need to E.713 Control plane traffic modelling
be considered. That is, the spatial effect has to E.716 User demand modelling in Broadband-ISDN
be taken into account. E.760 Terminal mobility traffic modelling
E.523 Standard traffic profiles for international traffic streams
The offered traffic varies as a function of time
in accordance with the activity of the environ- Traffic Measurements
ment of the system considered. For PSTN/ISDN E.490 Traffic measurement and evaluation general survey
the term Busy Hour has been defined. This E.491 Traffic measurement by destination
refers to the highest traffic during a day. The E.500 Traffic intensity measurement principles
Time Consistent Busy Hour (TCBH) is defined E.501 Estimation of traffic offered in the network
as those 60 minutes (determined with an accu- E.502 Traffic measurement requirements for digital telecommunication
racy of 15 minutes), which has the highest aver- exchanges
age traffic over a long period. With this defini- E.503 Traffic measurement data analysis
tion it may well happen that the traffic offered in E.504 Traffic measurement administration
the busiest hour on some days is higher than the E.505 Measurements of the performance of common channel signalling
traffic in the TCBH. network
E.743 Traffic measurements for SS no 7 dimensioning and planning
The traffic variations can also be divided into E.745 Cell level measurement requirements for the B-ISDN
different time scales such as:
Traffic Forecasting
E.506 Forecasting international traffic
The 24 hours daily variations
E.507 Models for forecasting international traffic
E.508 Forecasting new telecommunication services
The weekly variations
The yearly variations; some days or seasons

are expected to have higher traffic such as Time congestion: The fraction of time all
New Years Eve servers are busy.
Traffic trends (increases or decreases) Traffic congestion: The fraction of the offered
traffic that is not carried.
Depending on how the system at hand operates,
we may distinguish loss-systems from waiting- Typically, these indicators are used to establish
time systems. For the former, requests are dimensioning standards for server groups.
rejected if all servers are occupied; this is typi-
cally the case for calls offered to trunk groups. An inconvenience of waiting-time systems is the
For waiting-time systems the request is placed resulting waiting time. Not only the mean wait-
into a queue if it finds the servers occupied upon ing time is of interest, but also the distribution of
arrival. the waiting time. For example some short wait-
ing time would barely be noticeable. The rela-
A loss-system has a number of indicators to tion between the waiting time and the level of
express the level of inadequacy: inconvenience for a user is not obvious.
Request congestion: The fraction of requests Having finite financial means, it is often not
that observes all servers busy upon arrival. cost-effective to build a system that meets any
No answer by
B-user
Technical errors
A-user error B-user busy
and blocking
Figure 22 Sequence of
B-user answer Conversation potential events during a
PSTN set-up
imaginable situation of request rates. One of the Convergence to a state which maximises the
objectives of teletraffic theory is to find relations throughput of an overloaded resource subject
between the offered traffic, the system capacity to keeping its response times short enough to
and the resulting service level. reduce the occurrence of user abandons.
Achieving such convergence automatically
For a loss-system a block-diagram (in several over a realistic range of overload scenarios
ways similar to the ones used for dependability (including a range of overloaded resource
analyses) can be set up. An example is shown in capacitates and number of active traffic
Figure 22. Such diagrams may also be found for sources).
other services, e.g. for accessing a web page/site
when browsing, although this service is rather Minimising ineffective traffic flows by selec-
implemented by waiting-time servers. tively throttling flows experiencing high drop-
ping ratios.
Naturally, the schema depicted in Figure 22 can
also be used to study repeated call attempts. That Fairness between traffic flows in a certain
is, the users response to unsuccessful requests sense.
impacts the resulting load on the systems and
hence an essential part of dimensioning the system. Overload situations are typically stimulated by
a few events:
One may make a distinction between characteri-
sation and modelling, ref. [Macf02]. Characteri- Media-stimulated, e.g. invitations to initiate
sation refers to a phenomenological description sessions/accessing pages, casting votes, com-
of traffic, i.e. the analyses of measurement data, petitions, marketing;
and the production of high-level statistical
descriptions. Modelling covers the detailed low- Emergency situations;
level production of probabilistic models of traf-
fic that can actually be used to make predictions Network equipment failures, including soft-
about network and system behaviour. These two ware/configuration errors;
perspectives come together by the requirement
of compatibility the high-level profile derived Auto-scheduled session initiations, including
from the low-level model must fit the experi- Denial of Service attempts.
mentally observed characterisation.
In the absence of effective control mechanisms,
On the shorter time scales, traffic descriptors are such overload-triggering events would threaten
used as inputs for performance estimation and the stability of network systems and cause a
system sizing. A number of usefulness criteria severe reduction in service levels. Ultimately,
should be fulfilled for such descriptors [Macf02]. systems might fail and the service would not be
Of most basic nature, the descriptors should allow available to users.
for mathematically tractable handling and result-
ing in useful results. Moreover, they should be: End-users are commonly intolerant with respect
to long response times to their service requests.
Stable a given stationary traffic flow should Typically, when a user has to wait more than a
not have wildly fluctuating parameters from few seconds, re-attempts are made or the service
one day to another; request is abandoned altogether. Similar phe-
nomena are found for different types of services,
Parsimonious only a small number of such as telephone calls, web page accesses and
parameters should be necessary; so forth.
Comprehensible the significance of the User impatience combined with an excessively

parameters should be easily understood; long response time can cause the effective
throughput of a network resource to collapse
Aggregatable the parameters of the super- to a low level when the original request load is
position of two flows should bear a simple high enough. This may come from the resource
relation to those of its components; spending more of its capacity on partially pro-
cessing the requests, not finalising any. Hence, a
Scalable natural growth in traffic should not huge backlog of requests can occur, while most
result in complex changes in parameters. of the users requests are not being completed
before a certain threshold is met (the user
7 Controlling Network Load releases, re-tries, protocol time-out, etc.).
Generic requirements on the behaviour of a net-
work overload control include: The user model of this is often assumed to be
described by (see Figure 23), i) the persistence
Original Overall system Serviced
requests arriving requests
requests
p Retried Rejected
requests requests
probability, say pi, that a new attempt will be (1 B) (t + s (1 B p) Figure 23 Schematic

made if the ith attempt is not serviced, and, ii) (1 B p)2 illustration of re-attempts
the distribution of the time interval between the contributing to the overall load
ith and (i +1)th attempt. Keeping the notion of Ac for carried traffic and A on the system
for offered traffic, A = . s, gives:
If all attempts fail, then the mean number of

attempts per initial intent is given by the series t/s + 1 B p
Ac = A (1 B)
1 + p1 + p1(p2 + p1(p2(p3 + . If all pk = p, this (1 B p)2
adds to (1 p)-1. As an example, assume that p =
80 % (which is observed for some media-stimu- That is, the carried traffic can very well be-
lated events. Then, on average five attempts per come higher than the originally offered traffic,
original intent is seen (if all attempts fail). which may impact the dimensioning. Several
Although being a simple example, it is seen means could be introduced to combat this situa-
that such phenomena easily result in a drastic tion, e.g. to reduce the persistence probability
increase in the number or requests on a resource. (e.g. providing a prompt to the user that re-
Hence, user persistence may lead to significant requests should be postponed) or to increase the
higher load, which even may result in overload completion probability (e.g. attempted by pro-
in other parts of the network due to that those viding a voice-mail service).
resources are not freed in order to serve requests
that are successfully completed. During varying request mixes/demand patterns,
different types of resources in the network may
Expanding on the notion of an unsuccessful become the bottleneck. In particular, within a
request: network node, the resources may have different
tasks; some taking care of peripheral interfaces/
P the persistence devices, some protocol processing, others rout-
B the probability of not completed service ing updates, etc. Preserving a controlled service
(e.g. no reply by B-side) processing, it is essential to implement an ade-
quate overload control mechanism. This mecha-
Say that a request requires s time units service nism is to reject requests in order to keep
when not completed and t time units when com- response times low and ensure service comple-
pleted. Then the following can be seen: tion at satisfactory levels.
Occupancy time when successful first time: When media-stimulated events trigger a mass-
s; probability: (1 B) call (voting, quiz, etc.) the call arrival curve usu-
ally has a fairly steep edge to a peak and a
Occupancy time when successful second time: slower decrease, as illustrated in Figure 24. An
(t + s); probability: B . p . (1 B) essential feature of the overload mechanism is to
avoid that the peak results in a too severe degra-
Occupancy time when successful third time: dation in the service on this call type as well as
(2 . t + s); probability: (B . p)2 . (1 B) the other call types utilising the same resources.
Hence, the overload control kicks in at a certain
... load level which is again related to a certain
level of the call arrival rate. Then, a difference
Occupancy time when successful ith time: between the offered calls and the admitted calls
(i . t + s); probability: (B . p)i-1 . (1 B) can be observed at high load intervals.
The average occupancy time can then be found The relation between offered and admitted calls
to be: is shown in Figure 24. Naturally, the exact shape
of these curves depends on the resource types
Load Serviced load
Rejected load
Load control
activated
Capacity
Gaia by activity
load control
Load control
activated
Time Offered
load
Figure 24 Illustration of and call types in question. In general, however, trols have to react fast enough to prevent that
request arrival for mass-calls as the offered request rate increases, a point will the total load too greatly exceeds the pro-
and gains from load control be reached where the overload control is in- cessed load during the load increase phase.
mechanisms voked and some part of the offered load is
rejected in order to preserve the service capacity. As described earlier, using the reject rate as the
As the load increases even further, the relative controlled variable leads to convergence in the
share of admitted requests decreases, although admitted rate. This is frequently combined with
the trick is to keep the rate of admitted requests a leaky bucket algorithm due to its fairly simple
at almost the same level. Again, this depends on implementation and effective response.
the implementation. For example, in case the
same processor is used to run the overload algo- Overload controls are also expected in several
rithm, some initial processing would likely be coming applications, such as service control, e.g.
needed for all arriving requests. This is to see by SIP or HTTP requests. On the other hand, it
what to do with them as well as to complete pro- is important to distinguish between messages.
cessing/serving of requests. Due to the initial For example, messages releasing resources (e.g.
processing per request the single processor will circuit disconnect message) should not be throt-
eventually be overloaded. That is, the rate of tled as that could further worsen the load situa-
completed services will decrease, although this tion.
should happen at a higher load than would occur
when no overload control was activated. In order 8 Traffic Controls and
to prevent the rate of completed services from Dimensioning
decreasing, the load control as well as any ini- An objective of traffic engineering is to provide
tial processing of every request should be efficient design and operation of the network
placed externally to the bottleneck resource. In while assuring that the demand is served within
order for the overload control to be effective, it the requirements stated. Network dimensioning
needs to be automatic and embedded in the net- assures that the network has adequately config-
work resource. This is to react sufficiently ured resources, including the physical network
rapidly to changes in demand or resource capac- elements and the logical network elements (e.g.
ity. The mechanism must be able to handle vari- virtual paths).
ations in the request mixtures and the way
requests are distributed over the relevant Several items are included in the traffic controls
resources. For network planning objectives, all category ([ITU-D_TE]):
the parameters must be coordinately assigned in
correspondence with targets for the different Traffic routing: giving the set of routing
request types (reflected into service classes). choices and rules for selecting routes for dif-
These should then be according to the SLA con- ferent traffic types (including different source-
ditions. destination pairs). Routing principles may be
flat, hierarchical, fixed, dynamic, state-based,
A number of requirements for overload controls event-based, etc.
are (adopted from [Whit02]):
Network traffic management controls: ensur-
Maximise effective throughput of the bottle- ing that the traffic throughput is maintained
neck resource, during all ranges of offered under overload and failure conditions. These
request rates. may be restrictive or expansive. Restrictive
controls, such as blocking or call gapping,
Bounding response times related to the bottle- throttle the traffic sources. The expansive con-
neck resource, in particular during overload trols typically reroute the traffic towards the
intervals. This implies that the overload con- network portions that are less loaded. Often,
real-time monitoring of the performance/load
levels triggers these actions related to pre-
Box B Selected ITU-T Recommendations for Traffic
Controls and Dimensioning
defined levels. For example, when the number
of calls on a link is equal to a given threshold, Circuit-switched Networks
additional calls are rejected. E.510 Determination of the number of circuits in manual operation
E.520 Number of circuits to be provided in automatic and/or semi-automatic
Service protection methods: controlling the operation, without overflow facilities
grade of service of certain traffic types by util- E.521 Calculations of the number of circuits in a group carrying overflow
ising discriminatory restriction of access to traffic
resource groups. E.522 Number of circuits in a high-usage group
E.524 Overflow approximations for non-random inputs
One example is to balance GoS between traffic E.525 Designing network to control grade of service
types requesting bandwidth. Another example is E.526 Dimensioning a circuit group with multi-slot bearer services and non
to provide stability in networks without hierar- overflow inputs
chical routing schemes by restricting overflow E.527 Dimensioning a circuit group with multi-slot bearer services and
traffic to an alternative route that is shared with overflow traffic
firsts-choice traffic. E.528 Dimensioning of digital circuit multiplication equipment (DCME)
systems
When several traffic types compete for the net- E.529 Network dimensioning using end-to-end GoS objectives
work resources, the requirements of the individ- E.731 Methods for dimensioning resources operating in circuit switched mode
ual types have to be checked to ensure that their
service levels should be met. Several systems Broadband Integrated Services Digital Network
E.735 Framework for traffic control and dimensioning in B-ISDN
support prioritisation of traffic, including sepa-
E.736 Methods for cell level traffic control in B-ISDN
rate criteria for admission control, scheduling
E.737 Dimensioning methods for B-ISDN
and dropping rules.
Signalling and Service Control
Designing systems that continue to operate at E.733 Methods for dimensioning resources in Signalling System no. 7
maximum capacity even during overloads, vari- networks
ous overload strategies can be introduced. For E.734 Methods for allocating and dimensioning Intelligent Network (IN)
processing and signalling tasks leaky bucket, resources
gapping and level priorities are seen. For infor-
mation transfer modes, bandwidth reservation
and circuit channels protection can be used,
besides various routing separations. failures. Hence, dependability analyses and traf-
fic analyses should be considered together. For
Bandwidth reservations can for example be used dependability analyses the phenomenon that a
in hierarchical networks with alternative routing fault may appear might be modelled as a process
where the primary traffic could be protected with given distributions, as well as the repair
against load from overflow traffic (overflow time of the failure. A fairly simple model of a
traffic refers to flows that are rerouted onto sec- system is expressed by a block diagram, where
ondary choices, e.g. when the primary route is all essential components are reflected. Then,
congested). A common rule is that overflow traf- series and parallel branches are resolved to esti-
fic is only accepted if more than a given band- mate the overall system dependability.
width is available.
A growing part of telecom systems is made of
Virtual channel protection includes two schemes; software modules in different nodes interacting.
i) each traffic type is allocated a minimum band- Considering replication of data and program rou-
width ensuring a minimum service level, and, ii) tines, a certain level of dependency may be pre-
a maximum allocation per traffic type avoiding sent although from a (hardware) point of view
that a single traffic type dominates. the resources are completed. It is vital to con-
sider the commonalities and dependencies in a
Packet-level traffic controls assure the packet- dependability model to avoid that incorrect con-
level GoS objectives under any network condi- clusions are drawn.
tion and that a cost-efficient GoS differentiation
is made between services with different packet- In one way, technical risk analyses are related to
level QoS requirements. A selection of ITU-T the dependability. Then the risk can be defined
recommendations for traffic controls and dimen- as the product of the probability and the conse-
sioning is listed in Box B. quence of an event. The overall risk becomes the
sum over all the events. In several cases, quanti-
In some respects, an incorrectly dimensioned tative estimates of the events are challenging to
network will have a degraded performance level estimate, resulting in risk diagrams having con-
in a similar manner as a network experiencing sequence and probability on the axes cate-
gorised by small, medium and high. The risk one may say that a future standpoint is taken
profile of a company then decides which mea- looking back towards today to reveal the
sures to apply trying to eliminate the events that causes of that scenario to appear. A character-
are most unwanted. istic of Delphi and expert panels is that several
rounds are carried out, where the participants
A part of the network planning area is to formu- may get feedback on the statements of others
late contingency plans. These may be expressed in between. This is to try to achieve a conver-
as; what to do when XX happens. Then, respon- gence on opinions.
sible units and their actions have to be clear to
the parties involved. Moreover, all XX events Traffic forecasting is necessary both for strategic
should be captured. Examples of events are fire studies as well as when designing systems on a
in a station, major vehicle accidents involving medium term. Broadly, two basic approaches
police, ambulance, etc. might be identified for forecasting:
9 Forecasting Descriptive approach, where the basic

Considering the steadily changing environment assumption is that the underlying process gen-
of a telecom actor, it is essential to have ideas on erating the demand in the past continues into
what the future situation will be like. In particu- the future. Statistical model is for example
lar this is relevant for the demand patterns. Vari- applicable in this case, whereby the time
ous forecasting techniques are applied to get an series (or demand history) is analysed for
objective view of the future situation this is components such as the base or constant com-
commonly done by looking at the forces work- ponent, trends, etc. Then, this description is
ing in the past, present and future periods. extrapolated into the future. Naturally, knowl-
edge of any impending events should be
Two basically different approaches for making included in the extrapolation.
forecasts are:
Explanatory approach, where casual relation-
Quantitative including trend and factor meth- ships are built into the model. Several models
ods, analogy methods and potentials. A trend is can be found in the literature, such as wide
a statistical description of a historic and rele- econometric models to models predicting the
vant development that is prolonged into the purchasing patterns for a group of customers.
future. A basic precondition for doing this is
that factors that influence the development For all the forecasting a random error compo-
have been examined and their potentially nent (sometimes referred to as noise) is present.
changing magnitudes in the future captured in This composes the unexplained deviation of the
the predictions. An analogy method can be data from the basic demand generating process.
applied when the case at hand is believed to fol-
low a similar evolution as another known case When returning to the descriptive models, the
(e.g. another region/nation, another product, following patterns are commonly considered
another customer segment, etc.). This method when relevant, see Figure 25:
may also be applied when relating products
with other industries. A potential for a product Base representing a central tendency of the
or customer segment (also referred to as satura- time series at any given time
tion level) could possible be transferred from
other industries or other product groups. For Trend exhibiting a consistent long-run shift
example, the number of internet subscriptions of the average
will not likely exceed the number of PCs.
Seasonal showing a repetitive variation, e.g.
Qualitative including perspective and sce- a one-year periodicity. This is commonly
nario analyses, and interviews and question- superposed upon a background process, e.g.
naires (such as Delphi and expert panels). The showing a trend.
perspective method aims at describing likely
development tracks commonly based on dis- As time passes, new information is gained and
cussions on technological, market, social and the models can be updated. When doing this
economic evolution in the period. Hence, this either the time origin can be fixed or the time
will take todays standpoint and derive the origin can be shifted to the end of the current
potential future based on that view. In a sce- period. In the former, the parameters have to be
nario method a more or less remote future is recomputed every time in order to incorporate
taken on. Potential alternative futures sce- the latest observations. In the latter, also called
narios are then elaborated, possibly without adaptive smoothing, the parameters are updated
describing what happened between today and by combining the old estimates and the latest
that point in the future. Hence, for a scenario forecast error recorded.
Demand Demand Demand
Time Time Time

(a) (b) (c)
A few forecasting techniques are: In case there are no historical data available, Figure 25 Demand patterns:
qualitative forecasting models are most relevant. (a) base, (b) trend, (c) trend
Regression methods fit some hypothesized Time series models and Kalman filter models and seasonal variations
model, linear in its coefficients, to the time are not relevant as they need a substantial num-
series. Least square estimators are commonly ber of (pre-)observations.
used for the coefficients. Multiple regression
can also be used as an explanatory method of It is essential to know whether the peak periods
recasting when a causal relationship between for different services for different customer
the dependent variables and the independent groups coincide as this affects the capacity
variables is present. needed in the network. This is illustrated in Fig-
ure 26, demonstrating a daily profile for residen-
Moving average computes an average of the tial and business customers.
constant number of past observations in order
to eliminate the random variations or noise. The averaging effect when measuring for a given
As a new observation becomes available, the interval must be handled. Some effects are illus-
oldest observation is dropped (in order to keep trated in Figure 27.
their number constant) and a new average is
computed. ITU-T recommendations in the forecasting area
are:
Exponential smoothing assigning different
weights to the observations according to an E.506 giving directions on forecasting
exponential decaying factor as the time dis- including base data, social and demographic
tance increases. Hence, longer (historic) val- data. In case some samples are missing, guid-
ues get lower impact on the forecasted value. ance for obtaining these elements is found.
Bayesian methods useful when faced with a E.507 presents an overview of mathematical
lack of historical data at the beginning of the forecasting techniques: curve-fitting models,
forecasting process. The approach is then to autoregressive models, autoregressive inte-
start with some initial subjective estimate grated moving average (ARIMA) models,
of the parameters in the model, and use of state space models with Kalman filtering,
Bayes theorem to modify them in light of the regression models and econometric models.
actual data observed. This recommendation also contains methods
for evaluating the forecasting models and how Figure 26 Daily traffic
Box-Jenkins models set of autoregressive to choose an appropriate one depending on profiles for different
models in which successive observations are available data, forecast period, and so forth. customer segments
highly dependent. Three classes of models are
considered; i) the autoregressive process,
ii) the moving average (or the errors) process,
Traffic load
and iii) the mixed autoregressive-moving
average process. The choice of the correct
model is made by examining the autocorrela- 5 Sum of max (business) and max (residential)
tion coefficients.
4 Integration gain
Both quantitative and qualitative forecasting
techniques can be exercised. The first group 3
includes models based on smoothing models,
2
time series models (Box Jenkins approach), Residential
Kalman filter models, Regression models, Logit Sum of business

1 and residential
models, diffusion models and other econometric
Business
models. Examples of qualitative techniques in-
clude market surveys, scenario methods, analogy 2 4 6 8 10 12 14 16 18 20 22 24
methods, expert methods and Delphi methods. Time-of-day
Figure 27 Effects from
Kbps
averaging over intervals
(from [Soto02])
14000
12000 5 min
average
10000
1 hour
8000 average
6000 24 hours
average
4000
2000
Time
E.508 describes how to forecast new ser- network, leading to the equation being undeter-
vices where there are no historical data. Tech- mined when the traffic matrix A is to be found
niques described include market research, (hence, theoretically there is an infinite number
expert opinion and sectorial econometrics. of feasible solutions for the traffic matrix).
The recommendation also discusses how to
combine results from different forecasting Assuming that the total traffic originating and
techniques and how to adjust the results when terminating in nodes are known, this is the same
first observations are obtained. as saying that the column and row sums in the
traffic matrix are known. Then, applying
A central part of deriving demand characteristics Kruithofs method, individual matrix elements
is to elaborate a set of traffic matrices. These can be estimated. The overall procedure is to
matrices give the amount of traffic flowing iterately adjust matrix elements trying to make
between sources and destinations. For N nodes the column sums and the row sums to fit the pre-
N2 traffic elements may complete the matrix. described values, respectively. Applying
Basically, these traffic matrices do not give any Kruithofs double factor method, a matrix ele-
information of the network topology or the traf- ment between two iterations is adjusted by the
fic routing. Two approaches for obtaining a traf- formula:
fic matrix are i) direct measurements of source
destination pairs, ii) partial information to infer St
Ak+1
ij = Akij
the matrix elements. Sk
In several networks, measurements of aggregates where

are carried out, such as the total traffic originat-
ing or terminating in a node. The individual ele- Ak+1ij the adjusted element value in the traf-
ments in the traffic matrix may not be known, fic matrix (iteration k + 1)
however; that is, the traffic between specific
node pairs. Full details from measurements are Akij element value in the previous iteration
not frequently seen as they involve a significant
amount of measurement infrastructure, data stor- Sk actual sum of elements in iteration k,
ing and processing, hence likely to become a for either column or row
costly approach. Therefore, other procedures
have been proposed only based on certain St target (column or row) sum to be
known data. These procedures are then started achieved
from these informed priors (referring to given
information on the traffic matrix). Other approaches have also been proposed to
estimate the individual traffic elements, but
In a network with n nodes there are generally Kruithofs double factor method fulfils objec-
n . (n 1) sourcedestination pairs involving tives such as (based on [ITU-D_TE]):
links in the network. If L is the traffic on links
between nodes, A the traffic matrix between Uniqueness only one solution exists for a
node pairs and R the matrix indicating whether given forecasts.
a link is used on the way between a source and a
destination, this can be written as: L = R . A. The Reversibility the resulting matrix can be
problem, however, is that usually there are many reversed to the initial matrix with the same
more sourcedestination pairs than links in a procedure.
Transitivity the resulting matrix is the same As described above, defining the reference con-
independent of whether it is obtained in one nection is followed by elaborating GoS objec-
step or via a series of intermediate transforma- tives. A number of recommendations are found
tions. for this purpose too:
Invariance regarding the sequence of nodes; E.540 Overall grade of service of the inter-
their sequence may be changed without influ- national pert of an international connection
encing the results.
E.541 Overall grade of service for interna-
Fractioning the individual nodes can be split tional connections (subscriber-to-subscriber)
into sub-nodes or be aggregated into larger
nodes without influencing the result. Note that E.543 Grades of service in digital interna-
this is not completely met for Kruithofs dou- tional telephone exchanges
ble factor method, but deviations are small.
E.550 Grade of service and new perfor-
10 Performance Objectives mance criteria under failure conditions in
and Service Constraints international telephone exchanges
Grade of Service (GoS) is defined as a number
of traffic engineering parameters to provide a E.720 ISDN grade of service concept
measure of adequacy of plant under specified
conditions. GoS parameters are expressed as E.721 Network grade of service parameters
probability of blocking delay distribution, etc. and target values for circuit-switched services
Network Performance (NP) is defined as the in the evolving ISDN
ability of a network or network portion to pro-
vide the functions related to communications E.723 Grade-of-service parameters for
between users. GoS is the traffic-related part of Signalling System No. 7 networks
NP, although NP also covers non-traffic related
aspects (including dependability, charging, etc.). E.724 GoS parameters and target GoS
objectives for IN Services
NP and GoS objectives are derived from Quality
of Service requirements. Basically, QoS should E.726 Network grade of service parameters
be user-oriented and, in principle, independent and target values for B-ISDN
of the network. NP parameters though, are net-
work-oriented; i.e. they can be used when speci- E.728 Grade of service parameters for
fying performance requirements. B-ISDN signalling
A common approach followed so far by ITU-T E.770 Land mobile and fixed network inter-
is to define a so-called reference connection in connection traffic grade of service concept
order to derive GoS conditions for a network
portion. Then, an end-to-end connection is parti- E.771 Network grade of service parameters
tioned into portions and the QoS requirements and target values for circuit-switched land
are divided into contributions from each of the mobile services
portions. In several cases, an interface between
two portions refers to an interface between two E.773 Maritime and aeronautical mobile
operators/providers. A number of recommenda- grade of service concept
tions address reference connections, including:
E.774 Network grade of service parameters
E.701 Reference connections for traffic and target values for maritime and aeronauti-
engineering cal mobile services
E.751 Reference connections for traffic E.775 UPT Grade of service concept
engineering of land mobile networks
E.776 Network grade of service parameters
E.752 Reference connections for traffic for UPT
engineering of maritime and aeronautical
systems E.671 Post selection delay for PSTN/ISDNs
using Internet telephony for a portion of the
E.755 Reference connections for UPT traffic connection
performance and GoS
E.651 Reference connections for traffic
E.651 Reference connections for traffic engineering of IP access networks
engineering of IP access networks
Y.1540 Internet protocol data communica- ent things to different people. When new sys-
tion service IP packet transfer and availabil- tems are to be introduced the components
ity performance parameters capacity should be known. This can be obtained
by measuring, under controlled conditions, or by
Y.1541 Network performance objectives for modelling the components before the system is
IP-based services deployed. This allows for a more accurate di-
mensioning and provides a basis for following
Looking into these one finds a separation the performance during operation.
between the user data performance and sig-
nalling performance. Examples of parameters of Performance and traffic analyses should be car-
the former are information transfer delay, infor- ried out for several phases of a system life,
mation delay variation, information loss ratio, including:
etc. Examples of the latter are set-up delay, dis-
connection delay, connection set-up blocking, Design the design principles may be studied
etc. In most cases the GoS requirements refer to find which ones to choose
to a fully operational network, not taking into
account network failures. One exception is Delivery the specifications are to be verified
E.550 which combines the effect of availability by the receiver of a system
and traffic congestion and defines parameters
and target values that consider their joint effect. Operation performance under varying condi-
tions, also when the system is enhanced or
11 Performance Testing modified in other ways.
Performance testing belongs to the non-func-
tional verification of a system operation. That is, When an actual system is to be modelled, the
not the does the system deliver a proper result amount of technical specification and data to be
(being a functional examination), but rather how captured may be overwhelming, although more
does the system deliver a proper result. There intriguing to concisely formulate in a model. In
may be several goals for carrying out perfor- principle, a balance between the completeness of
mance testing, including: the model and its tractability can be faced. This
makes the art of system modelling even more
Identify bottlenecks and determine configura- pleasing.
tion for maximal performance
One goal of including performance engineering
Determine capacity of a system and effects at all stages of the product life-cycle is to pre-
under overload vent problems. This also applies to systems in
operation. Measuring during operation also gives
Examine effects of introducing additional additional insight as successful services tend to
modules, new releases, etc. grow whereas unsuccessful ones usually shrink.
Monitoring, comparing results to forecasts and
One advantage by performance testing is that it the capacity can give early warnings of any sys-
could work on the final system, avoiding testing tem enhancements to be planned. Depending on
on models which might or might not fully the lead-time for enhancements this can be cru-
reflect the behaviour of the system. However, cial during operation.
this is also a drawback as a complete testing then
must be delayed until the system is fully config- Detecting faults is also a result of performance
ured coming rather late in the development engineering. Fault-tolerant systems, distributed
cycle. Moreover, the testing may become more systems and data inconsistence etc. may lead to
expensive compared to testing on a model. This degradation only, which could be detected with
may also come in addition to real traffic being performance testing possibly assisted by trac-
offered to the system, potentially impacting the ing capabilities.
customer. Due to the steady pressure on releas-
ing new systems/products, vendors often choose In one way, when a systems data and process-
not to conduct full testing, but rather restrict the ing become more distributed, the performance
examinations to so-called representative tests. risk does as well. Rapid development and auto-
However, these tests may not fully cover the generation of program code may also add to the
operational situation the system might experi- performance risk. This underlines the need for
ence. performance engineering, in particular the per-
formance prediction using adequate tools and
Performance modelling is the abstraction of a models. The following benefits of performance
real system into a simplified representation to modelling are listed in [Sing02]:
enable the prediction of performance. Note,
however, that this term commonly means differ-
Relatively inexpensive prediction of future 12 Measurements and
performance Monitoring
Basically, measurements are carried out in order
Design support allowing objective choices to to obtain quantitative estimates about the system
be made behaviour, traffic characteristics, performance
levels and so forth. A number of objectives of
Decision support for further development of measuring can be identified, including estimat-
existing systems ing performance, characterising traffic flows,
charging inputs, documenting SLA conditions,
A clearer understanding of a systems perfor- and so forth. Hence, different measurement
mance characteristics schemes have been defined. Although a network
may be correctly dimensioned, overload and
A mechanism for risk management and reduc- failure situations are likely to occur where net-
tion. work traffic management actions have to be ini-
tiated. Moreover forecast errors or approxima-
Relations with different phases of the system tions, statistical variations in the demand, also
development are illustrated in Figure 28. lead to degraded service levels. GoS monitoring
is needed to detect these problems and provide
For analytic treatment, simulation and measure- feedback for network design and traffic charac-
ments several common aspects are seen, includ- terisation. Depending on the problems faced,
ing ([Sing02]): network re-configuration, changes in routing
patters, adjustment in traffic control parameters
I Model requirements what questions are to or network extensions may be initiated.
be answered by the model
In ITU-T a number of recommendations are
II System understanding gathering all the found relating to measurements and monitoring;
performance characteristics of the system or a short list follows:
process
E.490 Introduction to the series on traffic
III Model design and development translating and performance measurements, including
the system design into a model design, deter- a survey and application of results for short
mining what to include and to exclude term (input to network management actions),
medium term (inputs to maintenance and
IV Data collection usually takes the most reconfiguration), and long term (foundations
(elapsed) time and should be started as soon for network extensions).
as possible, e.g. low-level system data not
readily available, such as the number of CPU E.491 Application of traffic measurements
seconds taken by a process to perform a par- by destination for network planning, describ-
ticular task ing the approaches of call detail recording and
direct measurements.
V Model verification ensuring the model
design reflects the system design by testing E.492 Traffic reference period
the model at various levels and model walk-
throughs in conjunction with the system E.493 Grade of Service (GoS) monitoring
designers, developers and users
E.504 Describing operation procedures for
VI Model validation if possible running the measurements (by human operators and by
model to reflect the real system using the same underlying system).
input and checking the similarity of the output
what if scenarios, predictions and answers to E.503 Potential applications of the measure-
the questions exercising the model with ments and operational procedures for obtain-
specific inputs designed to answer the ques- ing the analysis results.
tions defined in the model r equirements
E.500 Describing the principles for traffic
Iterations are often done for several of the stages intensity measurements, including a generali-
in the process, e.g. as requirements are changed, sation of the busy interval approach. Terms
a greater understanding gained and improved like estimates for daily peak traffic intensity,
data quality of data is gathered. Simpler models normal load and high load for month and year
can be produced early to give an initial predic- are introduced.
tion. As time evolves more design information
and input data are available, models could be E.501 Describing methods for estimating
refined and accuracy gained. offered traffic to a resource group and the ori-
Feasibility Requirements External design Internal design Code and test System test Roll-out
Define overall
Performance tracking
strategy
Performance
Define and manage performance risks
supervision
Performance quality control
Business volumetrics
Business volumetrics
Performance
requirements
Business and IT requirements
Acceptance criteria
Initial sizing
Technology research
Performance Design participation

assessment
Performance estimation modelling
Is the proposed Can the system Can the system Can performance
solution feasible? design be meet each de- be maintained
improved tailed performance after system or
to increase and volumetric workload changes?
performance? requirement?
Define test
Provide test environment
strategy
Define and build test cases
Performance
test
Single thread tests
Volume tests
Define monitoring
strategy
Performance
monitoring
Develop monitoring capability Monitor pilot
Figure 28 Performance activities related to system development phases (based on [Sing02])
N
N
gin-destination traffic demand based on mea- 1 1
X = Xi , s2 = Xi2 N X 2 .
surements on a circuit group. N i=1 N 1 i=1
E.502, E.505, E.745 Specifies traffic and Both X and s2 are functions of a random vari-
performance measurement requirements for able and hence, random variables themselves,
PSND/SIDN, common signalling system no. defined by a distribution called the sample distri-
7 networks, B-ISDN exchanges, respectively bution. The first is the main estimator for the
population mean value;
E.743 Identifies the subset of measurements
E[X] = m
(of E.505) to be used for dimensioning and
planning of common signalling system no. 7 s2/N is an estimator of the variance of the sample
networks. mean, X , that is:
s2
The type of measurements and parameters to be 2 X =
estimated must be in accordance with the objec- N
tive of the measurements. Defining the measure-
ment procedure implies balancing the technical The accuracy of an estimate of a sample parame-
and administrative efforts against information ter is described by the confidence interval. The
gained. According to the stochastic nature of confidence interval specifies how the estimate
traffic, a time limited measurement gives obser- is placed relatively to the unknown theoretical
vation of certain realisation of the traffic pro- value with a given probability. Given the
cess. Thus, this can be seen as a sample of the assumptions above, this confidence interval of
underlying traffic process. Repeating the mea- the mean value becomes:
suring one often sees different values. Therefore,
s2
one can commonly only state statistical mea- X tN 1,(1/2) ,
sures of the traffic process, e.g. a mean value, N
variation. Having full information of the traffic where tN-1, (1-/2) is the upper (1 /2) per-
process implies that the distribution function is centile of the t-distribution with N 1 degrees of
known. freedom. (1 ) is called the level of confidence
and expresses the probability that the confidence
A traffic process that is discrete in state and con- interval includes the unknown theoretical mean
tinuous in time can, in principle, be measured by value. When N becomes larger the t-distribution
combining: converges to a Normal distribution.
Measurements of number of events, e.g. num- Note the assumption of independence of the
ber of requests, number of errors. sampling events. This assumption might be bro-
ken if the observations are made too close as the
Measurements of time intervals, e.g. process- values then can be correlated. Similar for simu-
ing times, session times, waiting times. lations, different starting seeds should be applied
to strive for independence of the simulation
Two broad categories of measurement methods traces.
are; i) continuous measurements, and, ii) discrete
measurements. Another categorisation is, i) pas- 13 Traffic Handling
sive measurements, and ii) active measurements. Mechanisms
In the latter case test traffic is typically injected Three main groups of traffic handling mecha-
into the system, possibly influencing the result- nisms are:
ing system behaviour and performance and fac-
ing the challenge of defining patterns of the test Classification: At the ingress, the traffic is
traffic for it to mimic the process that is to be classified on a packet-by packet basis into one
estimated. of the defined service classes. Typically an
application may be looked for by means of
A general challenge for sampling, including both static field information, e.g. in the packet
measurements and simulations, is to derive an headers.
estimate of the level of confidence that should be
placed on the values estimated. Let us assume Traffic conditioning (policing and marking):
that we have N independent and identically dis- Commonly some form of policing to some or
tributed observations {X1, X2, , XN} of a ran- all of the defined service classes, to ensure
dom variable with unknown finite mean value m that usage remains within prescribed levels.
and finite variance 2. This is particularly vital for the more appreci-
ated classes, which should often receive better
The mean and variance of the sample are defined treatment, and thereby result in higher cost to
as: the network operator. Policing refers to decid-
ing on a packet-by-packet basis whether the Adding more resources increases the carried
packet stream for a particular flow lies within traffic;
a specified contracted profile. The profile is
f
commonly defined by mean rate and peak = Ei f > 0.
n1
rate/peakedness. If the contracted terms are
broken, appropriate actions are taken. Actions In a pure loss system Ei . f expresses the
may be to drop packets, to delay packets and improvement function that is positive for a finite
to mark packets. number of circuits due to the convexity of
Erlangs B formula.
Differential treatment according to service
class: At every point along a path, mecha- One wants to minimise the cost C for a given
nisms are deployed to ensure that each class carried traffic A:
gets the appropriate treatment. This is com-
monly achieved by means of two complemen- Min{C} given A = f(n1,n2, nK).
tary mechanisms; i) establishing separate
queues for each class and using scheduling Introducing Lagrange multiplier and
algorithm, and ii) providing a means to intelli- G = C . f, that is:
gently discard within a particular queue. For
the former, an algorithm has to determine Min{G(n1, n2, , nK)} = min{C(n1, n2, , nK)
which queue to serve next. Examples of such . [f(n1, n2, , nK) A]}
algorithms are priority queueing combined
with class-based weighted fair queueing and a A necessary condition for the minimum solution
modified deficit round-robin. For these a pri- is:
ority queue is used for some classes and this
G f
queue is served whenever there are any pack- = ci = ci Ei f = 0
n1 n1
ets in it. For the second class of mechanisms it
is commonly differentiated between in-con- for all i. That is,
tract and out-of-contract packets.
1 E1 f E2 f EK f
= = = ... = .
c1 c2 cK
14 Network Dimensioning
A well-known principle was published by Moe Hence, a necessary condition for the optimal
in 1924: the optimal resource allocation is ob- solution is that the marginal increase of the car-
tained by a simultaneous balancing of marginal ried traffic when increasing the number of cir-
incomes and marginal costs over all sectors. cuits (improvement function) divided by the cost
This principle is applicable to all kinds of proof a circuit must be equal for all groups.
ductions, also when referring to network
resources and allocation of these onto various In case the different traffic directions have dif-
traffic types. ferent income levels, these have to be included
by replacing the unit cost ci by the fraction ci / gi
Two portions of the problem can be recognized: where gi is the income level.
i) given a limited amount of resources, how to
allocate these among the traffic types? ii) how Denoting the revenue with R(A) and the cost
many resources should be allocated in total? with C(A), the profit is given by:
An example (from [ITU-D_TE]) allocates traffic P(A) = R(A) C(A). A necessary condition for
onto links to nodes. Given a node having con- optimal profit is then given by
nections in K directions. Assume that the cost of
P (A)
a connection of a node i is a linear function of = 0,
A
the number of circuits ni:
implying
Ci = coi + ci . ni, i =1, 2, ..., K
R C
= .
A A
The total cost for that node is:
K
Hence, the marginal income should be equal to

C(n1 , n2 , ..., nK ) = C0 + ci ni the marginal cost to achieve optimal profit. This
i=1 can be expressed as:
where Co is a constant. P(n1, n2, ..., nK) =

K

The total amount of carried traffic is a function R(f (n1 , n2 , ..., nK )) C0 + ci ni ,
of the number of circuits: A = f(n1, n2, ..., nK). i=1
Network planning methodology Figure 29 Some optimisation
methods (based on [Soto02a])
Linear programming Simplex method

Operational research
Non-linear programming Gradient-based procedures

and optimisation
Flow optimisation Critical path, maximum flow, etc.
Combinatorial processes Branch and bound
Iterative processes Decision by successive comparisons
Heuristic procedures Hybrid with emphasis on constraints and

equipment characteristics

and hence the optimal solution is achieved when min fi wi + dki xki + cij yij
i k i i j
P R
= Ei f ci = 0.
ni A
where
Using the results above, this can be written as: fi cost of installing node at location i
R
= . wi variable indicating whether node i should
A
be established (wi = 1) or not (wi = 0)
The factor is the ratio between the cost of one
circuit for the traffic which can be carried addi- dki cost of link between location k and
tionally if the group were extended by one cir- location i
cuit. That is, circuits should be added to the
group until the marginal income equals the xki variable indicating whether link ki
marginal cost . should be established (xki = 1) or not
(xki = 0)
15 Optimisation Techniques
As described earlier, optimisation is a pivotal cij cost of link between location i and
element in network planning. Here a defined location j
objective is to be optimised, commonly under
a set of specified constraints. An overview of yij variable indicating whether link ij
some optimisation methods is shown in Figure 29. should be established (yij = 1) or not
(yij = 0)
One of the founding approaches is the simplex
method, which was invented in the late 1940s. Straightforwardly, this could be attacked by the
By this the use of mathematical programming simplex method. However, as integer constraints
techniques, and in particular linear programming are attached to all the variables, certain adapta-
methods quickly gained acceptance and popular- tions must be incorporated. One approach is the
ity. On the other hand, uncertainties are com- branch and bound technique, where an integer-
monly associated with the inputs and the rela- constrained variable is assigned the value 0 and
tions between subjects of the systems to be anal- 1, respectively, and the rest of the undetermined
ysed. Furthermore, in many applications of
mathematical programming, the performance
of the optimal solution could well be drastically
compromised if the actual state of the nature Box C Some Network Terms
turns out to be different than the specifications
Networks are commonly modelled by directed graphs (digraphs) whose vertices
of the model and input.
represent the network switching nodes and whose directed edges represent the
transmission links. A digraph is said to be 2-edge connected in case it takes the
15.1 Locations and Links removal of at least two edges to break the graphs into more than one discon-
Deploying a new network, both locating nodes nected component.
and interconnecting nodes are to be found. A
A graph is planar if it can be drawn on the plane in such a manner that no two
simple example is shown in Figure 31.
edges intersect (have a common point other than a vertex). The resultant em-
bedded graph is called a plane graph. Any connected planer graph embedded in
Assuming that the intermediate nodes, index i, a plane with n vertices (n 3) and m edges has f = 2 + m n faces. A face is a
are to be located as well as the connections region defined by the plane graph. f denotes Eulers number. The number of faces
between the two neighbour node sets, a simple f includes (f 1) inner faces and one outer face (the unbounded region). The pro-
optimisation formulation can be given as: tection cycles are then a set of facial cycles with special orientations.
sum of the needed capacity for these traffic
Box D Network Topologies flows individually.
A number of topologies are relevant for telecommunication networks, see Figure
30. These are often applied on different portions of the network also in a hierarchi- Considered a somewhat more complicated
cal manner. For example, the mesh network may be used between transit tele- example that the one shown in Figure 31, further
phone exchanges, the star network in the access network from a local exchange variables have to be introduced when the traffic
to its subtending concentrators, while a ring structure is commonly found in SDH
flows are to be routed between the nodes (note
networks for redundancy arguments. Bus networks may be seen for local area
that Figure 31 only gives one possible route
networks as well as subtending concentrators in a chain.
between a node i and a corresponding node k).
For a more complete network allowing for
any-to-any terminal nodes different routes
may exist between the terminal nodes.
As the parameters may be involved in a non-lin-

a) Full/complete mesh b) Ring c) Star d) Bus ear manner for more advance problems, several
principles have been proposed for solving such
Figure 30 Examples of network topologies problems:
Besides these fundamental topologies, hierarchical levels can be introduced Kuhn-Tucker

where each of these levels in principle can have different topologies.
Fibonacci search
Golden search
DSC-Powells method
variables are not necessarily obeying the integer Steepest decent
constraints. Introducing the integer constraints Simulated annealing
always limits the options and hence a worse Tabu search
objective function value can be achieved. There-
fore, all branches that give a worse optimum 15.2 Stochastic Programming
estimate than a complete solution found so far It can be shown that ordinary sensitivity or para-
can be omitted from the further search for the metric analyses do not provide an adequate app-
optimal configuration. roach for supporting decision-making when
uncertainties are involved. One way of app-
An extension of the problem formulation above roaching this is to define models that explicitly
is to introduce link and node capacities; mod- take the uncertainty into account. And, this is
elling the relations between node and link size where stochastic programming models come
and overall costs. This implies that the amount into the picture. Commonly levels of certainty
of traffic flowing is to be explicitly considered. are associated with all variables.
This also means that the cost parameters are not
given as a single value, but have to be calculated A schematic formulation of a random (linear)
depending on the overall traffic and capacity. programming problem is:
Calculating needed link and node capacity, one
is often faced with non-linear relations. For min c . x
example, the scale effect in several cases states
that the overall capacity needed when a number such that
of traffic flows are aggregated is less then the
xki
dki
i wi fi
yij
cij
Figure 31 Illustration j
of locations and inter-
connections for
intermediate nodes (index i)
A . x b, group, called first-stage decision, that must be
T(w) . x h(w), made without certain knowledge about some
xX random parameters of the mode, and another
group, called second-stage decision, that can be
Here c R n1 and b R m1 are known vectors, taken after uncertainties have been revealed.
A is a known m1 x n1 matrix, and X R n1 is
some subset that may contain integrality restric- Consider the example above; in the first stage
tions (that is the corresponding element of x the second set of constraints can be violated.
must then take on integer values only). The These constraints, however, are considered at
uncertainty is reflected through the way the co- the second stage through some corrective or
efficient matrix T and the corresponding right- recourse actions y(w) that are considered after
hand side h of the second set of constraints the actual outcome of w is known. Assuming
depend on the outcome of random event(s) w. one wants to minimise the sum of direct cost
This implies a measurable mapping between two (c . x) and expected resource costs, this can be
spaces such as T : W -> R m2 x n1 and h : W -> formulated as:
R m2, defined in a corresponding way.
min c . x + E[q(w) . y(w)],
An essential statement is that we seek related
values of x that are only based on information such that
known at the point of decision. That is, the deci-
sion cannot be based on the actual outcome of A . x b,
the random event w. The information concerning T(w) . x + W(w) . y(w) h(w),
this future event is assumed to be known through x X, Y(w) Y.
its probability distribution.
Here q and W are measurable mappings onto
This also implies that the notion of optimality is corresponding spaces, in a similar manner as
no longer absolutely obvious as the objective to h and T above.
be optimised as well as the formulation of the
constraints depend on the decision-makers atti- Referring to the formulation above, c is the first-
tude to risk. stage cost, q is the second-stage cost, T is the
technology matrix, W is the recourse matrix, and
15.2.1 Chance-constrained Programming h is the second-stage right-hand side. In general,
Consider the same optimisation formulation as the two-stage stochastic recourse model applies
above, except that the second set of constraints to a wider formulation.
is to hold with a probability of at least , where
[0,1]. That is, The division of decisions into two sets may
reflect an inherent nature of an overall problem
min c . x showing the timing of decision. That is, some
decisions may be made immediately whereas
such that others can be postponed until uncertainty has
been disclosed, or at least until additional infor-
A . x b, mation on uncertain parameters is available.
P[T(w) . x h(w)] , Hence, this represents a simple dynamic deci-
xX sion process. The first-stage decisions must be
made without certain knowledge of random
This is formulated as a chance-constrained parameters and must be chosen as to minimise
stochastic program where the second set of con- the sum of direct cost and the expected value of
straints is referred to as a joint chance-constraint future cost. The future cost, on the other hand, is
or a joint probabilistic constraint. One way of determined when an optimal second-stage deci-
generalising this is to introduce individual con- sion is made after observation of the outcome of
straints: the random parameters:
P[Tk(w) . x > hk (w)] k, k = 1, , m2 Decision on x -> observation of q(w), h(w), T(w)

and W(w) -> decision on y.
Where Tk is the kth row of T, hk is the kth compo-
nent of h, and k is the probability associated 15.2.3 Multi-stage Stochastic Programs
with the kth constraint. with Recourse
The decision process outlined in the previous
15.2.2 Two-stage Stochastic Programs section may be generalised to include a number
with Recourse of stages; that is, an alternating sequence of
For two-stage stochastic recourse programs the decisions and observations of random variables.
decision can be divided into two groups; one
such that
A . x > b,
x X,
where Q(x, f) expresses the probability that the

total cost exceeds the specified level f and k
gives the weight placed on this term.
In this manner both the expected total cost and

the probability of the budget being exceeded
are considered.
16 Example of Network
Planning Tool and
Approach
PLANITU ([PLANITU]) is a computer tool for
dimensioning and optimisation of networks,
aiming at minimum cost designs for:
0 1 2 3
Location and boundaries of exchanges

Selection of switching and transmission
equipment
Figure 32 Scenario tree with 4 Xt denotes the decisions to be made at stage t Circuit quantities, traffic routing, switching
stages and 9 scenarios and yt the random variables to be observed at the hierarchy
same stage. Similar as for two-stage programs Choice of transmission paths.
this can be illustrated as:
The main focus is telephone exchanges/nodes
observation of y1 -> decision on x1 (concentrators, local exchanges, transit
observation of y2 -> decision on x2 exchanges), transmission systems and traffic
routing.
observation of yN -> decision on xN
An overall picture of functionality and input/
For some cases it is assumed that the constraints output is shown in Figure 33. The planning
obey a Markovian criterion, i.e. the ones applica- process is outlined in Figure 34.
ble at stage t are coupled to previous stages only
through stage t 1. The main groups of input data, although depend-
ing on the situation at hand, are:
This formulation intuitively invites the scenario
tree description, see Figure 32. Existing network configuration: i) exchange
locations and boundaries, ii) exchange and
Even though this kind of formulation looks transmission equipment, iii) geographical lay-
inspiring for thinking of future options in net- out of subscriber and inter-exchange network.
work planning, it turns out that the algorithms
for solving the corresponding problems often Demand forecasts: i) subscribers location
must be tailored to the problem at hand. Hence, and category, ii) traffic quantities and dis-
few attempts have been made to devise general- persion (traffic matrices).
purpose algorithms for multistage stochastic
integer programs. Switching equipment: i) capacity subscriber
lines, trunk lines, call attempts, etc., ii) costs
15.3 Minimum Risk Problem subscriber trunk links, exchange units,
An objection to the recourse approach is that the ii) traffic handling specification, iv) floor
expected cost is to be minimised. However, a space requirements.
risk-resistant decision-maker may prefer to min-
imise the probability of total cost exceeding Transmission equipment: i) capacity, ii) cost
some prescribed threshold value, say f (e.g. the system, terminal equipment, repeaters, inter-
bankruptcy level or simply a budget limit). This faces to other systems, iii) attenuation and
could be considered by introducing the corre- loop resistance.
sponding measure in the objective function:
Buildings and ducts: existing situation, poten-
min c . x + E[q(w) . y(w)] + k . Q(x, f), tial extensions.
Quality considerations: i) Grade of Service, 17 Concluding Remarks
ii) transmission plan. This article has provided a brief introduction to
the topic of network planning. Although rela-
Due to the interdependencies between the cost tions with enterprise management, financial
contributions, such as number and location of issues and a number of more technical areas are
exchanges and transmission equipment, it is listed, there is a vast amount of literature and
hard to optimise the overall network in one go. lists of supplementary issues that could be in-
Hence, an iterative procedure has been defined cluded. For example, the optimisation methods
for PLANITU allowing for a modularised themselves have been treated in an extensive list
approach where sub-optimisation is used for of books over the years.
each stage. This is illustrated in Figure 35.
Telecom management
Forecast Scenario Policy, Final

approval approval resources decision
Detailed plan
Network planning management
Policy evaluation, Product research

strategic Planning,
digitalisation process
(incl. ISDN)
Fundamental plans
Network Planning criteria, Price lists

planning e.g. GoS
Land and building

planning
Products,
Forecasting
cost structures
Detailed network
data
Figure 33 Task blocks and
All other Data interfaces for network
data collection pre-processing Traffic
measurements planning
(adapted from [PLANITU])
External
sources
Trans-
Switching
mission
Traffic General Subscriber Network
measurements economy inventory Equipment costs structure
Subscriber
forecast
Exchange locations
Traffic and boundaries
forecast
Switching and
transmission
Quality and
equipment
resources Figure 34 The network
planning process
PLANITU
(from [PLANITU])
Input: Output: The increasing awareness of network planning
is particularly crucial when efficiency improve-
ments are to be implemented in an organisation.
Exchange Boundaries Locations In fact, the systematic views that network plan-
locations Circuits Distances
ning techniques provide also support the
smoother interactions between units within the
organisation as well as between the operators
and other actors in the market.
Exchange Locations Boundaries
boundaries Cost/erlang Subs/exchange
References
[DSL-058] DSL forum. Multi-Service Architec-
ture & Framework Requirements. September
Inter-exchange Boundaries Traffic 2003. Technical Report TR-058.
traffic Subs/exchange distribution (www.dslforum.org)
[Macf02] MacFadyen, N W. Traffic characteri-

sation and modelling. BT Technology Journal,
Inter-exchange Traffic Circuits 20 (3), 1430, 2002.
circuits Cost/circuit Routing
[PLANITU] ITU-D PLANITU Network Plan-

ning Software. Februar 25, 2004 [online] URL:
www.itu.int/ITU-D/planitu/index.html 2003
Transmission Circuits Cost/link
systems Distances Cost/circuit
[RFC372] Awduche, D et al. Overview and
Principles of Internet Traffic Engineering. IETF
RFC 3272, May 203.
Network Current Improved
changes network network
[Sing02] Singleton, P. Performance modelling
what, why, when and how. BT Technology Jour-
nal, 20 (3), 133143, 2002.
Figure 35 Activities for A key message in this presentation is that net- [Soto02] Soto, O. Integrated Planning method-
exchange network planning, work planning is multi-faceted and has relations ology. Scenario analysis and data gathering.
based on [PLANITU] with (almost) all other types of activities going ITU-D workshop on network planning strategies
on in a network operator. Hence, it is important for evolving network architectures, Nairobi,
to balance the system portfolio insight with Kenya, 711 October 2002.
skills in planning techniques as well as the finan-
cial means for comparing results. In addition [Soto02a] Soto, O. Requirements for decision
comes the different time horizons one has to making in network evolution, Strategic planning
deal with from strategic planning to tuning of and new Technologies, Solution mapping for
parameters for traffic handling. geo-scenarios. ITU-D workshop on network
planning strategies for evolving network archi-
A key point for an operator is to provide the set tectures, Nairobi, Kenya, 711 October 2002.
of services asked for by the customers and this
Figure 36 Illustrating the should be done more cost-effectively and faster
awareness of diverse demands than the competitors. As shown in Figure 36 the
and offering adequate service customers demands can often be arranged along
levels and price levels may a scale with the corresponding service levels.
improve the operators Whether this is done by a single network or not
margins is not shown in the illustration.
Price (service level) Price (service level) Price (service level)
Number of customers Number of customers Number of customers
Optimization-Based Network Planning Tools
in Telenor During the Last 15 Years
A Survey
RALPH LORENTZEN
1 Introduction modelling whilst Veritas was given the task of

The rapid technological development in the implementing the models. Later on the consul-
telecommunications field during the last two tants at Veritas moved to the consultant com-
decades made it necessary for the operators to pany Computas, and the cooperation with Com-
repeatedly reevaluate the structure, design and putas has continued to this day.
application of their networks. In order to estab-
lish cost-effective network design and utilization From the outset a good cooperation was estab-
many telecommunication operators developed lished with network operators in the field who
and used optimization-based network planning allocated time and energy to specification and
Ralph Lorentzen (67) is daily tools. This happened also in Telenor. The ratio- testing.
leader of the company
nale was that the frequent technological shifts
Lorentzen LP Service which
does consulting work within did not give the planners sufficient time to The initiative was triggered by a situation where
mathematical programming. He acquire an intuitive feeling of what constituted a local cable TV network planner compared two
retired July 1, 2003 from his job
good network designs. So the requirement for alternative network layouts which both seemed
as senior scientist in Telenor
R&D. Ralph Lorentzen gradu- comprehensive, cost-effective and robust reasonable, but discovered afterwards that the
ated in mathematics at the Uni- designs excluded simple back of the envelope cost of one of the layouts was twice the cost of
versity of Oslo, where he
or spread sheet calculations. The idea was that the other. The group worked together on and off
worked for some time as assis-
tant professor giving lectures in mathematical models could be used to optimize over a period of about 15 years and designed and
statistics and operations network structures and thus contribute to reduce implemented a series of network planning tools
research. He worked as distribu-
costs, ensure network reliability and improve based on integer programming. Some of the
tion planner in Norske Esso, as
principal scientist at Shape operational efficiency. tools will be described in the sequel, namely
Technical Centre, as systems KABINETT, ABONETT, FABONETT,
engineer in IBM, and as chief
The approach was said to be pragmatic but thor- PETRA, MOBANETT, and MOBINETT.
consultant in Control Data Nor-
way, before joining Telenor R&D ough; heuristic but nevertheless intelligent. The
in 1985. aim was not to give a black box answer, but a 2 KABINETT A Cable TV Net-
ralph.lorentzen@tiscali.no deeper understanding of the issues and of the work Planning Tool
consequences of proposed solutions. The opti-
mization tools were not meant to give a final 2.1 General
answer, but to constitute instruments to be inte- At the time when the cable TV network planning
grated into the decision process of the planners. tool KABINETT was developed the networks
should have a tree structure. Manual planning
Common to all the tools were that they were was time-consuming, and the planners had
based on mathematical programming. The dif- capacity for analyzing a few alternatives only.
ferent network components were modeled as One important feature of the planning problem
decision variables with associated costs. Techno- was to secure that the subscribers received suffi-
logical constraints were modeled by equations cient voltage with a minimum number of ampli-
and inequalities that the decision variables had fiers. Also, the cost of civil work constituted a
to satisfy. Commercially available optimization large portion of the total cost. Of course, the
software together with tailor-made program planning problem became much simpler later
modules were used to find cost-effective solu- when the decision was made to go for star net-
tions. The fact that the tools were to be used by works in lieu of tree networks.
others than the ones who developed them im-
plied that they had to be incorporated in high 2.2 The Cable TV Network Design
quality user interfaces and linked up to relevant Problem
data sources in an appropriate way. In general Here will be given a short description of the
the effort spent on the development of user inter- cable TV network design problem KABINETT
faces exceeded the effort spent on implementa- attempted to solve.
tion of optimization algorithms by a factor of
three to four. We were given a signal source and a set of sub-
scribers. The signal source was to be connected
In 1986 Telenor R&D together with the consul- to the subscribers by copper cables via ampli-
tant company Veritas established a group of fiers, splitters and taps.
three people who started building optimization
models for key network design problems. The amplifiers, splitters and taps had to be
Telenor R&D was mainly responsible for the placed in cabinets. There were two types of cabi-
Signal source assumed to have access to a calculation program
D1/D2
network (which did not perform any optimization) for
voltage and attenuation calculations between the
signal source and the D3 amplifiers, which they
could use as an aid for deciding where to place
additional amplifiers.
Splitter
(no attenuation
calculations From the signal source the signal might go via split-
performed) ters on its way to the amplifiers. This part of the
network will here be called the D1/D2 network.
D3 amplifier From an amplifier the signal could go via split-

ters on its way to the taps. There were two types
of taps, namely terminal taps and transit taps.
D3 and Terminal taps had subscriber ports from which
subscriber
network the signal could only proceed directly to sub-
Splitter
scribers. A transit tap had an additional port, the
D3 port, from which the signal could proceed to
a splitter or to another tap. Normally the D3 port
Transit tap with subscribers would be connected to another tap and not to a
splitter. A subnetwork connecting a D3 amplifier
(attenuation to the taps underneath it will here be called a D3
calculations
performed) network. If the D3 amplifier had sufficient volt-
age to support its associated taps with their sub-
Splitter (possible, but unusual)
scribers, the D3 network was said to be feasible.
Otherwise it was said to be infeasible. All the
D3 networks that KABINETT proposed had of
course to be feasible.
Every subscriber was connected directly to a

Termianl tap with subscribers subscriber port on a tap. The subnetwork con-
necting the subscribers to the taps will here be
called the subscriber network.
Based on the location of the signal source, costs

Figure 2.1 Possible network nets, large cabinets and small cabinets. Ampli- and locations of candidate trenches, costs and
structure fiers must be placed in large cabinets while split- candidate locations of D3 amplifiers, splitters
ters and taps could be placed in small cabinets. and taps, costs and attenuations of candidate
A cabinet could (in addition to the amplifier in cable types, and subscriber voltage require-
large cabinets) contain an arbitrary number of ments, KABINETT tried to find the least costly
splitters and taps. network design. The user could specify parts of
the design and leave it to KABINETT to com-
The cables were placed in trenches. The trenches plete the design at the lowest possible cost.
formed a network which was called the trench
network. The trench network was assumed to The algorithms used in KABINETT were heuris-
have a tree structure. tic, so a theoretically optimal solution to the net-
work design problem was not guaranteed.
In KABINETT the signal passed via exactly one
amplifier on its way from the source to the sub- Figure 2.1 shows a possible network structure.
scriber. These amplifiers were called D3 ampli- The splitters could have more than two exit
fiers and were placed in the network by KABI- ports, and the exit ports could have different
NETT such that the signal that the subscribers attenuations. There were also in general many
received had sufficient voltage. In reality it could types of taps where each tap type was character-
be necessary to place additional amplifiers be- ized by its number of exit ports and the attenua-
tween the signal source and the D3 amplifiers. In tion associated with each individual exit port.
KABINETT there was no voltage calculation on One could have several splitters and transit taps
the stretch between the signal source and the D3 in series, but normally the signal would not pro-
amplifiers. Consequently KABINETT did not ceed from a tap to a splitter.
make any analysis of the requirement for ampli-
fiers in addition to the D3 amplifiers. However, Ideally the trench network, the D1/D2 network,
once the D3 amplifiers had been placed, this was the D3 networks and the subscriber network
generally straightforward. The users were shold be optimized together. Instead a separate
cost minimization of the trench network was The output could be fed into an interactive cal-
done first. Then an optimization of the sub- culation program (not described here) for place-
scriber network was done followed by the opti- ment of amplifiers in the D1/D2 network. This
mization of the D3 networks, and finally the program also gave a schematic description of the
D1/D2 network was determined. The separate solution.
optimization of the trench network was partly
justified by the fact that usually well above 70 % 2.5 Determining the Trench Network
of the total network cost was the cost of trenches. The trench network cost minimization problem
is an example of the classical Steiner problem
2.3 Input in undirected graphs. A network of candidate
The input to KABINETT specified trenches was given, and the problem was to find
the cheapest possible subnetwork (which must
the location of the subscribers to be connected necessarily be a tree) which connected a subset
to the network of nodes. The nodes to be connected included
the location of the signal source node the signal source node and the subscriber nodes.
the location of the nodes where D3 amplifiers, In addition the planner could include amongst
splitters and taps could be placed the special nodes some of the equipment nodes,
trench route alternatives i.e. candidate nodes for placement of D3 ampli-
cable types fiers, splitters and/or taps.
splitter types
tap types There could be subscribers who were connected
the D3 amplifier type to the rest of the candidate trench network by
cabinet types two or more candidate trenches. However, one
might want to avoid using these candidate
The trench routes consisted of trench sections, trenches for transit cables to other subscribers.
each of which was characterized by geographical This was achieved by multiplying the cost of
location, length, and cost per metre. these trenches by a large number before the opti-
mization.
The cable types were characterized by attenua-
tion and cost per metre. In KABINETT an approximate solution to the
Steiner problem was found using Rayward-
The splitter types were characterized by the Smiths algorithm [1].
number of ports and the attenuation at each indi-
vidual port. 2.6 Determining the Subscriber
Network
The tap types were characterized by the number The model used for the subscriber network was
of ports and the attenuation at each individual a classical capacitated plant location model.
port and whether they were transit taps or termi-
nal taps. The subscribers were to be connected to taps by
cables. In this module all taps were considered
KABINETT operated with one type of D3 to be terminal taps. One cable type only was
amplifier which was characterized by cost and considered for each subscriber. In practice the
the voltage at the exit port. planner would assume a default cable type for
all subscribers in an initial run. As a result of in-
There were two cabinet types, namely large and specting the initial solution he could change the
small. Each type was characterized by cost. cable type for some of the subscribers and run
the optimization again.
2.4 Output
The output from KABINETT described the pro- Each candidate tap was given a cost which was
posed solution to the cable TV network design the sum of the tap cost, the cost of a small cabi-
problem. It showed net, and the cost of the copper cable needed to
the trenches to be dug connect the tap to the closest equipment node
what types of cables should be used where in the direction of the signal source node.
where D3 amplifiers should be placed
what type of splitters and taps should be The cost of connecting a given subscriber to a
placed where tap in a given equipment node was set to the sum
which subscribers should be connected to of the material and labour cost associated with
which ports on which taps connecting them with the appropriate cable.
a table showing labour costs and bill of mate-
rials Attenuation was not considered in the subscriber
network optimization. Voltage and attenuation
considerations were postponed to the D3 net- 2.7 Determining the D3 Network
work optimization.
2.7.1 General
The subscriber network optimization model is The method for determining the D3 networks
described in mathematical terms below. consisted of three algorithms, namely the con-
struction algorithm, the partitioning algorithm
The following notation is used: and the move/exchange algorithm.
Subscripts Typically, KABINETT started with attempting

to cover all the taps with one D3 amplifier. If
n denotes equipment nodes, this failed, two D3 amplifiers were tried, etc. If
s denotes subscribers, the minimum number of D3 amplifiers which
t denotes terminal tap types. KABINETT found was needed was K, and the
parameter extra was set to a positive integer,
Costs then KABINETT tried to find the least expen-
sive solution with K, K + 1, , or K + extra
cns denotes the cost of connecting subscriber s to amplifiers.
equipment node n;
cnt denotes the cost of placing a terminal tap of The construction algorithm took as input a sub-
type t in equipment node n. This cost is the sum set of the taps with their voltage requirements
of the tap cost and the cost of a cable from the and placed a D3 amplifier in a node such that
tap to the nearest equipment node on the route to either a feasible D3 network was formed or an
the signal source node. (infeasible) D3 network was formed where the
voltage requirement in the D3 amplifier node
Capacities was as low as possible.
pt denotes the number of subscriber ports on a The partitioning algorithm was a one pass algo-
tap of type t. rithm for partitioning the taps into a specified
number of subsets and constructing a (feasible
Variables or infeasible) D3 network for each of the subsets.
xns = 1 if subscriber s should be connected to a The move/exchange algorithm was an iterative

tap in equipment node n, and = 0 otherwise. algorithm which tried to improve on the solution
found by the partitioning algorithm. The move/
ynt = 1 if a tap of type t should be placed in exchange algorithm tried first of all to make an
equipment node n, and = 0 otherwise. infeasible solution feasible, and tried thereafter
to find a solution with reduced cost.
Using this notation the optimization problem can
be formulated as follows: Before we treat these algorithms we will
describe the splap concept.
minimize nscnsxns + ntcntynt (2.1)
2.7.2 Splaps
subject to In the D3 network KABINETT could place
more than one splitter and/or tap in an equip-
nxns = 1 (2.2) ment node. The result was a collection of split-
ters and taps which could be characterized by its
sxns tptynt (2.3) total cost, a set of D3 ports with their individual
attenuations, and a set of subscriber ports with
xns tynt (2.4) their individual attenuations. Such a collection
of splitters and taps in a node was in KABI-
where xns and ynt are 0 1 variables. NETT denoted a splap (see Figure 2.2).
This optimization problem was solved using a Before the D3 network optimization was started
conventional branch and bound code. KABINETT compiled a list of candidate splaps
based on the splitter and tap types in the input.
The planner could preset part of the solution by As mentioned earlier the D3 port of a transit tap
fixing some of the xns or ynt variables to 1. would normally be connected to another tap and
not to a splitter, and the splap list was based on
When the subscriber network was determined, collections of splitters and taps where this was
the voltage requirement at the taps were calcu- the case. The planner decided the size of this list
lated and recorded. by specifying the maximal number of subscriber
and D3 ports of the splaps.
A splap S was said to be inferior to another splap Figure 2.2 Transformation of
S if the numbers of subscriber and D3 ports of S a collection of splitters and
were less or equal to the corresponding numbers taps in a node to a splap
for S, and the attenuations of the ports of S were
greater or equal to corresponding ports of S.
The splap list would not contain any splap which
was inferior to another splap on the list.
Splap
2.7.3 The Construction Algorithm
First an outline of the algorithm will be given:
Initially the signal source node was used as what

is called an attraction node.
Subscriber D3 ports
ports
The voltage requirements in the nodes in the
trench tree were calculated starting from the
taps and proceeding towards the attraction node.
Whenever a junction node in the tree was
reached a splap which minimized the voltage
requirement was placed. ment of n, and establish this as the voltage
requirement of n. In both cases label node n
If a node was reached where the voltage require- as treated. Go to 3.
ment was greater than the output voltage of the
D3 amplifier, this node was made the attraction 3 The first time a voltage requirement of a node
node the first time this happened. All subsequent is calculated which exceeds the voltage deliv-
voltage calculations were then made proceeding ered by a D3 amplifier, change the attraction
towards the new attraction node. node from the signal source node to this node.
Go to 2.
Based on criteria to be detailed below a D3
amplifier was then placed in the vicinity of the 4 If the voltage requirement of the attraction
attraction node such that a (feasible or not feasi- node does not exceed the voltage delivered by
ble) D3 network was formed. a D3 amplifier (i.e. the attraction node has not
been changed from the signal source node),
The algorithm is described in more detail below. place a D3 amplifier in the signal source node
and terminate the algorithm. Otherwise, try
Algorithm for constructing a candidate D3 iteratively to place a D3 amplifier in nodes
network for a subset of taps: obtained by fanning out from the attraction
node until either a node is found whose volt-
1 Initialize the attraction node to be the signal age requirement does not exceed the voltage
source node, label all edges in the trench tree delivered by a D3 amplifier or a local mini-
as untreated, label all the taps as treated and mum of the voltage requirement is found
label the remaining nodes in the trench tree as (which is larger than the voltage delivered by
untreated. The treated nodes are characterized a D3 amplifier). In both cases a D3 amplifier
by the fact that their voltage requirement is is placed in the node found.
established.
If the voltage requirement of the node where the
2 If all nodes are treated, go to 4. Otherwise, D3 amplifier was placed exceeded the voltage
find an untreated node n which is farthest delivered by a D3 amplifier, the resulting D3
away from the attraction node. Pick an un- network was infeasible. In that case the infeasi-
treated edge connecting n to a treated node, bility was defined to be equal to the difference
calculate and save the contribution to the volt- between the voltage requirement and the D3
age requirement of n caused by this node (i.e. amplifier voltage. If the D3 network was feasi-
the voltage requirement of the treated node ble, the infeasibility was defined to be 0.
plus the attenuation in a cable between n and
the treated node), and label the edge as 2.7.4 The Partitioning Algorithm
treated. Repeat until all the edges coming into In KABINETT several variants of the partition-
n from treated nodes are treated. If then there ing algorithm were implemented. Here, how-
is only one treated node adjacent to n, set the ever, only one of these variants will be described.
voltage requirement in n equal to the contribu-
tion to the voltage requirement of n caused The partitioning algorithm took as input a posi-
by this node. Otherwise, place in n a feasible tive integer K which represented the number of
splap which minimizes the voltage require- D3 networks.
A description of the partitioning algorithm fol- If N1 is infeasible with t in N1 or if N2 is infeasi-
lows. ble after t is moved to N2, let rV(t, N1, N2) denote
the reduction in the sum of voltage requirements
Partitioning algorithm for creating K obtained by moving t to N2.
(feasible or infeasible) D3 networks:
Now we can describe the move/exchange algo-
1 Start with all the K networks empty. Select the rithm.
K taps which are farthest apart, i.e. the K taps
t1, , tK which maximize mini<j d(ti, tj). Move/exchange algorithm between a
Include one of these taps in each of the K collection of D3 networks:
networks. Go to 2.
1 If all networks are feasible, go to 2. Other-
2 For each network find its close node, i.e. the wise, calculate r = max r1(t, N1, N2) where the
closest node which has not been included in maximization is done over all N1, N2 and t in
any of the networks. If none of the networks N1. If r > 0, make a move which gives the
have a close node, terminate (every node be- infeasibility reduction r. Repeat until either all
longs to a network). Otherwise, order the networks are feasible or r 0. If all networks
close nodes in a list according to how close are feasible, go to 2. If r 0 and 3 has been
they are to their associated network. Label executed stop, otherwise, go to 3.
all the close nodes as untreated. Go to 3.
2 Calculate r = max rC(t, N1, N2) (as usual
3 If there are no untreated nodes, go to 5. Other- maximization over the empty set is defined
wise, go to 4. to be ), where the maximization is done
over all t, N1 and N2 for which rC(t, N1, N2) is
4 Pick the first untreated close node on the list. defined.
If this close node does not contain a tap, in- If r > 0, make a move which gives the cost
clude the node in its associated network, and reduction r. Repeat until r < 0. If 4 has been
go to 2. If the close node contains a tap and executed, terminate. Otherwise, go to 4.
the node can be included in its associated net-
work without making the network infeasible 3 Calculate max rV(t, N1, N2) and max rV(t, N2,
(this is checked by the construction algorithm), N1), where the maximizations are made over
make this inclusion and go to 2. Otherwise, all t, N1 and N2 for which the expressions are
label the close node as treated and go to 3. defined. Let t1 and t2 be taps which maximize
the two expressions respectively. If moving t1
5 Include the first close node on the list and from N1 to N2 and t2 from N2 to N1 results in a
go to 2. reduction in the sum of infeasibilities, do these
moves. (This is an exchange.) Repeat until
2.7.5 The Move/Exchange Algorithm all networks are feasible or no reduction in
When a solution was found by the partitioning infeasibility results. If all networks are feasi-
algorithm, it was natural to try to improve on it. ble, go to 4. Otherwise, go to 1.
Whether the solution is feasible or not, attempts
were made to change it into a feasible solution 4 Calculate max rC(t, N1, N2) and max rC(t, N2,
with the same number of D3 amplifiers, and N1), where the maximizations are made over
with the lowest cost possible. This was the func- all t, N1 and N2 for which the expressions are
tion of the move/exchange algorithm. defined. Let t1 and t2 be taps which maximize
the two expressions respectively. If moving t1
In the description of the move/exchange algo- from N1 to N2 and t2 from N2 to N1 results in
rithm the following concepts will be used: a reduction in cost, do these moves. Repeat as
long as cost reduction results. Then go to 2.
Let N1 and N2 denote D3 networks and let t
denote a tap. 2.8 Determining the D1/D2 network
Once the D3 networks were determined the de-
If N2 is infeasible with t in N1, let r1(t, N1, N2) termination of the D1/D2 network was straight-
denote the reduction in the sum of infeasibilities forward. KABINETT just cabled up the trench
obtained by moving t to N2. tree which connects the D3 amplifiers with the
signal source node, and placed appropriate split-
If N1 is feasible with t in N1, and if N2 is feasible ters wherever necessary.
after t is moved to N2, let rC(t, N1, N2) denote the
reduction in cost obtained by moving t. As mentioned earlier no voltage and attenuation
calculation was done in the D1/D2 network, so
KABINETT did not place amplifiers in this net-
work.
Subscriber Figure 3.1 Outline of a
subscriber network with one
supply point at subscriber
switch level
Distribution
points
Cross connection
points
Supply
point
2.9 Possibilities for Manual The terminology used will first be described.
Modification of the Solution
One of the weaknesses of KABINETT was that We were given a set of new subscribers, each of
the placement of the splaps in the construction whom was to be connected by cables to one of a
algorithm was not optimized. The solutions set of alternative supply points.
tended to have too many splaps, which implied
too many cabinets, in the D3 networks. The network between a supply point and the
subscribers connected to this supply point had
Facilities were therefore incorporated into a hierarchical structure.
KABINETTs user interface which made it easy
for the planner to change locations of splaps and The subscribers could be connected to distribu-
amplifiers in the D3 networks and to check the tors placed in distribution points or directly to
feasibility of modified solutions. The user could cross connectors placed in cross connection
also change the cable type to be used on a speci- points. The subscribers which were connected
fied stretch. These facilities were described in to distributors were called ordinary subscribers
KABINETTs user manual. whilst the subscribers which were connected
directly to cross connectors were called special
3 ABONETT A Subscriber subscribers. The distributors were connected to
Network Planning Tool cross connectors, and the cross connectors were
connected to subscriber switches or RSUs. All
3.1 General connections were made with cables selected from
After the modeling of KABINETT was com- a set of cable types with different capacities.
pleted it was found that a similar tool could
solve the problem of extending a telephone sub- A supply point could be a cable from a sub-
scriber network to connect new subscribers in a scriber switch, an RSU, an existing cross con-
cost-effective way. The absence of attenuation nector or an existing distributor. Figure 3.1 illus-
problems made it possible to be somewhat more trates a subscriber network where the supply
ambitious in the modeling and go for a simulta- point is a subscriber switch or an RSU.
neous optimization of the trench and cable net-
work. The resulting tool was called ABONETT. The cables were placed in trenches and possibly
It was found that ABONETT resulted in solu- in ducts. There could be a requirement for plac-
tions which on the average were 10 % less ing cables or empty ducts in trenches leading
costly, and that the work involved in the design from supply points up to specified cable or duct
phase was reduced by at least 50 %. termination points in order to facilitate future
extensions of the network.
3.2 The Subscriber Network Design
Problem There was also a requirement for reserve capac-
Here will be given a short description of the sub- ity in cross connection points and distribution
scriber network design problem that ABONETT points, also in order to facilitate future exten-
attempted to solve. sions.
The ABONETT user specified the subscribers, The distribution points were likewise character-
candidate supply points, candidate trenches, can- ized by their geographical location, the types of
didate cross connection points and distribution distributors which could be placed there, the
points, and requirements for cables/ducts to required spare capacity in percent, and maximal
specified cable/duct termination points. distance to subscribers they could offer connec-
tion to.
ABONETT tried to find a network design which
satisfied all the given requirements and which The cross connector types, distributor types,
was as inexpensive as possible. cable types, and ducts were characterized by
capacity and cost.
The ABONETT user had full control over the
solution in the sense that he could specify as 3.4 Output
many characteristics of the design as he wanted. The output from ABONETT described the pro-
In the extreme he could specify the solution posed solution to the subscriber network design
completely and use ABONETT just to check the problem.
validity of the solution and calculate its cost.
It listed the supply points, cross connection
3.3 Input points and distribution points with the equipment
The input to ABONETT described the sub- selected by ABONETT, which trenches should
scribers to be connected to the network, the sup- be dug, and which cables and ducts should be
ply point(s), cross connection and distribution used. A simplified bill of materials was also
point alternatives, reserve capacity requirements, given.
cable or duct requirements (if any), and trench
route alternatives. 3.5 The Trench and the Cable
Networks
The trench routes consisted of trench sections, Two networks were introduced, namely an un-
each of which was characterized by geographical directed trench network and a directed cable net-
location, length and the required trench type. In work.
certain cases it was desired to reserve trench
usage. If a subscriber could connect to the net- The edges in the trench network were the trench
work via more than one trench alternative sections, and the nodes were the end points of
through his property, it was important to prevent the trench sections.
ABONETT from considering the possibility of
using these trench alternatives as transit The nodes in the cable network represented
trenches for reaching other subscribers. Techni- existing supply points, existing and candidate
cally this was treated in ABONETT by adding a cross connection and distribution points, sub-
high artificial cost to such trenches. scribers, and cable/duct termination points. In
addition an extra node was introduced, namely
The trench types were described by cost per the root node.
metre.
The arcs in the cable network connected the root
The subscribers were characterized by their geo- node to the supply points, the supply points to
graphical location, the capacity they required, the cross connection points and cable/duct termi-
whether they should be connected to a distribu- nation points, the cross connection points to the
tion point or directly to a cross connection point, distribution points and special subscribers, and
and whether the adjacent trench sections should the distribution points to the ordinary subscribers.
be reserved.
The cable network thus constituted a rooted
The supply points were characterized by their directed acyclic graph.
geographical location, whether they were
switches/RSUs, cross connectors or distributors, 3.6 The Trench, Cable and
capacity, reserve capacity requirement, and max- Combined Optimizations
imal distance to cross connection points, cable In order to keep the computation time within
termination points, distribution points or sub- reasonable limits three optimizations were made;
scribers they could offer connection to. namely the cable optimization, the trench opti-
mization, and the combined optimization.
The cross connection points were characterized
by their geographical location, the types of cross In the cable optimization ABONETT tried to
connectors which could be placed there, the re- find the cheapest cabling which satisfied all the
quired reserve capacity in percent, and maximal requirements under the assumption that all the
distance to distribution points or special sub- trenches are available free of charge.
scribers they could offer connection to.
In the trench optimization ABONETT first tried pj is the capacity required in the cable leading to
to find the cheapest trench network which con- subscriber or cable termination point j.
nected together the subscribers, the supply
points, the cross connection points, the distribu- cijt is the cost of a cable of type t carrying signals
tion points, and the duct/cable termination from node i to node j. This cost included the cost
points. Then ABONETT tried to find the cheap- of any technical equipment (cross connector,
est cabling in this trench network which satisfied distributor) this cable necessitated in node j, and
all the requirements. it was calculated under the assumption that the
cable followed the shortest path from i to j in the
In the combination optimization ABONETT given trench network. For the trench optimiza-
considered both trench costs and cable costs. tion this shortest path is the only path leading
However, the only trenches which were taken from i to j. For the cable optimization the shortest
into consideration were those selected either in path was calculated using Dijkstras algorithm.
the trench optimization or in the cable optimiza-
tion (or both). Trenches which were selected Ducts were looked upon as a cable type which
both in the trench and cable optimization, and in served a special class of subscribers, namely
the trench optimization were selected to contain duct termination points.
either two or more cables or at least one cable
terminating at a subscriber, are made available at The integer linear programming problem can be
no cost. All other trenches which were selected formulated as follows:
either in the cable or the trench optimization
were made available at their real digging cost if minimize z = ctij xtij
they belonged to a circle in the resulting trench ijt
network, and at no cost if they did not belong to subject to

a circle.
at xtij ntjk (3.1)
The solution resulting from the combination it kt
optimization was ABONETTs suggested solu- if j is a cross connection or distribution point,

tion to the network design problem.
ntij pj (3.2)
3.7 Finding the Cheapest Possible it
Trench Network if j is a subscriber or a cable termination point,

In the trench optimization the cheapest trench
network connecting the subscribers, cross con- ntij bi (3.3)
nection points, distribution points and the duct/ jt
cable termination points was sought. It is obvi- if i is a supply point,

ous that the cheapest network is a tree.
ntij at, (3.4)
This was an example of the classical Steiner
problem in undirected graphs. In ABONETT xtij = 0 or 1, ntij integer. (3.5)
an approximate solution to the Steiner problem
is found using Rayward-Smiths algorithm [1]. 3.9 Approximate Solution of the
Integer Programming Problem
3.8 The Integer Programming Model by Lagrange Relaxation
for Finding the Cheapest Cabling In order to reduce the integer linear program-
in a Given Trench Network ming problem in the previous chapter to a prob-
The problem of finding the cheapest cabling in a lem which can be handled by graph theoretical
given trench network was in ABONETT formu- algorithms, Lagrange relaxation combined with
lated as an integer linear programming problem. subgradient optimization was used. We relaxed
the constraints expressing capacity requirements
The following notation was used: while we retained the requirement that the net-
work should connect the root node with all the
xijt = 1 if there is a cable of type t carrying sig- subscribers and the cable/duct termination points.
nals from node i to node j, and 0 otherwise.
Before relaxation we made an approximation:
at is the capacity of cable type t. We replaced the constraints
nijt is the capacity used in cable type t from node

i to node j.
bi is the capacity of supply point i.

at xtij ntjk by (3.6) nijrt is the capacity used in cable type t from
it kt node i to node j following trench path r.

at xtij at xtjk , (3.7)
it kt bi is the capacity of supply point i.
and we disregarded the other constraints involv- pj is the capacity required in the cable leading
ing nijt. to subscriber or cable termination point j.
This had the effect of reducing the subproblem yg = 1 if trench section g is dug, and 0
to a pure cabling problem. The n-s, which con- otherwise.
stituted the corresponding usage solution, were
uniquely determined by the cabling. dg is the cost of digging trench section g.
However, when we updated the dual variables ijgr = 1 if trench section g is on the trench
corresponding to the modified constraints above path r from i to j, and 0 otherwise.
in the subgradient method, we did this based on
to what extent the following inequalities are sat- The variables yg were only defined for the trench
isfied: sections which were available at their true dig-
ging cost in the combination optimization.
ntij ntjk . (3.8)
it kt The integer linear programming problem was
In this way we were led to solve a sequence of formulated as follows:
cabling subproblems which were Steiner prob-
lems in directed acyclic graphs. We elected to minimize z = crt rt
ij xij + cg yg (3.9)
use Wongs method [2] augmented by Pacheco ijrt g
and Maculans solution improvement procedure subject to

[3] to solve these subproblems. For the sake of
at xrt
ij nrt
jk
(3.10)
completeness we describe in 3.11 and 3.12 the
irt krt
specialization of these methods to acyclic graphs.
if j is a cross connection or distribution point,
3.10 The Integer Programming
Model for Simultaneous Cable nrt
ij pj (3.11)
and Trench Optimization irt
In the combined optimization cables and if j is a subscriber or a cable termination point,

trenches were optimized simultaneously. The
problem was again formulated as an integer lin- nrt
jk bi (3.12)
ear program. This model is an extension of the krt
model described in the previous chapter. if i is a supply point,
Several trench paths between nodes were made nijrt at, (3.13)
eligible as alternatives in the optimization. First
of all the trench paths found in the cable and ijgr xijrt yg, (3.14)
trench optimizations were eligible. New paths
up to a prescribed maximum number were gen- xijrt, yg = 0 or 1 integer. (3.15)
erated by shortest paths whilst successively
blocking trench sections which were given a cost Ducts were again looked upon as a cable type
in the combined optimization. Here paths con- which served a special class of subscribers,
taining few of these trenches were given priority. namely duct termination points.
The following notation was used: This problem was again solved by Lagrange
relaxation where in addition the relationship
The eligible trench paths between node i and j between cables and trenches were relaxed, and
are referred to by the superscript r. where the same type of approximations were
made as when the trench network was given.
xijrt =1 if there is a cable of type t carrying sig-
nals from node i to node j following trench
path r, and 0 otherwise.
cijrt is the cost of a cable of type t following

trench path r from i to j.
3.11 Wongs Dual Ascent Method for Definition:
the Steiner Problem Specialized The Steiner length of a directed spanning tree in
to Rooted Directed Acyclic G(N, A) is equal to the sum of the length of the
Graphs arcs in the Steiner tree it contains.
Here we describe Wongs dual ascent method
specialized to rooted directed acyclic graphs in Definition:
the form it is implemented in ABONETT. The Steiner function f(i, j) defined for the arcs in
a directed spanning tree is equal to the number
Let G(N, A) be rooted directed acyclic graph of special nodes which can be reached via the
where arc (i, j) has length c(i, j) 0. arc (i, j) in the Steiner tree it contains.
Let r be the root node, S the set of special nodes Definition:

to be connected to the root node and v(s) an aux- A fragment in a directed spanning tree with
iliary function to be defined for the nodes s in S. Steiner function f is a maximal chain of arcs
with the same value of f. A node belongs by
Initialization: definition to the fragment which the incoming
Set v(s) = 0 for all s in S. arc belongs to.
Set d(i, j) = c(i, j) for all lines (i, j).
Let G be the subgraph with node set N and no The definition above secures that every node
lines. belongs to one fragment only.
1) For all nodes s in S, let C(s) be the set of Definition:

nodes from which there are directed paths to The segment of a node is the subchain of the
s in G. Choose an arbitrary s in S such that r fragment to which the node belongs that termi-
does not belong to C(s). If there is no such s, nates at the node.
go to 3). Otherwise, find nodes i and j such that
i does not belong to C(s), j belongs to C(s), In a directed spanning tree where (u, v) is an arc,
d(i, j) = min d(k, 1) for k not in C(s) and 1 in the predecessor u of v is denoted by p(v).
C(s).
The algorithmic iterates from one spanning tree
2) Set to another such that the Steiner length either is
reduced or stays constant while the prospect of
v(s) = v(s) + d(i, j) obtaining Steiner length reductions in later itera-
d(k, l) = d(k, l) d(i, j) for k not in C(s) tions is increased.
and l in C(s).
We operate with type 1 and type 2 iterations
Include the line (i, j) in G and go to 1). which we describe below:
3) Find the shortest directed spanning tree in G Type 1 iteration:

from r to the nodes which can be reached
from r in G and prune this tree. We consider a node i and an out-of-tree arc (i, j).
If we obtain reduction in the Steiner length by
It can be remarked that v(s)becomes a lower replacing (p(j), j) by (i, j), this is done, and the
bound on the minimum length of a Steiner tree. iteration is terminated.
3.12 Pacheco/Maculans Improve- If f(p(j), j) = 0, the Steiner length will remain

ment Algorithm for the Steiner unchanged by such a replacement. Nevertheless.
Problem Specialized to Rooted the replacement is carried out and the iteration
Directed Acyclic Graphs terminated if the length of js segment is reduced
Pacheco and Maculan have designed an algo- by it.
rithm which in many cases improves signifi-
cantly the solution of the Steiner problem which Assume now that f(p(j), j) > 0, f(p(i), i) = 0, and
results from Wongs algorithm. We describe that the change in the Steiner length which
here the specialization of this algorithm to root- would result by replacing (p(j), j) by (i, j) is
ed directed acyclic graphs in the form it has been 0. Then we look for nodes j such that (i, j)
implemented in ABONETT. are out-of-tree arcs and such that replacing
(p(j), j) by (i, j) would contribute to a reduc-
Let G(N, A) be a rooted directed acyclic graph tion of the Steiner length. If we can find a set
where arc (i, j) has length l(i, j) 0. Let r be the of such nodes whose total contribution to the
root node, and let S be the set of special nodes to reduction of the Steiner length is greater than ,
be connected to r in a Steiner tree. all the replacements are carried out, and the iter-
ation is terminated.
If the largest total reduction we can find is , 3.14 Related Work
we back up one node from i. If l(p(i), i), In the combination optimization, only a subset of
then the iteration is terminated without carrying the trench sections are considered at their real
out any replacement. Otherwise we look for out- cost. A formulation was implemented where all
of-tree nodes j such that replacing (p(j), j) by cable and trench options were considered simul-
(p(i), j) would contribute to a reduction of the taneously.
Steiner length. If we can find a set of such nodes
whose total contribution to the reduction of the 4 FABONETT Planning the
Steiner length is > , all the replacements Access Network
are carried out, and the iteration is terminated. When it was decided to place Service Access
Otherwise we back up one node from p(i) and Points (SAPs) and ring structures in the access
continue in the same way. If this process does network, Telenor R&D was requested to make a
not terminate in accordance with the criteria planning tool, which could assist in finding cost-
given, we terminate it without replacements effective access network designs.
when we reach the beginning of is segment.
A local switch (LS) and a set of main distribu-
Type 2 iteration: tion points (MDs) together with a set of what we
call special subscribers (SSs) were given. The
Here we consider a node n not in S with f(n, j) > MDs and the SSs could be connected directly
0 for at least two j-s. We will try to find an alter- to LS or via service access points (SAPs) where
native spanning tree with less Steiner length multiplexing was done. FABONETT operated
where f(p(n), n) = 0. This is done by searching with copper cables, fibre cables and, by abuse
for one suitable replacement arc for each frag- of language, radio cables. An SS should be con-
ment with f > 0 which succeeds n. nected to LS or to a SAP by either fibre or radio
cables. An MD should be connected to LS or to
First we calculate the length l of the fragment a SAP by copper cables. A sequence of cables
which ends in node n. For each fragment starting which connected an MD or an SS to a SAP or to
at node n we then search for the candidate out- LS, or which connected a SAP to another SAP or
of-tree replacement arc leading to the fragment to LS, was called a connection. The SAPs may
which would increase least (or reduce most) the belong to SDH rings, which must go through LS.
Steiner length. If the sum of these increases is In an SDH ring there were connections between
less than l, the replacements are carried out, and pairs of contiguous SAPs in the ring, and con-
we obtain a reduction of the Steiner length. nections between LS and SAPs adjacent to LS in
the ring. Each cable was placed in a sequence of
The improvement algorithm consists of carrying contiguous trace sections. A trace section was
out iterations of type 1 and 2 until no replace- characterized by its cost, length, type and one or
ments can be made. more section codes. The trace section type deter-
mined inter alia which cable types that could be
3.13 Post-Processing the Solution placed in the trace section. Typical trace section
Experience has shown that Lagrangian relax- types were conduits and ducts (existing or new),
ation and subgradient optimization not necessar- trenches of different categories, air cable sec-
ily yield acceptable primal solutions. Therefore tions and radio sections. A section code was
a simple post-processing of a selection of solu- simply a positive integer. Two trace sections
tions obtained in the final stages of the iteration shared a common section code if events causing
process was done, and the cheapest solution damage to the two trace sections were assumed
obtained in this way was selected. to be positively correlated. A connection inher-
ited the section codes from the trace sections
Typical elements in the postprocessing were: used by the cables forming the connection. Two
connections belonging to the same SDH ring
Increase the cable capacity to all nodes with could not share a section code. FABONETT
insufficient cable supply; operated with PDH SAPs and SDH SAPs. All
SDH SAPs were assumed to contain add/drop
Reduce the cable capacity to all nodes where multiplexers (ADMs). If an MD was directly
this is possible; connected to a SAP, the SAP had to contain
RSS/RSU. An SDH SAP could be a Transmis-
Move a subscriber to another distributor if this sion Point (TP). A TP did not contain RSS/RSU.
is profitable; Consequently, SSs and other SAPs, but no MDs,
could be directly connected to a TP.
Replace a cross connector or distributor with
one of another type if this is profitable; All SDH rings passed through LS and were
either STM1 or STM4 SNCP rings. The SDH
Try to eliminate nodes with low utilization. SAPs were ordered hierarchically: SDH SAPs
MD SS MD SS
SAPR
SAPR
PSAP SAP3 SAP2 SAPR LS SAP1 SAP2 SAP3 PSAP
SAPR
SAPR
which could only be connected directly to LS or stead. Furthermore, FABONETT/SDH did not Figure 4.1 Schematic view of
placed in SDH rings through LS were called S1 invent possible locations for candidate SAPs the access network structure
SAPs. SDH SAPs which could only be con- and candidate trace sections. All candidate SAPs
nected directly to LS or S1 SAPs were called S2 and trace sections must be provided by the user.
SAPs. SDH SAPs which could only be con-
nected directly to LS, S1 SAPs or S2 SAPs were Based on the location of LS, locations of MDs,
called S3 SAPs. S1 SAPs could belong to STM1 SSs, existing cables, existing and candidate trace
or STM4 SDH rings or not belong to rings at all. sections, existing and candidate SAPs, FABO-
S1 SAPs which belonged to rings were called NETT tried to find the least costly network
ring SAPs. By abuse of language LS was also design. The design problem was formulated as
denoted as an S0 SAP. If an S2 (S3) SAP S was an integer program which was solved by a com-
connected to an S1 (S2) SAP S, it was said that bination of linear programming with dynamic
S was subordinate to S. PDH SAPs could be row and column generation, branch and bound,
directly connected to LS or to SDH SAPs. There and heuristics.
were two categories of PDH SAPs, namely those
which required only a single connection, and Figure 4.1 gives a schematic view of the net-
those which required double connection (i.e. two work structure.
connections with no section code in common)
back to LS. The PDH SAPs would normally be Since FABONETT did not necessarily solve the
connected to LS in a way specified by the user. design problem to a theoretical optimum, the
If this was not done, a PDH SAP which required planner had to inspect the solution and some-
double connection to LS would be allowed to be times make model reruns with slightly altered
singly connected to a ring SAP. An MD could be input. The planner could for example question
directly connected through copper cables either FABONETTs selection of a particular SAP can-
to LS or to a SAP which was not a TP. A PDH didate and wish to make a rerun with this SAP
SAP could only have connected to it MDs which excluded. FABONETTs input format made this
either were connected to it in the existing net- possible without erasing the SAP candidate from
work or which the planner explicitly connected the input. Or the planner could question the cor-
to it. There could be subscribers connected to an rectness of connecting a particular SS directly to
MD who required the MD to be connected LS. A rerun could then be made where the plan-
directly either to LS, to a ring SAP or to a PDH ner specified which SAPs the SS should be
SAP with double connection to LS. A circuit allowed to connect to.
connecting an SS to LS belonged to one of two
types, namely regular circuits and singular cir- The planner had a problem of a dynamical
cuits. The singular circuits had to be directly nature. In establishing the best network structure
connected to LS through fibre. The regular cir- the development of the demand structure over
cuits could be connected directly to LS or to a time had be taken into consideration. FABO-
SAP through fibre or radio. Some SSs could NETT was a static one shot model. Some sim-
require that their regular circuits were connected ple features for dynamical use were, however,
directly either to LS, to a ring SAP or to a PDH built into FABONETT. The planner could give
SAP with double connection to LS. A PDH SAP selected SAP and trace section candidates the
could only have connected to it SSs which were label preferred and give them a bonus. Then
connected to it in the existing network or which two FABONETT runs were made. First a future
the planner explicitly connected to it. By con- run was made where the circuit demands repre-
vention it was said that an SS was connected to sent some future point in time. Then the main
a SAP (or LS) if the SSs regular circuits were run was made where some or all the candidate
connected to the SAP (or LS). SAPs and trace sections chosen in the future run
were labelled preferred and given a suitable
FABONETT did not propose new PDH SAPs. It bonus.
could, however, propose that PDH SAPs, which
in the existing network were directly connected A detailed description of the FABONETT model
to LS, should be connected to an SDH SAP in- and solution algorithms may be found in [4].
5 RUGSAM/RUGINETT/PETRA routed on a route object of type R. One or more
Transport Network Planning weight factors could be associated with arc (R,
Up to the early 1980s the routing and grouping R) in the graph indicating the fraction of the dif-
of circuits in what is now called the transport ferent capacities of type R taken up by a route
network was done manually. Telenor had experi- object of type R routed on R.
mented with the use of planning tools from other
telecommunication administrations without Connections that service circuit demands
much success, and it was decided to give Tele- directly were called demand connections. The
nor R&D the task of developing an optimization planner could specify that demand connections
tool. The development of the tool went through associated with a certain demand could or could
several phases. In the first phase routing and not be routed on by other connections. If a
grouping of circuits in a pure PDH network was demand could be routed on, it was denoted
considered. The tool RUGSAM consisted of two accessible. Otherwise it was denoted inaccessi-
optimization models where the first model opti- ble. Accessible demands were treated in the fol-
mized routing without considering grouping. lowing manner in the optimization module:
Then the model was changed into a single
model, RUGINETT, which optimized routing For every accessible demand, one or more
and grouping simultaneously. RUGINETT was connections (which might be routed on) with-
generalized to consider a combined PDH and out routing were established which the opti-
SDH network. In RUGINETT the cable and mization regarded as existing. Normally no
radio connections were considered as given, and existing connections were routed on these
the tool proposed cost-effective routing and connections, but the planner could specify
grouping of demands in the given network. any partial use of them.
Eventually it was found that this scope was too
narrow, and that the RUGINETT methodology Each accessible demand was changed into an
could be generalized to also propose new inaccessible demand where a possible require-
trenches, ducts, cables, and radio connections, ment for diversified routing was maintained.
and thus become a combined network design
and network utilization tool. Even if the basic After the optimization (which operated with
ideas were the same, the generalization was so inaccessible demands only), whatever was
fundamental that the tool was renamed and was routed on the accessible connections without
called PETRA. PETRA contained an optimiza- routing was shifted over to the connections
tion model based on integer programming which which covered the created demands.
proposed the installation of new components like
multiplexers, cables and radio connections and A connection belonged to exactly one connec-
ring structures, and at the same time how the tion type. Typical connection types were
network should be used to service forecasted
demands. PETRA did not pretend to solve the different variants of 2 Mb groups/circuits
integer program to optimality. However, the user x Mb PDH groups/circuits/transmission
interface allowed the planner to make modifica- systems
tions to the proposed solutions and check feasi- x Mb SDH virtual containers/multiplex
bility and costs. sections/groups/circuits
x Mb SDH ring sections
We shall briefly describe the basic concepts that x Mb WDM groups
were used in PETRA. The term route objects cables and radio links
were used as a common name for connections XDSL connections, point-to-multipoint con-
and equipment. The route objects were parti- nections, conduits and ducts
tioned into connection objects (short: connec-
tions) and equipment objects (short: equipment). A connection could be one-way, two-way or
Equipment objects were situated in nodes. Every undirected and was characterized by capacities,
route object belonged to a route object type. costs and how it was governed by the routing
rules. In particular, one-way connections could
Connections were generally routed on other con- be routed on one-way, two-way, and undirected
nections in the connections hierarchy and possi- connections, whilst two-way connections could
bly on nodes according to a set of inputted rout- be routed on two-way and undirected connec-
ing rules. In general, however, existing connections only, and undirected connections could
tions or connections specified by the planner be routed on undirected connections only.
could have partial routing only, or no routing
at all. The routing rules were visualized by an The number of possible connections is huge, and
acyclic routing graph. The nodes represented it would be impossible to introduce them all as
route objects. An arc from node R to node R decision variables in the optimization. They
indicated that a route object of type R could be were therefore generated dynamically through
a column generation technique. Another means 6.2 The GSM Access Network
of keeping the number of variables down was Design Problem
to operate with connection sets in lieu of connec- The mobile subscribers can connect to a set of
tions, where a connection set was a set of con- given Base Transceiver Stations (BTSs).
nections with identical routing.
Several BTSs can be connected together forming
The (mixed integer) optimization problem could a rooted tree. The BTS sitting at the root of such
in principle be solved to optimality by a branch- a tree is called an anchor BTS.
and-generate algorithm. However, the very size
of the problem made this prohibitive, and the Each anchor BTS must be connected to a Base
problem was instead solved by a combination Stations Controller (BSC), possibly via a Digital
of linear programming and heuristics. Access Cross Connect System (DACS). In
MOBANETT each BTS is characterized by a
Like the access network planner, the transport number of 64 kbit/s radio channels. The connec-
network planner had a problem of a dynamical tion must have sufficient capacity to be able to
nature. In establishing the best design, the devel- carry the total number of radio channels for all
opment of the demands over time had to be taken BTSs in the rooted tree associated with the
into consideration. Like FABONETT, PETRA anchor BTS.
was a static one shot model, and the same fea-
tures for dynamical use were built into PETRA. Each BSC must in turn be connected to a Mobile
Two PETRA runs could be made, first a future Services Switching Centre (MSC). The connec-
run where the demands represent some future tion must have sufficient capacity to carry the
point in time, and then a main run where some traffic from the BSC subject to a given blocking
or all the route objects chosen in the future run probability.
could be labelled preferred and given a bonus.
Each MSC must be connected to one or two
A detailed description of an early version of Main Switches (FS2s). The connection must
RUGINETT may be found in [5]. have sufficient capacity to carry the traffic from
the MSC subject to a given blocking probability
6 MOBANETT GSM Access and a given so-called redundancy factor. To
Network Planning simplify the presentation we shall assume that
Like for the other networks we have discussed, each MSC should be connected to two FS2s.
planners of mobile networks in Telenor had
found that manual planning of the GSM access An MSC and a BSC can be collocated in order
network was time-consuming, and that they to reduce cost.
would have capacity for analyzing a few alterna-
tives only. Therefore Telenor R&D was again Thus the GSM access network has a tree struc-
given the task of developing a suitable PC-based ture. This structure is shown in Figure 6.1.
planning tool. The result was the tool MOBA-
NETT which attempted to find a GSM access 6.3 Cost Minimization,
network which minimized total cost. Like the General Description
other tools, MOBANETT did not pretend to In Figure 6.1 the network to be designed lies
solve the cost minimization problem to optimal- between the two horizontal lines.
ity. However, the accompanying user interface
allowed the planner to make modifications to the The locations of the FS2s and the BTSs and the
solution found by MOBANETT and check feasi- number of radio channels between the anchor
bility and cost. BTSs and the BSC/DACS were input to MOBA-
NETT. MOBANETT tried to determine
6.1 General
MOBANETT consisted of several modules where MSCs, BSCs and DACSs should be
which could be put into one of two categories: placed;
optimization modules which found solutions which BTSs should be connected to which
to the cost minimization problem by mathe- DACS/BSCs;
matical and graph teoretical methods;
which BSCs should be connected to which
modules which enabled the user to interface MSCs;
with MOBANETT via (input and output)
tables and which prepared the data formats in order to obtain a feasible access network at
suitable for the optimization modules. the lowest possible cost.
Figure 6.1 GSM access
network structure FS2
MSC
BSC
DACS
Anchor BTS
BTS
The cost minimization was done with algorithms one BSC only. So if one wanted to model sev-
which relate to a particular graph called an eral optional connections for a candidate DACS,
options graph. An example of an options graph the DACS had to be duplicated (an example of
is shown in Figure 6.2. Here the nodes between which is shown to the right in Figure 6.2).
the two horizontal lines represent candidate
DACSs, BSCs and MSCs, and the edges repre- 6.4 Mathematical Model
sent candidate connections. Each of the candi-
date connections can carry a number of 64 kbit/s 6.4.1 Notation
circuits. Between BTSs and BSCs each circuit The following notation will be used:
can carry one radio channel, while between
BSCs and MSCs, and between MSCs and FS2s, Subscripts:
each circuit may carry F (usually 1 or 4) chan- t: BTS
nels where F is set by the user. Each of the edges d: DACS
in the options graph had a cost function associ- m: MSC
ated with it which gives the connection cost as a b(d): the BSC to which DACS d is connected
function of the number of circuits connected.
The algorithms tried to find the subtree of the Constants:
options graph which at the lowest possible cost kt: number of radio channels to be con-
connects the BTSs via DACS/BSCs and MSCs nected from BTS t
to two FS2s. The two FS2s which a candidate et: the traffic measured in erlang at BTS t
MSC should be connected to (if it was selected) f: 1 blocking probability in BSC and
were input to the optimization although MOBA- MSC
NETT would propose which two FS2s to use. In F: number of channels per BSC MSC
the options graph a DACS may be connected to circuit, and per MSC FS2 circuit
Figure 6.2 Options graph
FS2
MSC
BSC
DACS
Anchor BTS
BTS
Rm: redundancy factor between MSC m and cmk: cost of connecting k channels between
FS2 MSC m and FS2
Decision variables: The cost coefficients will reflect connection

xtd = 1 if BTS t is connected to DACS d, and costs as well as costs of DACS, BSC and MSC.
0 otherwise We shall return to the cost structure later.
xtb = 1 if BTS t is connected to BSC b, and 0
otherwise Auxiliary variables:
xdk = 1 if there are k channels between eb: offered traffic at BSC b
DACS d and BSC b(d), and 0 other- em: offered traffic at MSC m
wise
xbmk = 1 if there are k channels from BSC b to Linearized erlang functions:
MSC m, and 0 otherwise T: traffic in erlang
xmk = 1 if there are k channels from MSC m k: number of channels
to FS2, and 0 otherwise E(T): number of channels needed as a func-
tion of T (for given f)
Cost coefficients associated with the variables: E0 and E: constants in the linearization of E(T)
ctd: cost of edge between BTS t and DACS d E(T) = E0 + ET for T > 0, E(T) = 0 for T = 0
ctb: cost of edge between BTS t and BSC b with inverse
cdk: cost of connecting k channels between E-1(k) = E0 / E + k / E for k > E0, E-1(k) = 0
DACS d and BSC b(d) for T = 0
cbmk: cost of connecting k channels from
BSC b to MSC m
6.4.2 Optimization Model a constant part = C,
The cost minimization problem was formulated a linear part Lc, and
as follows: a saw tooth part T(c) = T . (1 c(mod 30) / c)
minimize ctdxtd + ctbxtb + cdkxdk + Thus C(c) = C + Lc + T(c).

cbmkxbmk + cmkxmk
The saw tooth part reflected cost elements which
subject to depended on the number of 2 Mbit/s circuits.
tktxtd kkxdk (balance in DACS d) (6.1) We shall now describe the individual cost func-
tions and how they were allocated to the edges
E(eb) Fmkkxbmk (balance in BSC b) (6.2) in the options graph.
RmE(em) kkxmk (balance in MSC m) (6.3) 6.5.1 DACS Cost

The cost of a DACS consisted of a cost per
where 2 Mbit/s circuit connected to a BTS or a BSC.
The cost decomposed into a linear part and a
eb = t,dbetxtd + tetxtb saw tooth part. The linear part translated into a
constant part which was put on each of the edges
and connecting to BTSs. The saw tooth part was put
on the edge connecting to the BSC.
em = bfE-1 (Fkkxbmk)
6.5.2 BSC Cost
Using the formulas for E(T) and E-1(k) trans- The cost of a BSC consists of a fixed cost plus
forms (2) and (3) to (4) and (5): a cost per radio channel coming from the BTSs
which are connected to it (possibly via DACS).
E t,dbetxtd + Etetxtb mk(Fk E0)xbmk (6.4) The cost decomposed into a constant part and
a linear part. The constant part was put on the
Rmfbk(Fk E0)xbmk k(k RmE0)xmk (6.5) edges connecting to MSCs. The linear part trans-
lated into constant parts on the edges between
Note that we must have k E0 / F in BSCs and BTSs and the BSC and on the edges between
k RmE0 in MSCs. This defines the minimal BTSs and the DACSs associated with the BSC.
values kmin which k can take.
6.5.3 MSC Cost
Thus the optimization model becomes: The cost of an MSC consisted of a fixed cost
plus a cost per 2 Mbit/s circuit connecting to a
minimize ctdxtd + ctbxtb + cdkxdk + BSC or FS2. The cost decomposed into a con-
cbmkxbmk + cmkxmk stant part on the edges to the FS2s, a linear part
on the edges to the BSCs, a saw tooth part on
subject to each of the edges to BSCs and a saw tooth part
for each of the edges to FS2s.
tktxtd kkxdk (balance in DACS d) (6.6)
6.5.4 Cost of Connection Between BTS
E t,dbetxtd + Etetxtb mk(Fk E0)xbmk and DACS/BSC
(balance in BSC b) (6.7) The cost of the connection between a BTS and
a DACS/BSC consisted of a cost per 64 kbit/s or
Rmfbk(Fk E0)xbmk k(k RmE0)xmk per 2 Mbit/s. In both cases this translated into a
(balance in MSC m) (6.8) constant cost on the edge between the BTS and
DACS.
where
6.5.5 Cost of Connection Between DACS
k E0 / F in BSCs and k RmE0 in MSCs. and BSC
The cost of the connection between a DACS
6.5 Cost Structure and a BSC consisted of a cost per 2 Mbit/s. This
A detailed presentation of the cost elements decomposed into a linear part which translates
which the planner gives via tables to MOBA- into constant parts on the BTS DACS edges
NETT is not given here. Prior to the optimiza- and a saw tooth part on the DACS BSC edge.
tion the costs of components (DACS, BSC,
MSC) and connections as a function of the num- 6.5.6 Cost of Connection Between BSC
ber of radio channels c were approximated by and MSC
functions C(c) which in general were sums of The cost of the connection between a BSC and
up to three terms. These terms were an MSC consisted of a cost per 2 Mbit/s. This
decomposed into a linear part which translated Linear costs on the edges between MSCs and
into linear parts and saw tooth parts on the BSC BSCs cannot be moved towards the DACS/BTSs
MSC edges. in the same way because a BSC can be con-
nected to several alternative MSCs. The best we
6.5.7 Cost of Connection Between MSC can do is for a BSC to move a part of each linear
and FS2 MSC BSC cost equal to the minimal linear
The cost of the connection between an MSC and MSC BSC cost for this BSC. This was done
an FS2 consisted of a cost per 2 Mbit/s. This after all other costs had been allocated and pos-
decomposed into a linear part which translated sibly shifted. Let Lmin be the minimal coefficient
into a linear part and a saw tooth part on the for the linear costs on the edges from MSCs and
MSC FS2 edge. let L be the corresponding coefficient for the edge
to an arbitrary MSC. The linear cost coefficient
6.5.8 Collocation of BSC and MSC on the BSC MSC edge was changed to L Lmin.
It could be cost effective to place a BSC and an
MSC in the same node. This was treated simply Considerations analogous to those above give
by introducing into the options graph additional the number k of channels between BSC and
nodes for a candidate MSC and a candidate BSC MSC equal to
with reduced costs and a zero cost connection
between them. k = E0 / F + Et,dbet / F + Etet / F.
6.5.9 Shifting Linear Edge Costs The cost shifting thus resulted in
Towards BTS
It was believed that the optimization algorithms a constant term LminEet / F on each BTS
function better if the cost elements were shifted DACS edge
to edges in the options graph which are as close
to the BTSs as possible. We shall now describe a constant term LminEet / F on each BTS
how this shifting is done in principle. BSC edge
We have already seen how linear costs associ- a constant term LminE0 / F on each BTS
ated with connections between DACS and BSC DACS edge
were changed to constant costs on DACS BTS
edges. We similarly translated the linear costs on a linear term (L Lmin)k on the BSC MSC
the FS2 MSC edges to corresponding MSC edge
BSC edges. In order to do this it was necessary
to find the number k of channels between FS2s 6.6 Candidate BSC and MSC Nodes
and an MSC as a function of the number k of and Connections
channels between the MSC and MSCs. Inequal- In order to relieve the planner of the tedious task
ity (6.5), which could just as well have been of inputting all locations for candidate BSC
written as an equality, gives locations a simple algorithm was developed
which proposed candidate BSC nodes.
Rmfb(Fk E0) = k RmE0 (6.9)
The geographical locations of the BTSs and
where the sum is over the BSCs connected to the FS2s were given as input to MOBANETT. In
MSC. addition it was possible to give as input the loca-
tions of NMT base stations and NMT switches
This gives (MTXs). All these geographical points had the
possibility of becoming candidate BSC nodes.
k = RmE0 + b(RmfFk RmfE0) (6.10)
The MTXs were automatically made into candi-
A linear term Lc on the edge MSC FS2 thus date BSC nodes, and the planner would add on
translates into more candidate BSC nodes of his choice. There-
after he could apply the algorithm below.
a constant term LRmfE0 on each BSC MSC
edge Algorithm which proposes additional
candidate BSC nodes:
a linear term LRmfFk on each BSC MSC
edge 1 Draw the largest circle possible around every
candidate node (with the node as centre) so that
a constant term LRmE0 on the FS2 MSC the total traffic generated by the BTSs in the
edge circle is below a given limit (the exclusion
limit), and exclude all points in the circles from
the possibility of becoming candidate nodes.
2 If every geographical point is either a candi- had earlier experimented with algorithms for
date node or excluded, stop. Otherwise draw finding the least number of frequencies neces-
the largest possible circle around every geo- sary to carry a given traffic in a network with a
graphical point (with the point as centre) given set of BTSs with sufficiently low level of
which is not yet a candidate node and which interference. Telenor R&D was then given the
is not yet excluded, so that the total traffic task of instead making an optimization tool
generated by the BTSs in the circle is below which assigned a given number of frequencies to
a given limit (the inclusion limit). Define the each BTS such that the level of interference was
point with the smallest circle as a candidate acceptable and minimized. The input was a set
BSC node. Go to 1. of admissible frequencies (which needed not
be contiguous) and a symmetric compatibility
After the candidate BSC nodes had been decided matrix which for each pair of BTSs indicated
upon the planner selected a subset of these, pos- whether they interfered on neighboring frequen-
sibly augmented by some other geographical cies, on same frequencies only, or not at all.
points, as candidate MSC nodes. MOBINETT went through several stages.
The planner could also get some assistance in The first variant considered only BTSs which
setting up candidate connections. MOBANETT supported neither baseband nor synthesizer hop-
would for every MSC candidate propose connec- ping. This problem was formulated as a pure
tions to the two closest FS2s and to the closest integer programming problem which was ini-
candidate BSC nodes up to a given number and tially solved by a combination of linear pro-
within a certain distance. Furthermore MOBA- gramming and heuristics. For realistic cases the
NETT would for each candidate BSC propose number of variables was reasonably small (about
connections to the closest BTSs up to a given 10,000) whilst the number of constraints could
number and within a given distance. Connec- be rather large (> 150,000). Since the linear pro-
tions via DACS were also proposed according to gramming software we used performed better
certain criteria which we shall not go into here. with a high number of variables and a low num-
ber of constraints than vice versa, the transposed
6.7 Solving the Optimization Model problem (where the dual variables had to be inte-
gers) was solved instead. In order to solve the
6.7.1 Use of Lagrange Relaxation problem to optimality the heuristic was replaced
The optimization model as defined in 3.2 was by a dual branch and cut and generate software
solved by using Lagrange relaxation and subgra- package that was designed and implemented at
dient optimization. The technicalities connected Telenor R&D.
with the use of the subgradient method are stan-
dard and will not be described here. Inequalities As new BTSs which supported baseband and
(1), (4) and (5) were relaxed. However, the con- synthesizer hopping became available, MOBI-
dition that the BTSs must be connected to FS2s NETT had to be modified accordingly. The com-
in a tree structure was retained as a constraint. patibility matrix was replaced by two asymmet-
The relaxed problem thus became a classical ric interference matrices, one describing interfer-
Steiner problem in a rooted directed acyclic ences from interfering BTSs to victim BTSs on
graph. The root node was an auxiliary node con- the same frequency, and one describing corre-
nected to the candidate MSC nodes where the sponding interferences on neighboring frequen-
cost of the connection reflected the cost of con- cies. Thresholds were set for acceptable interfer-
necting the MSC to its two FS2s, and where the ence. This necessitated a complete remodeling
special nodes to be connected to the root repre- of the problem, and new solution algorithms had
sented the BTSs. to be implemented. The new MOBINETT tried
to allocate frequencies to cells such that the fre-
6.7.2 Solving the Steiner Subproblem quency requirements were satisfied and such that
Since the Steiner subproblem had to be solved a a weighted sum of interference contributions
substantial number of times a heuristic is used. above given thresholds were minimized.
The heuristic chosen was Wongs dual ascent
algorithm [2] followed by Pacheco-Maculans The planner could partition the cells into subsets
solution improvement algorithm [3], both spe- and solve the allocation problem for one subset
cialized to rooted directed acyclic graphs. at a time. The basic approach was to formulate
the subproblems as integer programs which was
7 GSM Frequency Planning solved by classical branch and bound. In addi-
by MOBINETT tion, postprocessors based on tabu search and
The requirements for GSM services increased simulated annealing were implemented. MOBI-
rapidly in the early 1990s, and it was realized NETT is at the time of writing still in use.
that it was crucial to have access to a good PC-
based tool for frequency planning. Telenor R&D
8 Conclusion 9 References
We see that Telenor R&D over the years has 1 Rayward-Smith, V J, Clare, A. On Finding
been involved in the establishment of network Steiner Vertices. Networks, 16, 283294,
planning tools for most parts of the physical net- 1986.
work. This had not been possible without the
participation and support from dedicated net- 2 Wong, R T. A Dual Ascent Approach for
work planners in Telenor. The main hurdles Steiner Tree Problems on a Directed Graph.
have been: Mathematical Programming, 28, 271287,
1984.
The varying quality of the data in the network
databases. A spin-off effect of the planning 3 Pacheco, O I P, Maculan, N. Metodo Heuris-
tool development has been a substantial in- tic para o Problema de Steiner num Grafo
crease in the accuracy of some of the data Direcionado. Proceedings of the III CLAIO,
sources. Santiago, Chile, August 1986.
The responsibility for the planning of the dif- 4 Lorentzen, R. Mathematical Model and
ferent networks was in the past decentralized. Algorithms for FABONETT/SDH. Telek-
The local planners had a variety of responsi- tronikk, 94 (1), 135145, 1998.
bilities, and it was often difficult for them to
allocate the time necessary to acquire and 5 Lorentzen, R. Mathematical Methods and
maintain the necessary familiarity with the Algorithms in the Network Utilization Plan-
tools. Also, the ICT equipment available to ning Tool RUGINETT. Telektronikk, 90 (4),
the planners at the local level was not always 7382, 1994.
sufficient for making effective use of the tools.
There are reasons to believe that these hurdles

will be easier to overcome in the future:
The requirement for high accuracy in the net-

work databases is becoming more and more
pronounced, not only because of requirements
from planning tools.
There is a tendency to centralize the network

planning activity. This implies that planners
can dedicate more of their time to get the most
out of sophisticated planning tools.
The reliability, capacity and speed of ICT

equipment has become more than sufficient to
support the optimization algorithms that form
the motor in modern network planning tools.
Finally it should be observed that the develop-

ment, implementation, and use of network plan-
ning tools ought to be an ongoing process. Both
the demands for new services and the technol-
ogy that can be used to serve them evolve at
increasing speed.
Network Strategy Studies
TERJE JENSEN
Every network operator needs to have a network strategy covering the complete network portfolio. The
strategy must also be operational, meaning to be related to decisions and actions in near-time. One
main goal for having a strategy is to be prepared for chances that can be revealed with time. That is,
the strategy is likely to assist in detecting business opportunities.
A number of methods can be applied when elaborating the strategy, including scenarios, cost/benefit
calculations and risk assessment these aspects are briefly presented in this article.
1 Introduction different options, as elaborated in Section 4. As

Dr. Terje Jensen (41) is In todays world, change and uncertainty seem presented in Section 5, a sound strategy must
to be constants. On the one hand this is nice, as also relate to trends at various levels. Section 6
recent years he has mostly been improvements are captured in the evolution pro- presents how scenario work can be applied in
engaged in network strategy cess. On the other hand, the ongoing dynamics order to assist when deriving strategies, followed
is hard to capture in network planning with a by examples in Section 7. A quantitative
Besides these activities he has longer time horizon. In fact, a claim could be approach is then given in Section 8. Assessing
been involved in internal and that there is no point at all in looking a few years risk is also an essential element in any strategy
ahead as for (almost) certain, the future will not as described in Section 9. Some overall discus-
ling/analyses and dimensioning. be exactly as predicted. It is fair to say that this sions are then given in the last two sections.
terje.jensen1@telenor.com claim is missing the main objectives of carrying
out strategic planning, for networks as well as 2 Relating Long and Short
for other areas. Besides being incorrect, the Term the Scope and the
claim disregards the benefits following from a Challenges
systematic analysis of a companys surroundings It is fundamental to understand that the strategic
and future options. evaluations and planning are not carried out for
their own purposes per se, but need to be related
A motivation for evaluating the coming avail- to the current situation in order to get practical
able choices is to use the results to elaborate a implications. Naturally a number of steps in the
robust action plan to cover triggers for making strategy can be located some time ahead, and
decisions or to initiate actions as well as optional hence do not necessarily influence todays activ-
actions. As described later a robust plan supports ities. This means that one result is to sort out the
a flexible roadmap of system/network evolution, decisions to be made in a timely manner. How-
that is what to do for each of the systems given ever, one main result is to devise the actions
certain factors. Hence, a consequence of this is needed in the actual situation an actor finds itself
to try to postpone final decisions until the in. The overall process is depicted in Figure 1.
actions have to be started. This gives the option
to not start the action or carry out another action. A number of general steps can be identified:
In an economic sense, there are at least two i. The frame of the task is described by the
motivations why flexibility should be included elements: a) Current network portfolio, b)
in an investment evaluation: Firstly, the estimate Forecasts and trends, c) Set of (optional) tar-
of the activity value is improved, better reflect- get states. Naturally, these are interrelated as
ing the actual characteristics of the challenge. a potential target state is influenced by ones
Another motivation is the improved insight position in the existing situation and a sce-
gained for uncertainties and hence a better nario for further development. That is, a sce-
understanding of the flexibility and the options nario-based approach will likely apply for
expected. Commonly real options are applied to this step. Included in the current portfolio
include these options in the evaluation. Bringing goes also an assessment of short-term devel-
real options into the equation increases the level opment, e.g. given through a historic
of activity values, although the increase differs description in the traffic/user development
for the different nature of activities. in a system.
The following sections give an overall descrip- ii. The gap between the current situation and
tion of strategy work, both the intention (Section the target states is explained with a set of
2) and the overall approach (Section 3). A num- possible paths. That is, a path will tell a story
ber of criteria are needed to choose between the for how the current network portfolio will
Now Target time Now Target time
Target networks options Target networks options
Status of Status of
network network
portfolio portfolio
Decision point
- question
- consequences per option (technical,
economics, products etc)
Scenarios to derive options Scenarios to derive options
Forecasts market, trends etc Forecasts market, trends etc
I. II.
Now Target time Now Target time
Target networks options Target networks options
Status of Status of
network network
portfolio portfolio
Basis for deciding

upon the first steps
Optional (recommended) Optional (recommended)
migration tracks migration tracks
Risks: Technical, business,

strategy, regulatory
Scenarios to derive options Scenarios to derive options
Forecasts market, trends etc Forecasts market, trends etc
III. IV.
migrate in order to reach a target state. Dur- present value, internal rate of return, finan- Figure 1 Illustration of
ing a migration, a number of decisions has to cial needs, etc. step-wise approach using
be made and solutions implemented. A deci- the target options for deriving
sion is indicated by a diamond in Figure 1. iv. Looking at the candidate tracks, a number of recommended next steps for
For each decision, in principle, other choices risk factors can be attached to each track todays networks
could have been made, possibly resulting in describing technical, business-related, strate-
branches to the path. A decision point should gic, regulatory and other risk phenomena.
be estimated in time (i.e. when should the As shown later, these risk factors can be
decision be made), in addition the relevant described qualitatively or included in a quan-
question has to be described together with titative way. So, carrying out this exercise, it
consequences of each of the relevant out- is essential to relate the observations made
comes of the choices made (consequences in to how to make the first step for the current
terms of technical, economic, product, etc. network portfolio. That is, the results are
issues). used when making the choice of how to pro-
ceed with todays networks. It is also impor-
iii. Considering the set of optional (migration) tant that the information is revisited and
paths and set of decision points, a decision revised as necessary on a regular basis;
map can be drawn. This then contains the set including the target options, the decision
of candidate paths than can be followed map and the risk evaluations.
towards the targets. Commonly, quite a few
of the theoretical paths are not likely for Working with a horizon for a given number of
practical reasons, including financial mea- years, it is natural to start by gaining insight into
sures. Hence, a number of real candidates trends and key drivers for the network evolution.
can be described, of which fewer can possi- However, quite a few optional network solutions
bly be recommended migration tracks. These are expected. Being able to select an appropriate
candidates may further be compared accord- migration of an operators network portfolio, a
ing to agreed upon measures, such as net set of evaluation criteria has to be defined. To
As is - Obtain lower overall investment levels, bal-
- Network elements/capacities anced between different areas with corre-
- Traffic demands, products, customers
- Traffic variations, plans sponding timing (access/transport/service
platform, support systems),
Trends,
functionality Prognoses - Ensure complete solutions; consider run-
roadmaps Candidate evaluations
ning costs as well as investments.
Qualitative Quantitative
Other evaluations evaluations
Scenarios Ensure that an operator maintains the leading
sources
position for introducing innovative/holistic
Overall network roadmap network solutions adapted by the broadband
market (and others), by enabling rapid service
provision.
The priority placed on the different initiatives

Figure 2 Input and results for some extent, these criteria can be quantified has to be future-proof, meaning that the steps
evaluation (qualitative and although some qualitative assessments will also taken have to fit into a longer-term journey
quantitative) be necessary. approaching a target state of an operators opera-
tion. This basically captures the motivation as
An overall objective of an actor is to level the well as the overall challenge currently faced by
profit level, both short- and long term. Then, several operators.
facing the brutal reality of long-term economics,
the basic question of which networks and sys- 3 General Approach
tems to base the operators future operation on is An overall schematic illustration of the strategic
catching up. As commonly seen elsewhere, man- evaluation task is given in Figure 2. This is to
aging to terminate a (still) profitable system describe evaluations needed to make the deci-
seems like a harder decision to make than to sions shown in Figure 1. The following main
start up a new system. Hence, the conglomerate groups of work items are:
of systems that emerge may grow. In addition,
several of the systems seem to gradually a) Description of status and current plans
increase overlap of their application area. This (as-is) for the networks/systems looked at.
means that in the future more candidates will Configurations at the selected locations are
enter the arena, capable of supporting a spectre described in terms of network elements
of products including those already carried (manufacturer, capacity, links, etc.), traffic
today. One consequence is that the answer to flows and traffic variation.
which system to base the future on is steadily
more difficult to find. b) Elaboration of optional futures/scenarios
in terms of i) service demands/prognoses,
On the other hand, a few trends seem to domi- ii) trends for network solutions (and corre-
nate the industry, such as IP, optics, DSL and sponding functionality), and, iii) major
mobile/wireless. But, in view of the current net- uncertain issues (captured by the scenarios).
work portfolio managed by an operator and the
need to constrain investments and operations to c) Other sources, e.g. from analysis companies,
profitable ones, there is still a question in where standards, vendors, and so forth.
to invest in order to arrive at a future-proof port-
folio. Moreover, the overall risk should be d) Elaboration of coherent plans for network
acceptable, considering the profit level sought. roadmaps considering the dependencies
between different systems/networks utilising
Bearing in mind the overall objective, the goals their strengths.
of the strategic evaluations are to:
In order to elaborate plans for network roadmaps
Describe an operators target network portfo- the candidates have to be evaluated. For this pro-
lio, ject, both qualitative and quantitative evaluations
have to be conducted. Placing this evaluation in
Identify smart timing of new investments and the centre, several inputs have been identified, as
migration of existing platforms, shown in Figure 2. The overall results, implying
the recommended roadmaps, are then supported
Establish a foundation for decisions on invest- by economics as well as functionality argu-
ment and operational aspects to keep a low ments, and other arguments related to business
cost level: and regulatory issues.
4 Criteria for Selecting As for risks, the technology risk addresses one
Migration Branches area. However, a number of additional areas
could also be treated:
4.1 Overall Financial
In detailing future network candidates, a number Political and market economic risk
of options are revealed in terms of choosing
which candidates to base a further network Market and commercial risk (including regu-
migration on. Hence, there is a need to define latory)
a set of criteria to use when selecting which
options are the better ones. Some criteria are Partner risk (also including vendors)
described in the following.
Financial risk
One basic criterion is the profit levels expected
by taking certain steps. Requirements on the ex- Organisational risk (to follow the activities
pected profit are given considering the accompa- and realise profits)
nying risk level. Hence, higher risk would likely
ask for a higher profit level than a lower risk In order to realise the profit, appropriate steps
action. Profit can be estimated in different ways. have to be prepared for in a timely manner, like
For example, assuming that network solution is organisational efficiency, increased revenue,
irrelevant to income level, the cost of the solu- reduced cost, etc. Moreover, exit strategies have
tion is to be minimised. In general, however, to be elaborated to cover alternative steps/paths
various solutions may support different levels of to follow in case some of the planned steps turn
service and product type. The income side must out to be unwanted later on in the process.
therefore also be considered in the equation. Which exit strategies that are possible should
also be discussed. This corresponds with the
A fundamental challenge on the income side is map of decision points shown in Figure 1.
that the price level is influenced by many factors
outside the operator considered, such as com- The interplay with vendors and customers must
petitors and regulator. Theoretical models and also be obeyed. That is, trends among the users
analyses exist for similar configurations and will have to be observed and possibly matched in
not be dealt with in this article. order to increase the service demands. Likewise,
choices and prioritisations made by the vendors
Concerning financial aspects a number of topics have to be followed, as it could turn out very
should be looked at, such as: costly to install and maintain a system from a
vendor that is leaving the market or decides not
Financing needs, e.g. peak funding, to support that system in the future.
Value and profitability evaluations, e.g. net Keeping in mind the broader set of criteria, a
present value, internal return rate, payback few criteria groups are treated in the following
time, (Figure 3):
Sensitivity analyses of major factors. Product-related: capabilities of supporting rel-

evant products,
Profit
Cost-related Product-related
Flexibility (exit options)
Coordination-related
Risks,
Regulatory-related
What are others (vendors, competitors etc) doing?
Figure 3 Selected
criteria classes
Real-time Timing requirements: Covers real-time re-
requirements
quirements, potentially with further variants
on timing values (delay and delay variation).
None
Messaging Volume/bit-rate: The volume is an indicator of

the amount of data to be transferred. Consider-
File transfer ing constraints on the timing, this can also be
Transaction expressed as requirements on the range of bit-
Less strict
Leased capacity rate needed.

Web-http
Interface type/technical solution: Covers types
of protocol, etc., and will likely be mostly
Audio streaming Video streaming
related to products between operator units.
Multimedia conference Some of these may be used to separate products

Strict
Voice Live video within categories. The timing requirements

Live bcast
Leased line would likely be a significant factor in addition to
Volume/bitrate the volume/bit-rate. Some concerns may also be
related to the content/medium carried and how
the two end-points relate to each other (e.g. top-
Figure 4 Organising Cost-related: the investment and operational ology/scope covering relations between the end-
individual end-user services costs associated with the network solution points, like one-to-one, one-to-N and M-to-N).
(and migration),
Some examples using the first two factors are
Regulatory-related: would there be any regu- looked at in the following (Figure 4):
latory restrictions or exposure involved,
Voice: Carrying speech conversations, which
Co-ordination-related: are there any effects implies interactivity between a set of partici-
from the network solutions onto business mat- pants and strict real-time requirements on car-
ters (internal trading, etc.). rying the voice samples. Currently, most of
the voice traffic is carried in PSTN/ISDN and
The actual criteria weights to use might also GSM. Typical bit-rates are in the range 8
depend on the market situation/position, like 64 kbit/s.
whether an established operators situation or
a green field region is considered. Hence, a Live video (VoD): Representing transfer of
somewhat wider scope may be taken on where video in real time between a source and a
there is no history/legacy systems. To some receiver. This implies fairly strict real-time
extent this would also be valid when the overall requirements. Typical bit-rates vary in the area
service portfolio and operation of an actor are of 500 kbit/s 6 Mbit/s.
looked at deciding upon which products to
offer and which market segments to address. Multimedia conference: Representing a set of
When a set of legacy systems is involved, more media types (voice, video, text, image, etc.)
emphasis might be placed on managing the sys- that are to be transferred between a set of par-
tems efficiently and relating the products and ticipants in the conference. Typically there
systems (production means). This also includes are interactivity requirements, although some
decisions regarding the operations support sys- media types will have less strict real-time
tems. requirements. An example of aggregated bit-
rates from a single participant is in the range
4.2 Product-related 64 kbit/s 2 Mbit/s. Some services are offered
All product types relevant for the evaluation in a dedicated videoconference network today.
have to be considered. This includes not only In addition, a bundle of ISDN channels may
products seen by end-users, but also products to also work as transport. A NetMeeting-like
other operators or departments within an opera- application would also address this product
tor when relevant. Moreover, a product may also category.
be bundled together with other products to com-
pose a new product. Video-streaming: Representing the transfer
of video content with a play-out buffer in the
In the following are treated more product charac- receiver. Hence, this buffer captures some
teristics relating to transport. Basically, such a slack in the transfer-rate variation. The aver-
categorisation could be done in a number of age bit-rate may be the same as video, 500
ways. However, to simplify the task, the follow- kbit/s 6 Mbit/s.
ing factors could be used:
Audio-streaming: Representing the transfer of In several of these products, a number of traffic
audio with a play-out buffer in the receiver. flows could be involved, each with its separate
Typical applications would be radio and characteristics. In addition, other aspects could
music. Today radio stations can be listened also be considered, including:
to through the Internet. Typical bit-rates 16
128 kbit/s. To what extent mobility (and portability) is
supported. The highest bit-rates would not be
Web http: Browsing implies some level of easily provided on wireless/mobile links
interactivity with a user selecting objects to be (except for broadcast networks).
transferred. This means some requirements on
time for transferring the objects, although no The topology between the involved communi-
real-time requirements as such. Typical effec- cation parties. For conference, multicast and
tive (average) bit-rates in the range 15 100 broadcast services, there would be several par-
kbit/s (although peak bit-rates are higher, say ties involved. Hence, allowing for multi-party
64 500 kbit/s). destinations may require corresponding func-
tions in the network. Collection networks may
File transfer: This product category covers also be considered, where a single receiver
regular file transfers and more business criti- gathers information from several sources.
cal applications, e.g. related to outsourcing.
For the latter a high-bandwidth connectivity of The dynamics in establishment/release of
100 Mbit/s or more could be demanded. File communication sessions. Two variants are on-
transfer represents transfer of (larger) files demand (controlled by user) and permanent
without real time requirements. Examples of (fully controlled by the operator/provider).
file sizes are 0.5 Mbyte (document), 4 Mbyte Traditionally signalling has been used for
(mp3 file), 1 Gbyte (video movie). the former, while management activities are
invoked for the latter. However, a combina-
Messaging: Exchange of information between tion of signalling and management procedures
users without real-time requirements. A store- may the applied.
and-forward principle is introduced, allowing
for storage of messages in intermediate servers. Degree of dependability. Two main depend-
ability measures are availability and reliabil-
Transactions: Transactions represent messages ity. Requirements on dependability may differ
containing text, images, etc. Examples of vol- for the different products and also vary for the
umes are found in the range a few hundred different customer types. An example of a
bytes to a few Mbyte. No strict real-time product with fairly high availability is an
requirements are attached, although an ac- alarm service (considered to belong to the
knowledgement is commonly conveyed to transaction product category).
the source.
For products between operators/providers,
Live TV broadcasting: Representing todays these may be aggregation of the end-user
broadcasting and future digital systems. Ex- products (like for a wholesale operation). The Figure 5 Matching
ample of capacity is 6 Mbyte per channel. interface type may also be specified. A num- applications to be
Strict real-time requirements (direction ber of interface types could therefore be speci- supported and access
towards the receiver). fied. solutions (examples)
Leased line: Representing leased line services

provided, e.g. by the SDH network. Todays
downstream upstream
bit-rates include 64 kbit/s 155 Mbit/s and up
to wavelengths. Strict real-time requirements 15 10 5 2 0 2 5 10 15 Mbit/s
are given. IP-TV
Applications
Video conferencing
Leased capacity: Representing a pipe from
Video on demand
ingress to a set of egress points. Bit-rates may
be as for leased line; however, the real-time HDTV
requirements are less strict. P2P publishing
More detail can be considered when both direc- Coax with return
tions are quantified, see Figure 5. Considering
Systems
ADSL
the set of applications used simultaneously also
VDSL
gives indications of which access solutions (bit-
rates upstream and downstream) can be chosen. Fibre to the home (upgradable > 100 Mbit/s
Figure 6 Factors related On the other hand, when replacing/modernising
Service/products
to co-ordination between an an existing system, there is also the potential re-
operators business units Product bundling use or sale of redundant equipment which could
Responsiveness
Market (market dynamics) be taken into the equation. Then, the revenue/
income side may also be included in the overall
equipment cost calculations.
Trading-related
(systems, procedures) 4.4 Co-ordination-related
Operation/ For an operator having a wide set of systems, the
networks set of customers to support is also likely to be
diverse in terms of requirements. However, there
Scale-/scope-effects
(competence)
is a steady trend to integrate common solutions,
arriving at single systems within an operator,
potentially supporting a range of customer seg-
Interfaces, topology, ments and product types. This is likely to require
Integrating dedicated solutions co-ordination between business units within the
operator in order to successfully deliver services
to their respective customers. A simple example
is the use of a common IP network supporting
both residential and business customers, poten-
4.3 Costs tially having different dependability require-
There are a number of reasons why a network ments.
operator will invest in telecom equipment.
Firstly, new equipment could be installed pro- A number of factors influence the co-ordination
viding the new services that cannot be provided between units co-operating to deliver a product,
by current systems. Secondly, an existing system see Figure 6. The main ones discussed in the fol-
could be replaced in order to produce the ser- lowing emphasise trading between units:
vices more efficiently in a new system. Natu-
rally, the first and the second will also be com- Centralisation vs. distribution: Centralised
bined in several cases. A third reason is to ex- solutions are better for relatively simple, com-
pand an existing system in order to cover more mon tasks with small requirements for indi-
traffic load or customers. This is also likely to vidual changes and adaptation. For some other
provide the new versions (and hence a combina- cases, distributing the solutions is the better
tion of the first and second as well). choice.
The cost components come in various flavours, Centralised solutions imply a risk of becoming
including: least common denominator, i.e. not fully
compliant with individual needs. Also, (too)
Equipment cost both hardware and software late introduction of new functionality that only
as well as other vendor costs. In several cases applies to some application areas is a potential
recurrent license and maintenance costs are outcome when focusing on centralised solu-
given, or a cost level depending on the amount tions only. The downside of distributed solu-
of traffic/carried load. tions, obviously, is the risk that functions or
tasks that otherwise could have been com-
OSS costs, as a consequence of introducing bined, may be duplicated. On the other hand,
new equipment this can be either completely distributed solutions allow for different time
new OSS, or adaptations and integration frames when introducing changes and
related to existing systems. upgrades, taking into account varying local
needs. That is, needs reflecting individual
Installation costs business units.
Operational costs such as: As long as the size of the task/area is above a
- Cost of human resources minimum, a distributed solution will be best
- Other necessary infrastructure (e.g. trans- suited to adapt to variation in requirements
port layer support, fibre, copper line, etc.) and changes in environment. The scale/scope
- Necessary vendor support and licenses effect, however, would influence the mini-
- Training mum level, including equipment cost/utilisa-
- Equipment footprint, reflecting building tion, human resources, etc.
rental fees
- Power and cooling requirements Outsourcing: Outsourcing a particular task
- Recurring way of right costs and thus becoming one of several buyers pro-
vides some additional challenges:
- A well-defined and professional supplier/ Effort when changing Figure 7 Trade-off effort
customer relation must be defined. during initial phase and
modification/upgrading/
- The buyers possibilities of having prob- operational phase
lems solved rapidly may be reduced.
- The buyers possibilities of prioritising and

controlling error corrections and upgrades
are limited.
Internal trading: A demander of a certain

service within an operator has fundamentally
three alternatives: i) buy from an external unit,
ii) buy from another internal unit, iii) build it Initial effort
yourself. Whenever constraints are imposed
on internal trading (i.e. internal trading has
to be applied), the challenges in defining the
relationship are particularly great. A healthy
climate for internal trading requires freedom
and openness for all parties, i.e. any agree- ciently designed procedure commonly pays back
ment must be commercially viable for supplier in the long run.
and customer. This requires flexibility and
incentives for the business units involved. 4.5 Regulation-Related
(Including Competition)
Scale and scope issues: The following factors Providing public telecommunications services
may be considered for the scale effect (the and use of radio frequency spectrum are fairly
bigger, the better): strictly regulated. Operators have obtained
licenses in Norway to provide public services.
- Co-location of equipment may be easier, Moreover, a few operators are considered to
including common power supply, ventila- have significant market power in several areas
tion, etc. and thereby have to obey additional constraints,
like open and announced conditions for intercon-
- Operation/maintenance staff and gathering nection.
of competence
Examples of areas subject to regulation are:
- Planning expertise
Provision of public mobile services (NMT,
- Vendor contacts prices on equipment and GSM, UMTS)
support Provision of public telephony service
Provision of leased line services
- OSS-related number of systems and need Unbundling of local access lines (LLUB)
for interconnecting the systems
In addition, an operator may have to obey the
- Common arrangements for interconnection Universal Service Obligation (USO) with
with other operators/providers respect to public telephony, leased line services
up to 2 Mbit/s and access to digital telecommu-
The scope effect, advocating collective func- nication network (the latter referring to intercon-
tions, may allow for an easier introduction of nection as opposed to leased line).
new products and transfer of products between
different networks. Hence, this would affect Following additional rules due to holding Signif-
the service bundling challenge, in particular icant Market Power (SMP) imply that more
across different business units responsibili- issues have to be taken care of:
ties.
Transparency
The co-ordination effects also address a generic Non-discriminating
challenge for a system design, sketched in Fig- Cost-based pricing
ure 7; the effort spent during the initial design Separate accounting for different services
and set-up of a system commonly eases the
effort needed to introduce changes to a system. SMP regulation is valid both for fixed network
This also goes for the procedures and schemes to and mobile network services. For these services
follow between units both within an operator reporting to the regulating authority has to be
and between the operator and others; an effi- taken care of.
5 Trends are being bought out by others. On the other
hand, there are also several actors entering the
5.1 General telecom area such as the power supply compa-
Trends can be observed in several areas, includ- nies and local communities installing high
ing customer demands, market situation and capacity network capabilities at the same time
technology/system. Besides these, others may be as other facilities are placed into the ground.
regulatory, financial, macro-economic, etc. A
few samples of trend statements are given in the In addition to these come the global players
following. These are by no means exhaustive. and companies operating in other regions.
These may, assisted by their sheer size, be
Customer demands: What will customers ask able to operate at lower cost bases, realise
for in the future? Besides being a million- higher development capacity and exercise
dollar question on the business level, it is greater purchasing power. Hence, an on-going
also a topic that concerns many of the groups alliance trend, mergers and acquisitions are
engaged in network evolution. A few exam- expected. At the same time there will fre-
ples of trend statements are: quently be newcomers starting up within cer-
tain niches of the market.
- The overall market for fixed telephony has
become rather flat (in terms of revenue) in Technical issues: Steadily miniaturized elec-
developed countries and a decrease is tronics allow higher processing power and
expected. memory/storage capacity. Hence, more intel-
ligent terminals emerge, including terminal
- The rapid growth of voice in GSM seen pre- types with potential communication needs.
viously seems to be taking a break in West- This is seen by wireless communication being
ern Europe while the steady migration of integrated in various device types and that
voice and narrowband traffic from fixed to machine-to-machine communication seems
mobile networks has characterized the last to grow.
decade.
In the network equipment area, convergence
- New broadband accesses are primarily of equipment capabilities is observed. That is,
based on xDSL. Steadily increasing band- several systems (and vendors) have included
width demands due to increased penetration most options within their roadmap. Two
of broadband access and heavier use of the examples are provision of speech (telephony)
network. and Ethernet-based services. These may be
provided by several combinations of network
- Increasingly more cost-oriented business equipment. Another development is manifes-
customers, implying a trend of moving tation of acknowledged interfaces between
towards on-demand services to partly re- modules allowing for interconnecting units
place leased-line services. This is particu- from different vendors. It also allows poten-
larly true for small and medium enterprises tially different market segments to converge in
(SMEs). Other customers will also look for the sense that more services can be offered
means to reduce cost, like changing to less and requested in several segments.
expensive interface cards (e.g. Ethernet-
interfaces). The struggle between services provided and the
application of services should also be noted, see
Market situation: In several regions, the Figure 8. That is, services and applications may
providers are about to, or have recently gone be seen in a continual pursuit; where the service
through a consolidation phase. This implies provider tries to offer adequate services (bun-
that several smaller actors have given up or dles) while the applications try to utilise expect-
Service
bundling
User/ Networks/
Applications Services systems
customer
Network
portions
Figure 8 Relation between
services (delivered by Customer
value
networks/systems) Production
and applications means
2001 2000 1999 Table 1 Telenor operational-
revenue per cent revenue per cent revenue per cent
related revenue (excluding
(mill NOK) (mill NOK) (mill NOK) asset sales)
Analogue/Digital 13,668 36 % 12,802 38 % 13,313 45 %
(PSTN, ISDN, ADSL)
Mobile telephony 9,531 25 % 7,197 21 % 5,468 18 %
Leased line 1,065 3% 902 3% 810 3%
Satellite and TV 3,879 10 % 3,245 10 % 2,584 9%
distribution
Other network-based 2,633 7% 2,215 7% 1,593 5%
IT service and 5,009 13 % 4,738 14 % 3,501 12 %
installation
Advertisements, etc. 1,266 3% 1,555 5% 1,588 5%
Others 1,388 4% 1,040 3% 987 3%
Sum 38,439 100 % 33,694 100 % 29,844 100 %
2001 2000 1999 1998 1997
GSM subscriptions in
Norway, end of year
Fixed subscriptions 1,210,000 1,145,000 1,003,000 944,000 803,000
Prepaid 1,027,000 911,000 732,000 316,000 68,000
Churn rate (related to fixed) 12.5 % 12.7 % 14.2 % 13.1 % 13.9 %
Mobile originated traffic,

Norway, mill. minutes
GSM 2,969 2,298 1,801 1,279 711
NMT 64 108 174 271 331
ARPU GSM per month, NOK

Total 340 3,381 341 366 401
Fixed subscriptions 494 473 440 400 401
Prepaid 154 1,652 157 169 0
Telephony lines (fixed)

in Norway, end of year
Analogue (PSTN) 1,527,000 1,680,000 1,908,000 2,167,000 2,324,000
Digital (ISDN) 1,735,000 1,590,000 1,228,000 755,000 410,000
Telephony traffic (fixed)

in Norway, mill. minutes
Domestic, excl. Internet 10,567 11,612 12,371 12,911 11,923
Internet dial-up 4,974 5,667 4,255 2,059 1,079
International 383 387 415 386 379
To mobiles 1,412 1,295 1,246 967 727
Value added services 624 599 447 287 191
Pay-TV, number of
subscribers in Nordic,
end of year
Cable-TV 561,000 357,000 282,000 270,000 244,000
Smaller, closed Cable-TV 1,105,000 1,086,000 937,000 686,000 0
Satellite to residentials 657,000 506,000 405,000 352,000 251,000
Total 2,323,000 1,949,000 1,624,000 1,308,000 495,000
Internet, end of year

Subscribers/registered
users, Norway 831,000 625,000 400,000 260,000 165,000
Churn rate (subscriptions) 20 % 25.5 % 14 % 11.7 %
Nextra business
subscriptions, Norway 16,000 13,000 8,000 4,000 2,000
Nextra subscriptions,
outside Norway 106,000 104,000 57,000 0 0
Table 2 Trends in
Work force, man years,
end of year 21,000 20,150 21,968 20,226 19,598 Telenors operation
(from annual report 2001)
GSM prepaid the following, obtained from Telenors annual
Trends (1997 = 100)
1400 report for year 2001. As a perspective on the
dimensions behind different areas of Telenor,
1200
Table 1 shows turnover/revenue (million NOK)
1000 in 2001 compared to two previous years.
800 Nextra business subscribers
Table 2 gives more technical details, in particu-
600 Internet subscribers/users
lar for mobile systems (GSM, NMT), ISDN,
Total Cable-TV and satellite
400 PSTN, cable-TV and satellite, and Internet/IP-
ISDN lines
200
related operation. Numbers from the years 1997
PSTN lines
GSM fixed
2000 are included to indicate trends.
1997 1998 1999 2000 2001
Years The numbers clearly point out areas of growth
600
and reductions. In brief, analogue PSTN lines
have a fairly drastic decrease, compensated for
500 Trends (1997 = 100) by the increase of ISDN lines. However, the
Fixed internet dial-up
GSM overall voice-related traffic in PSTN/ISDN has
400
decreased (about 10 % in 2001). Areas of
Value-added services growth are GSM, Internet and cable-TV/satellite
300
in addition to ISDN. As expected, a significant
200 Fixed to mobile
decrease is also seen for NMT (to be phased out
Fixed international
100 Fixed domestic
a few years after 2001).
0 NMT
The trends are shown in Figures 9 and 10, hav-
1997 1998 1999 2000 2001
ing year 1997 and year 2001 as basis, respec-
Year tively. The former emphasises growth areas,
while the latter clearer shows the areas in
Figure 9 Trends in key indica- ed (future) services. Hence a spiral effect can be decline.
tors, referring to year 1997 as seen, although new technical solutions (and
basis (100 level) commercial concerns, billing, etc.) may intro- 5.3 System Life-cycle
duce more abrupt changes. Turning it around, The different systems have different maturity
in certain areas, this may also be seen as the levels and also address different levels of user
chicken and egg problem. needs. Some networks and associated technolo-
gies are still in the emerging phase, e.g. ADSL,
5.2 Status Background for Trends while others are quite mature, such as PSTN.
Figure 10 Trends in key indi- A sound basis for discussing trends of a com- Traditionally, a system has gone through a com-
cators, referring to year 2001 pany is to assess the actual status and concise petitive growth phase gaining as many cus-
as basis (100 level) historic numbers. A few illustrations are given in tomers/coverage as possible by customer acqui-
sition and network infrastructure roll-out. Fol-
lowing that phase, more emphasis is put on rev-
160 enues per customer, margins, churn rates and
Trends (2001 = 100)
GSM fixed subscr OPEX/CAPEX per customer. As a system even-
120 GSM prepaid tually enters a decreasing phase, the providers
PSTN lines attention is gradually shifted towards customer
80 ISND lines retention and system consolidation. This is illus-
trated in Figure 11, also sketching a shift in posi-
Total
40 tions in a future period.
Internet subscr
Nextra business sub The systems may not follow the same tempo
0
1997 1998 1999 2000 2001 through the different phases. That is, a system
Year might well pass another system, as illustrated,
600 comparing now with Y years ahead. During
Trends (2001 = 100) GSM mill min the different phases it is natural that the network
NMT mill min development is carried out according to different
400
Domestic
motivations. However, it is important that the
Internet
overall life span is considered in the overall net-
International
work planning avoiding that solutions are chosen
200
that makes the operation very complex or expen-
To mobiles
sive at a later stage.
VAS
0
1997 1998 1999 2000 2001 Another factor is that the decreasing phase
Year should be observed carefully, also reducing the
investment levels correspondingly, in order to ISDN/
ATM PSTN
steer clear of having much surplus equipment
GSM Leased
towards the end of system life. line
Now
Establishing such system life-cycles, scenario

exercises are central. The observations made ADSL
from scenarios will assist also when estimating IP
the tempo of the systems during the different X.25
phases. It should also be noted that not every UMTS VDSL WLAN
system has to follow the phases, in particular Emerging Growing Mature Declining Obsolete
systems that are considered as flops may have
IP GSM
difficulty reaching a growing and hence a Y years ahead
xDSL ISDN/
mature and declining phase.
PSTN
ATM
6 Scenario Work Qualitative WLAN
One should bear in mind that an approach based Leased
UMTS line
on scenario is commonly considered subjective.
Hence, several of the steps and choices made Optical
X.25
may not be uniquely inferred from the previous
steps. Still, the scenario approach can be applied Emerging Growing Mature Declining Obsolete
to capture a possible future in order to identify
the choices that one should prepare for.
An overall illustration of an approach is given in Using the sample of network roadmaps corre- Figure 11 Illustration of
Figure 12. As shown, the goal is to arrive at net- sponding to each scenario, the roadmaps are system life-cycles; state of
work roadmaps. Here the network roadmap revisited to find candidates. As expected, fewer the systems today and a
shows how the networks in the portfolio should sets of candidates are seen than there are scenar- possible shift some years
be developed in the time period considered. ios. Hence, these candidates are basis for elabo- ahead
Hence, for each scenario, a network roadmap is rating a decision tree and the migration plans.
derived. Deriving the network roadmaps, the
evaluation criteria and technology time lines
have to be taken into account.
Scenario A Scenario X
Scenario factors
Asdf asdf Asdf asdf

Sdf dfaf Sdf dfaf
Evaluation
criteria Fgoqe dfoqr Fgoqe dfoqr
Asdf dfasd Asdf dfasd
Gqert igwt Gqert igwt
Fglk oiurt Fglk oiurt
Trends,
technology
etc
A- Network roadmap X - Network roadmap
2002 2007 2002 2007
Collating candidates
Candidate X - Network roadmap

Candidate 1 - Network roadmap Figure 12 Overall approach
Input to
2002 2007 decision tree followed for deriving
and migration
plans candidates for network
roadmaps and mapping
of product categories
(could be iterative)
6.1 Arriving at Scenarios migrate and be applied. That is, when new func-
Deriving a set of scenarios, one faces the trade- tionality will be introduced (and in what year),
off between finding as few scenarios as possible and is the specific solution expected to increase,
at the same time as the complete feasibility be stable, or decrease in level of importance
space is covered. Here, the feasibility space (number of subscribers, traffic volume, etc.).
refers to all possible situations which may arise Moreover, a network roadmap also explains how
in the future. Naturally, a rather limited number the different solutions relate to one another. For
of situations have to be selected in order to have instance, an IP-based network might be carried
a tractable task. Still these situations should directly on an optical network and use ADSL,
include most of the challenges that can emerge. SHDSL, VDSL, Ethernet and WLAN as access
Deriving an adequate set of scenarios may be forms.
seen as an art, as several subjectively determined
conditions have to be included. However, when When devising the network roadmaps, informa-
open for iterations, one learns for each step, tion from the expected trends/time lines as well
incorporating these lessons when the scenarios as criteria is used. For example, the technology
are adapted. trends express views on when certain functions
will be available including statements on when
In order to describe a scenario, a number of sce- the solutions will be applied. This input is a nat-
nario factors have to be defined. These factors ural starting point to describe the migration, and,
will also to some extent assist when exploring considering the scenario characteristics, issues
whether or not the set of scenarios covers the from the technology time lines can be selected
feasibility space. Again, trade-offs are seen; the accordingly.
factors should be as few as possible (to be tract-
able) and still capture the main aspects. A high 6.3 Collating Network Roadmap
level of subjective judgement is involved when Candidates
expressing the scenarios by the scenario factors. For each of the scenarios, a network roadmap
(and product mapping) has to be derived. It is
A brief scenario description of a scenario defini- likely that fewer different network roadmaps
tion would then contain: will appear than the number of scenarios. There-
fore, a reverse processing could be applied,
List of key characteristics meaning that the resulting roadmaps are put
Short story of what happened/is happening together. Given that these have the same starting
Short description of operators situation point, they will differ by taking certain (differ-
Grading of the scenario factors ent) steps at certain instances in time. In princi-
ple, this can be thought of as a decision tree; fac-
Even though one would start with a set of sce- ing an instance when two scenarios differ
narios and carry on deriving network roadmaps, expresses that a decision has to be made. The
it is likely that the scenarios would be adjusted decisions may also be more related to certain
based on the lessons learned. Hence, a few itera- events and less strictly to time instances.
tions might take place until the scenarios seem
to address the most likely and central issues. An Correlating these network roadmaps with the
alternative would be to derive the scenarios from operators situation, more figures will be intro-
the scenario factors. Although these factors may duced as the method so far has been followed
not be independent, quite a few combinations mostly in a qualitative manner. This is done as
will appear, leading to an intractable number of part of elaborating network migration plans.
scenarios. On the other hand, these could still be
used as a starting point for identifying the sce- To speed up this process, it may be more effi-
narios. cient to start by raising the main questions relat-
ing to network migration. That is, devising these
6.2 Deriving Network Roadmaps questions and then adapting the scenarios corre-
from Scenarios spondingly.
Given a scenario, certain key characteristics and
main trends are included. The idea is then that 6.4 Scenario Factors
these would motivate for a corresponding migra- Working on scenarios, a set of factors has to be
tion of the network portfolio. Again, subjective devised. These factors apply when the scenarios
considerations are behind the story of expressing are described, i.e. how the scenarios are placed
a network roadmap corresponding to the sce- into the space spanned by the factors. In prin-
nario. ciple, one could use the set of factors to identify
potential scenarios. However, considering the set
A network roadmap shows how the different of factors the number of potential scenarios
network solutions (e.g. ADSL, VDSL, Cable might then become too large.
network, WLAN for access network) will
The work on scenarios would likely be carried Figure 13 Scenario factors
out iteratively, meaning that lessons learned Environment express how uncertain aspects
from describing scenarios and belonging net- evolve as seen by an operator
work roadmaps are used to return to the set of
factors and potentially revise them. Note, how-
Operator
ever, that the scenarios as such are not the main
- network roadmaps
outcome of the network planning, but are mostly - .....
used when dealing with the strategic perspective.
Still, observations made could well be forwarded
to other processes within the operators sphere.
The scenario factors express the uncertainty

attached to a future evolution. Hence, assigning
values/grades to a factor fixes how that aspect
= Scenario factor
evolves. The factors selected have been formu-
lated as questions; like A versus B, where A and
B are opposite choices on a scale. It is also
implied that the scenario description refers to the
environment as observed by an operator (Fig- demands in a common implementation.
ure 13). Hence, a common packet-based carrier net-
work could be realised, transporting most of
In some of the cases the different factors may the client traffic flows.
not be orthogonal (or independent). In these
cases, further work may be done to revise the set Scenario II: TDM-based services dominate the
of factors based on the knowledge gained and market although packet technology is capable
the results one wants to look more closely into. of transporting the different traffic flows. This
Again, however, the scenarios themselves are could for example happen when the TDM
not considered the main results in the network equipment is much cheaper than the packet
planning; rather these results are the potential equipment.
and possibly recommended roadmap of the net-
work portfolio. Scenario III: TDM-based services dominate
the market at the same time as the packet tech-
Examples of scenario factors and corresponding
scenarios are given in the following section.
7 Two Scenario Examples
7.1 Scenario Space With Two Axes TDM TDM Packet Packet
Assuming that one main question is whether or
not a common packet-based (core) network
should be used to carry the traffic, one should Figure 14 TDM or packet as carrier; left TDM only,
bear this in mind when deriving the scenarios. middle TDM and packet in combination, right packet only
Here it is assumed that the current situation is to
have a TDM-oriented common carrier. These
technical combinations are depicted in Figure
14. The blue boxes on top refer to client sys-
tems, e.g. ISDN exchanges, ATM switches, IP
Packet technology
routers, customer accesses. sufficient for all needs
Then we assume that there are two main uncer-

tainties; i) the technical feasibility of supporting Scenario II Scenario I
all the traffic on the packet-based network, and
ii) the market demand for packet-based services TDM market dominates Packet takes significant
versus more TDM-oriented services. The sce- market share
nario space could then be made as in Figure 15.
Scenario III Scenario IV
Illustratively the four scenarios can be described

as:
Packet technology needs
Scenario I: Packet services have taken a sig- to be specifically adapted
nificant market share at the same time as
packet technology is able to service the Figure 15 Two axes to illustrate the scenario space
Packet technology
Figure 16 Referring network
sufficient for all needs Scenario I
configurations to scenarios Scenario II
(T = TDM-based,
P = packet-based)
TDM T Packet
TDM market dominates Packet takes significant

market share
TDM T P P P

to be specifically adapted
Figure 17 Describing Packet technology

potential network roadmaps Scenario II sufficient for all needs Scenario I
referring to the scenarios Path II Path III
2008 2008
2008
TDM T Packet
TDM market dominates Packet takes significant

market share
Path I Path IV
2005
2008 2008
TDM T P P P

2003 to be specifically adapted
nology needs to be specialised to its applica- The network configurations relating to the four
tions. For example, common packet networks scenarios are depicted in Figure 16. Having
for different market segments cannot be described these configurations, potential migra-
realised. tion paths could be introduced as well. For ex-
ample, assuming that the current configuration is
Scenario IV: Packet services take on a signifi- at the lower left square, theoretically four paths
cant market share, although the packet tech- (and hence target states) could be identified,
nology needs to be adapted to its use. This each path reflecting a scenario. A more involved
could for example happen if the packet imple- case could be present as well, such as when
mentation is cheaper than the TDM-oriented some of these configurations could be a state on
way of constructing a network. the way towards a target in another square (see
path V in Figure 17). Another variation is that
The four scenarios can be looked upon as the same path is followed in the early migration
extremes to illustrate the host of possibilities that phases, although one does not reach as far along
exist. Practical and optimised solutions from a this path as believed; that is, the target state may
technical and commercial viewpoint may how- be plotted at a closer stage along the migration.
ever show that intermediate solutions between
these four are to be preferred.
7.2 Applying More Scenario Factors Drivers for evolution Scenario span
For a more elaborative scenario description, a
User behaviour Best-effort data, voice Reach real-time media,
number of factors are commonly defined. One moving to mobile seamless mobility
example is given in Table 3. Again, it has to be networks
remembered that the most significant factors
Customer ownership Network operators Value-added service
should be described, which both are chosen
providers
based on subjective evaluations and decided
by the main network-related questions faced. Business model Integrated carriers Fragmented value chain
Regional balance US stays leading Asia-Pacific drives

Utilising these scenario factors, a number of sce-
market innovations
narios can be devised two examples are illus-
trated in Figure 18. In addition to grading the Network-or Intelligence in Intelligence in network
scenario factors, a story has to accompany each terminal-centric terminal
of the scenarios. This should be a likely explana- Regulatory impact Laissez faire Strict regulation
tion how that scenario may occur. Naturally, on
the management level, one would also like to Rise of global Fragmented, national Coordinated, global
know the distribution of market power between operators
the actors present in a market and the roles in the
value chain. An illustration of this is given in Table 3 Example of scenario factors
Scenario A - operator as bit carrier Figure 18 Samples of

User behaviour -
Best-effort data, voice to mobile Reach media, seamless mobility
scenarios defined by the
scenario factors
Customer ownership Network operator Service provider
Business model Integrated carriers Fragmented value chain
Regional balance US in lead Asia-Pacific innovates
Intelligence dominance Terminal-centric Network-centric
Regulatory Laissez faire Strict regulation
Operator footprint Fragmented, national Coordinated, global
Scenario B - operator supports full-service
User behaviour Best-effort data, voice to mobile Reach media, seamless mobility
Customer ownership Network operator Service provider
Business model Integrated carriers Fragmented value chain
Regional balance US in lead Asia-Pacific innovates
Intelligence dominance Terminal-centric Network-centric
Regulatory Laissez faire Strict regulation
Operator footprint Fragmented, national Coordinated, global
Value
distribution
Historic, e.g. voice
Tomorrow
Application, e.g. Application/service Connectivity, IT-equipment, IT skill,

creative software provisioning infrastructure, software integration
operation,
connectivity Figure 19 Shift in value chain
(although depending on
market/competition situation)
Figure 20 Qualitative illustra- Access networks
tion of system trends for 2003 2004 2005 2006
different access networks
Power line DTT-IP
related to a set of scenarios
Broadcast
Cable network
Feeding CDN
Satellite
ADSL
HDSL
SHDSL
Driven by TV
VDSL
Rural areas?
LMDS Business segment
G.Ethernet
PSTN/ISDN
WLAN
GERAN
GSM - BSS
UTRAN
Figure 19, showing a potential shift in value dis- tional expenses must also be considered, given
tribution related to a set of scenarios. by staffing, license agreements, and so forth.
After describing the scenarios, actions and con- The description is based on an example where
sequences on the network portfolio must be the question is how to carry traffic flows for a
derived. Figure 20 shows an example of how a number of client systems between a number of
number of access systems could evolve related locations in a core network. However, the same
to a set of scenarios. As described above, each principles apply to other areas as well. Three
of the scenarios should have a network roadmap types are assumed, SDH, ATM and IP/MPLS.
associated. Then, these resulting network road- These are also referred to as aggregators in the
maps are collated to identify in what manner following.
they differ and the phenomena that influence the
decisions to be made for the different systems. Figure 21 depicts the overall procedure followed
when carrying out the calculations. An end-
The scenario-based approach provides insight result is the investments needed in order to carry
into the factors that influence the system evolu- the traffic loads. A study period of a number of
tion. This is commonly done on a qualitative years is given, hence, all the major input data
level. The approach has to be complemented must be specified for each of the corresponding
with calculations as described in the following years. This also allows for an evolution of the
section. The qualitative observations, however, input data, which is essential at least for traffic
can limit the number of candidates that should demands and for component prices.
be input for the calculations. Moreover, they
also try to systematise the uncertainties and risk A major bulk of input data is composed of the
factors to be followed. traffic matrices specified as traffic load (e.g. in
Mbit/s) between the locations (all locations as
The qualitative evaluations may also reveal a specified for the network, also giving the amount
number of winning systems that seem to be of traffic to/from abroad). A number of traffic
safe to instal or enhance. This is also a valuable types are specified.
result that should be considered in the following
work. A number of parameters give the dimensioning
requirements for each of the traffic types. These
8 Qualitative Evaluations parameters are:
Calculating Cost of Network
Roadmaps The ratio of traffic that should be kept during
A complete calculation of network roadmaps a failure situation (in the range 0 .. 1),
should ultimately fill requirements for business
case studies. However, there is often much The maximum link load that can be realised
uncertainty, particularly related to revenues. The (in the range 0 .. 1),
following sections only include the investment
aspects. To capture the total cost picture opera-
Protection ratio
... 200x Max link load ratio
Traffic Busy interval ratio

from-to
Per traffic type and per year
Various traffic sources (services/customer groups)
Effective traffic per traffic type

and per year
Relating locations
Each location
Mapping traffic types onto
... 200x ... 200x ... 200x Mapping traffic(per
aggregators types onto
year)
aggregators (per year)
Traffic on SDH Traffic on ATM Traffic on IP/MPLS
equipment equipment equipment
... 200x
Configuration
- IP/MPLS Geographic layout
- ATM e.g. distances
- SDH
- WDM
... 200x
... 200x
Overall equipment (IP/MPLS, ATM, SDH, WDM): _
- base Installed base (first year) and store ;
- interfaces IP/MPLS, ATM, SDH, WDM
... 200x
Additional equipment; IP/MPLS,

ATM, SDH, WDM
... 200x
Price lists; IP/MPLS,

ATM, SDH, WDM
Investment/ economics
Case Investment in Mill. NOK

Spare Invest Sum
2002 2003 2004 2005 2006 2007
Investment equip. -Spare 02-07
IP-related 9.6 2.9 4.7 6.5 11.3 8.6 4.9 38.7 43.6
SDH-related 3.2 0.7 2.3 0.0 0.9 1.5 0.0 8.7 8.7
ATM-related 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
WDM-related 1.1 0.5 1.1 3.2 3.4 5.2 0.0 14.6 14.6
sum 14.0 4.0 8.2 9.7 15.7 15.3 4.9 62.0 66.9
Investments 2003 - 2007

18
IP-related
14
SDH-related
10 WDM-related
2
Figure 21
2003 2004 2005 2006 2007 Schematic
Year
flowchart for
calculations
The relative portion of the peak traffic load The traffic loads on the aggregators give the
that is present in the dimensioning interval, configurations of each aggregator. Directions
i.e. a busy interval indicator (in the range 0 ..1). must also be included (that is where links are
going) as well as the distances between the loca-
In combination, the two former indicate the traf- tions (in particular for WDM equipment).
fic situation during a failure situation. For in-
stance when a failure occurs, a ratio of the traffic Adding equipment types on all locations gives
(on the failed link) is carried by an alternative the overall needed equipment (for base configu-
link and the maximum load on that link is given ration, interfaces, etc.). In order to consider
by the maximum link load parameter. actual deployment and equipment reuse between
different years, an equipment store is present.
The result of taking these parameters into Hence, only when needed equipment of a certain
account is called the effective traffic load. This type in a year exceeds the sum of that equipment
traffic load is given for each traffic type and type in the previous year and the number in the
each year. A step further is to consider the rela- store is additional equipment bought.
tions between the locations. This means to con-
sider where links are placed (present between The result after looking at installed equipment,
certain locations). Relating the effective traffic stored equipment and needed equipment is a
loads, mapping of traffic types onto aggregators list of additional equipment that needs invest-
and presence of links decide the traffic load on ment funds. Considering the price of that equip-
each aggregator type at each location. ment for the corresponding year gives an invest-
ment level. These investments can be aggregated
in different ways, for instance per equipment
class, per year, for the whole period, etc.
Internet access pessimistic, L1LL baseline Internet access optimistic, L1LL baseline
100 100
IP
IP
ATM
80 ATM 80
SDH
SDH
60 WDM
WDM 60
40
40
20
20
0 0
Case A1 Case A2 Case B1 Case B2 Case C1 Case C2 Case D Case E Case A1 Case A2 Case B1 Case B2 Case C1 Case C2 Case D Case E
Internet access optimistic, L1LL maximum

Internet access optimistic, L1LL minimum
100
100 IP IP
ATM ATM
SDH
80
80 SDH
WDM
60 WDM
60
40 40
20 20
0 0
Case A1 Case A2 Case B1 Case B2 Case C1 Case C2 Case D Case E Case A1 Case A2 Case B1 Case B2 Case C1 Case C2 Case D Case E
Internet access optimistic (50 % abroad), L1LL baseline

100
IP
80 ATM
SDH
60 WDM
40
20
Figure 22 Costs of IP/MPLS, ATM, SDH and WDM

0
Case A1 Case A2 Case B1 Case B2 Case C1 Case C2 Case D Case E equipment for cases and demand variations
In addition to economic output, traffic and baseline and Internet access optimistic + L1LL
capacity results can also be recorded. maximum.
Typically, calculations are carried out for a num- For the variant Internet access optimistic +
ber of cases and the results are compared. These L1LL maximum cases A1 and D come out with
cases can be specified by the technology used, equal overall investments when the IP/MPLS
years for introducing functionality, variation in cost is reduced by 25 %. For other combinations
traffic demands, changes in equipment prices, the IP/MPLS costs have to be reduced by about
and so forth. In addition different network archi- 75 % in order for the lines to intersect.
tectures would be examined. An intention is to
find out the more critical parameters, that is, 9 Risk Analysis and Exit
those significantly influencing the results and the Strategies
network roadmaps that should be preferred. At As described earlier, a number of decisions will
the end of the calculation period, the installed commonly be revealed during the network strat-
equipment could be assigned a value (terminal egy evaluations. For each step, a certain risk
value) depending on how investment levels level is associated. In this section a general
should be estimated. introduction to incorporate risks into the evalua-
tion given.
For most transmission systems the capacity often
comes in certain granularities, e.g. 155 Mbit/s, 9.1 Risk General Introduction
622 Mbit/s for SDH. This has to be included in Considering the constant change and uncer-
addition to any consequent costs such as man- tainty, an integrated risk management practice
agement systems and installation work. within an organisation is required to strategically
deal with uncertainty and thereby capitalise on
An example of investment results is given in opportunities. The stakeholders in the decision
Figure 22. These are shown for five different process will also be involved in order to ensure
combinations of traffic demands and where eight better decisions in the future.
cases are examined (A1, .., E). One reason be-
hind breaking up the investment cost showing A risk management arrangement should cover
network types is to give ideas on where more all types of risks that face the organisation,
investment savings could be obtained. including policy, operational, human resources,
financial, legal, health and safety, environment,
An immediate observation from the results reputation. Hence, it is essential to integrate risk
above is the fairly high ratio of costs relating to management into strategic decision-making. By
IP/MPLS equipment. In order to look more establishing a risk management framework, a
closely into the effect of varying costs of this mechanism will be in place allowing to discuss,
equipment the following calculations were con- compare and evaluate different risk types.
ducted; all IP/MPLS equipment costs were
changed by x % (in the interval from 75 % to Deploying a risk management regime also
+50 %). Only cases A1, A2 and D are shown in allows for a so-called risk-smart workforce that
Figure 23 as the objective was to estimate points supports innovative and responsible risk-taking
of intersections. Two demand variants are while ensuring legitimate precautions.
included: Internet access pessimistic + L1LL
mill. NOK
140
IP/MPLS price variations
120 Intern-pes; L1LL-base A1 sum
100 Intern-pes; L1LL-base A2 sum
80
Intern-pes; L1LL-base D sum
60
Intern-opt; L1LL-max A1 sum
40
Intern-opt; L1LL-max A2 sum
20
Intern-opt; L1LL-max D sum Figure 23 Impact on IP/MPLS

0
cost variations for Cases A1,
-75 % -50 % -25 % 0% 25 % 50 % A2 and D
Risk is unavoidable and present in almost every els as it concerns making decisions that con-
situation in every-day life, for private as well as tribute to the achievement of an actors objec-
business roles. A number of definitions of risk tives. Hence, integrated risk management is a
are in use. However, a common concept in all continuous, proactive and systematic process to
definitions is that a certain level of uncertainty understand, manage and communicate risk from
of the outcome is involved. One way the defini- an actor-wide perspective. It is about making
tions differ is the characterisation of the out- strategic decisions that contribute to the achieve-
comes; some describe risk as having only ment of an actors overall corporate objectives.
adverse consequences, while others are neutral. Integrated risk management requires an ongoing
An ongoing discussion takes place arriving at an assessment of potential risks. Hence, it should be
acceptable (common) generic definition of risk embedded in the actors strategy and risk man-
that recognizes the fact that when assessed and agement culture. As stated above it is not limited
managed properly, risk management can lead to to minimising the risks, but rather to foster inno-
innovation and opportunity. This appears even vation in order to achieve greatest returns with
more prevalent when dealing with operational acceptable results, costs and risks. Hence, an
risks and in the context of technological risks. optimal balance is strived for at the corporate
One definition found is the expression of likeli- level.
hood and impact of an event with the potential
to influence the achievement of an actors objec- As pointed out in [Cari01], four key elements
tive. Hence, risk refers to uncertainty around are included in an integrated risk management
future events and outcome. framework:
Other definitions found are e.g.: 1. Develop the corporate risk profile consider-
ing objectives and available resources
Combination of the probability of an event a. Identify risks; resulting in description of
and its consequences. Note in some situa- threats and opportunities, i) type of risk
tions, risk is a deviation from the expected, technological, financial, human resources,
health, ii) source of risk external, internal,
The chance of something happening that will iii) what is at risk area of impact/type of
have an impact on objectives. It is measured exposure, iv) level of ability to control the
in terms of consequences and likelihood, risk high (operational), moderate (reputa-
tion), low (natural disasters).
The chance of injury or loss defined as a
measure of the probability and severity of an b. Assess current risk management status;
adverse effect to health, property, the environ- resulting in descriptions of challenges/
ment or other things of value, opportunities, capacity, practices.
The possibility that one or more individuals or c. Identify risk profile; description of key risk
organizations will experience adverse conse- areas, risk tolerance, ability and capacity to
quences from an event or circumstance. mitigate and needs for learning. In general,
there seems to be lower risk tolerance for
A risk measure can be formalised as the unknown, where impacts are new, un-
observable or delayed. In addition a higher
risk = risk tolerance is observed where people feel
more in control (e.g. for car travel com-
probability(event i) consequence(event i)
event i
pared to air travel).
The summation is taken over all events consid- 2. Establish an integrated risk management func-
ered and the product of each events probability tion
of occurring and consequence is added. a. Communicate, understand and apply man-
agement direction on risk management; risk
In order to take advantage of the present risk fac- management needs to be aligned with an
tors a management framework should be intro- actors overall objectives, corporate focus,
duced. Risk management is defined as a system- strategic direction, operating practices and
atic approach to setting the best course of action internal culture. When a strategic risk man-
under uncertainty by identifying, assessing, agement direction is set up, both internal
understanding, acting on and communicating and external concerns, perceptions and risk
risk issues. As risk management is directed at tolerances are taken into account. It is
uncertainty related to future events and out- imperative to identify acceptable risk toler-
comes, it is implied that all planning exercises ance levels so the unfavourable outcomes
encompass some form of risk management. Risk can be remedied promptly and effectively.
management is also impacting people at all lev-
b. Implement operational integrated risk man- - Decide on necessary people, expertise,
agement through existing decision-making tools and techniques (e.g. scenarios,
and reporting structures. Integrating the risk brainstorming, checklists).
management function into existing strategic
management and operational processes - Perform a stakeholder analysis (deter-
ensures that risk management is an integral mine risk tolerances, stakeholder posi-
part of day-to-day activities. tion, attitudes).
c. Build capacity through development of B. Risk assessment

learning plans and tools. Building risk man- 2. Assess key risk areas:
agement capacity is an ongoing challenge - Analyse context/results of environmen-
even after integrated risk management has tal scan and define types/categories of
become firmly entrenched. Environmental risk to be addressed, significant organi-
scanning will continue to identify new areas sation-wide issues, and vital local
and activities that require attention, as well issues.
as the risk management skills, processes,
and practices that need to be developed. 3. Measure likelihood and impact:
- Determine degree of exposure, ex-
3. Practise integrated risk management pressed as likelihood and impact, of
a. Consistent application of common risk assessed risks, choose tools.
management at all levels. A common, con-
tinuous risk management process assists an - Consider both the empirical/scientific
actor in understanding, managing and com- evidence and public context.
municating risk.
4. Rank risks:
b. Integrate results of risk management prac- - Rank risks, consider risk tolerance, use
tices at all levels into informed decision- existing or developing criteria and tools.
making and priority setting
C. Respond to risk
c. Apply tools and methods 5. Set desired results:
- Define objectives and expected outcome
d. Consult and communicate with stakeholders for ranked risks, short/long term.
4. Ensure continuous risk management learning 6. Develope options:

a. Establish supportive work environment - Identify and analyse options ways to
where learning from experience is valued, minimise threats and maximise opportu-
and lessons are shared nities approaches, tools.
b. Build learning plans into actors risk man-

agement practices
c. Evaluate results of risk management to sup-

port innovation, learning and continuous
improvement
Continuous learning
d. Share experiences and best practices and communication
A potential risk management process wheel is

1: Identify
illustrated in Figure 24. Internal and external issues, setting
communication and continuous learning improve 9: Monitoring, context
evaluating and
understanding and skills for risk management adjusting 2: Assessing
key risk
practice at all levels of an organisation, from
areas
corporate through to front-line operations. The 8: Imple- Practise integrated risk
menting management (from
following steps may be included in a risk man- the strategy 3: Measuring
corporate strategy and
agement process (from [Cari01]): likelihoood
plans to front-line
and impact
7: Selecting operations people and
processes)
A. Risk Identification a strategy
1. Identify issues, set context: 4: Ranking
6: risks
- Define the problems or opportunities, Developing 5: Setting
scope, context (social, cultural, scien- options desired Figure 24 A potential risk
tific evidence, etc.) and associated risk results management process
issues. (adapted from [Cari01])
Risk management actions Modelling tools: such as scenario analysis and
Impact
forecasting models to show the range of possi-
Considerable Extensive bilities and to build scenarios into contingency
Must manage and
Significant management management
monitor risks plans.
required essential
Risks may be Framework on the precautionary approach: a

Moderate Management Management
worth accepting
effort worthwhile effort required
principle-based framework that provides guid-
with monitoring ance on the precautionary approach in order to
improve the predictability, credibility and con-
Minor Accept risks
Accept, but Manage and sistency of its application across several units.
monitor risks monitor risks
Qualitative techniques: such as workshops,
Low Medium High questionnaires and self-assessment to identify
Likelihood
and assess risks. Internet and organisational
intranets: promote risk awareness and man-
agement by sharing information internally
Figure 25 A potential risk 7. Select a strategy: and externally.
management model - Choose a strategy, apply decision crite-
ria result-oriented, problem/opportu- Each assess risk can be illustrated according to
nity driven its likelihood and impact as depicted in Figure 25.
- Apply, where appropriate, the precau- 9.2 Risk Allocation

tionary approach/principle as a means Quantitative Aspects
of managing risks of serious or irre- In principle, two risk types can be defined,
versible harm in situations of scientific depending on the source of the risk:
uncertainty.
Systematic risk coming from influences on
8. Implement the strategy: the market in general. An example is change
- Develop and implement a plan. of interest rate.
D. Monitor and evaluate Non-systematic risk coming from other

9. Monitor, evaluate and adjust: sources than those affecting the market in
- Learn, improve the decision-making/ general.
risk management process locally and
organization-wide, use effectiveness cri- Considering a portfolio, the risk can be de-
teria, report on performance and results. scribed by its exposure to the systematic factors,
the volatility of those systematic factors, and the
Each function or activity considered has to be residual (or non-systematic risk). Regarding the
examined from three perspectives: systematic factors, an actor can gain passive
exposure to these with little effort (or cost).
Its purpose: risk management would look at
decision-making, planning, and accountability The non-systematic risk, in contrast, comes from
processes as well as opportunities for innova- the on-going decisions actors make in managing
tion. portfolios over time. When the actor is actively
revising portfolio exposure to sources of system-
Its level: different approaches are required atic risk, these decisions generate non-systematic
based on whether a function or activity is risk in a portfolio in addition to any systematic
strategic, managerial or operational. risk embedded.
The relevant discipline: the risks involved The pay-off for bearing risk is commonly scaled
with technology, finance, human resources, according to the level of risk assumed. For sys-
and those regarding legal, scientific, regula- tematic risk, the ratio involves dividing the risk
tory, and/or health and safety issues. premium by the level of systematic risk:
A number of techniques can be used to assist in Risk premium

Sharpe ratio =
risk management, including: Systematic risk
Risk maps: summary charts and diagrams that For non-systematic risk, the ratio involves divid-
help organisations identify, discuss, under- ing the non-systematic return by the level of
stand and address risks by portraying sources non-systematic risk. This ratio is often referred
and types of risks and disciplines involved/ to as the Information Ratio (IR):
needed.
N on systematic return The information ratio is commonly a measure of
IR =
N on systematic risk how much non-systematic return that is expected
relative to the amount of non-systematic risk. In
For a portfolio, the return (pay-back) effects are general it is more preferable to obtain a higher
additive between systematic and non-systematic information ratio:
risk. Hence, in variance terms:
The information coefficient indicates how
2P = 2S + 2N + 2CovSN, accurate the actor is in forecasting future
returns. This is commonly estimated by com-
where 2 indicates the variance (subscripts: puting the correlation between forecast returns
P total portfolio excess return, S systematic and subsequent non-systematic returns. Dif-
excess return, N non-systematic return) and ferent options with a high level of predict-
CovSN indicates the covariance between the sys- ability have higher ICs than those with lower
tematic and non-systematic returns. level of predictability.
For a typical well-diversified portfolio, the non- When an actor wants to increase the informa-
systematic risk makes up a relatively small portion ratio of an active strategy, this is to
tion of the total variance of the portfolios increase the information coefficient, either
return. by finding better predictors of future relative
returns or by recruiting individuals with better
For an actor to allocate total risk in a portfolio insight than others. Given the competitive
between systematic and non-systematic sources, nature of the market it is not that easy to
it is important to know the sources of risk for increase the information coefficient.
each alternative strategy that is considered as a
candidate for inclusion in the portfolio. An im- The transfer coefficient indicates how effi-
portant characteristic of most alternative strate- ciently the actors information is used in form-
gies is that a greater portion of the total risk ing portfolio positions. The more constraints
comes from non-systematic risk instead of from that are placed on the portfolio, the lower the
systematic risks. transfer coefficient usually becomes. In a gen-
eral case, constraints are commonly placed on
An actor has the decision of how to i) allocate the size of individual positions or combina-
the risk budget across systematic factors, and ii) tions of positions (e.g. engagements in differ-
establish trade-off between systematic risk and ent countries, obligations to provide services,
non-systematic risk. This is the case even if the etc.). Relaxing any of these constraints tends
investor is not considering investing in any non- to increase the transfer coefficient and
traditional strategies. improve the information ratio.
In these discussions it is essential to estimate the The breadth indicates how many opportunities
expected return from active management of each the actor has in applying the information
strategy (to exploit the non-systematic factors). gained. Hence, this shows the number of deci-
One source of the expectations is the historical sions that can be made during a process. This
performance of different types of actively man- number is a function of both the number of
aged strategies. In some segments, possible elements in the portfolio and the frequency of
gains may be high, while in other segments, decision-making. On the other hand, there is
there is less opportunity for additional gains. often a cost side of increasing the frequency
of decision (such as preparing decision basis).
An expected information ratio may support the
understanding of actively managed portfolio. Considering these factors, in terms of risk bud-
This can be illustrated as: geting, it is observed that a greater amount of
non-systematic risk is allocated to the strategies
E[] that: i) focus on more inefficient markets (higher
IR = = IC T C N ,
expected information coefficient), ii) are subject
to fewer restrictive constraints (higher transfer
where: coefficient), iii) have greater breadth (combina-
IR = information ratio tion of a large universe to choose from and high
E[] = expected non-systematic return frequency of portfolio management decisions).
= non-systematic risk
IC = information coefficient
TC = transfer coefficient
N = number of decision points
(often referred to as the breadth)
Cash flow Performing the NPV calculations, using a proper
value for the periodic (discount) rate, r, is essen-
tial. This rate must reflect the cost of capital, that
is an investors rate of return when making an
NPV investment. The discount rate should then both
cover a compensation for the time-value of
money as well as a risk premium as seen by the
0 1 2 3 4 5 6 9 10 period
7 8 investor.
The Capital Asset Pricing Model (CAPM) is a

frequently applied approach for assessing the
discount rate. This postulates that the cost of
capital, r, equals the return on risk-free securi-
Figure 26 Net Present Value 9.3 Associating Gains with ties, , plus the market price of the risk (market
capturing the cash flow of Options/Choices premium), ( ), multiplied by the companys
an activity for a number On the one hand, relating to uncertain factors systematic risk, . Hence,
of periods represents a risk to be included in the evaluation.
a
On the other hand, in case a final decision can be r = + ( ) , where = .
postponed until more information is available,
one would also consider a future option to Here the systematic risk, , is expressed as the
decide as a positive effect. Hence, the latter ratio between the covariance of the activity, a,
should be included in the evaluation of a port- and the variance of the market, . The risk-free
folio consideration where all future decisions return, , is the rate one should obtain for an
could be added to the overall calculations. investment without any risk and can be approxi-
mated with government securities.
A fairly standard way of evaluating an activity
is to apply the Net Present Value (NPV). This Then, the market price of risk becomes the dif-
measure collapses an activitys monetary timely ference between the expected rate of return and
schema into a single value. This is done by ad- the risk-free rate; E[] .
justing future cash flows and reflecting them
into a current time value, see Figure 26. As seen from above, the systematic risk is the
correlation of the activity return with the market
Noting the cash flow in period i by CFi and the divided by the variance of the market return.
periodic rate by r, a common way of expressing This may be a challenging task, allowing for
the NPV is: subjective decisions by selecting activities
belonging to the market (i.e. a portfolio of
CFi comparable activities).
NPV =
i
(1 + r)i
The CAPM approach is applied to estimate the
By simply using this measure as a decision foun- cost of capital and thereby the discount rate to
dation, the activity with the highest NPV is cho- apply in the NPV analysis. Introducing uncer-
sen. Naturally, uncertainties are attached both to tainties in the NPV approach may be done in
the cash flow in a certain period and to the set of several ways; One way is to use simulations for
rates (which might vary between periods). generating cash flows in each period. The
result is a distribution for the NPV which then
Basing decision on a single parameter like NPV, can be used to deduce observations with respect
a separation between the investment and finance to whether or not to go for an investment. An
matters has been assumed. That is, the invest- example of such a measure is the probability of
ment decision is separate from the financing having an NPV below a certain threshold.
decision. An effect of this is that only the result-
ing NPV is looked at without considering the One of the drawbacks of the traditional NPV cal-
(cumulative) cash flows in selected periods. The culation is that the future cash flows are often
cumulative cash flows indicate the amount of modelled as deterministic. Moreover, once an
money needed to finance an activity. activity is initiated, it can simply be passively
managed. In practise, however, the cash flows
Comparing NPV to 0 (zero) means that the deci- are uncertain and the activity may commonly be
sion rule is that investments could be increased adjusted/changed by more active management.
as long as the marginal return on investment is Often traditional NPV can be referred to as static
greater than the cost of capital (determined by NPV calculations.
the finance market).
In trying to capture the uncertainties, a set of
scenarios can be devised. These scenarios could
be identified in a number of ways; i) defining a 9.5 Real Options
set of uncertain factors and applying combina- A net-present value (NPV) factor is usually
tions of these to derive the scenarios, ii) defining derived for an activity and decisions made on this
a most likely outcome, together with a worst and basis. A traditional NPV forces a decision to be
a best outcome, iii) defining a set of scenarios made on the current set of information about the
depending on the outcome of (external) major future. On the other hand, option valuation allows
factors. for additional flexibility of main decisions in the
future, as more information will be available. This
The NPV can then be calculated for each of the option of postponing a decision is captured in the
scenarios. As described earlier, the scenarios option method that results in two additional types
help understand the uncertainties and derive the of values due to the simple fact that one will
major factors to be observed and trigger actions. always want to postpone a final decision as long as
On the other hand, a number of drawbacks can possible. Firstly, one can earn the time value of
be claimed, mainly due to i) the NPV calculation money (e.g. interest). Secondly, as time passes
for each scenario is based on the same assump- until the decision has to be made, more informa-
tions as the static NPV (deterministic cash flows tion would be available such that a firmer founda-
and passive management), and ii) not capturing tion is achieved for making a decision.
the option of jump between different scenarios
during the course. Both arguments favour deferring a decision as
long as possible. The traditional NPV criterion
9.4 Decision Tree Analysis misses such an aspect. The real option methodol-
The decision tree analysis is one approach for ogy presumes the ability to postpone a decision
including the options revealed during the course and provides a way to quantify the value of
of the activity. Setting up an event tree does this deferring.
where the branches represent a cash flow outcome
attached with the probability for that event to take Using real option pricing, the value of an activ-
place. The decision nodes are added to the tree ity will always be greater than or equal to the
where the management may choose to change the value of a project using NPV. In case the identi-
activity in order to respond to (external) factors or fied flexibility by real option is unlikely to be
results of the activity. Such a method could for used, the difference in pricing would be small.
example be applied when analysing a complex This is also likely to happen when the NPV is
sequential investment scheme where several deci- very high or very low (much negative). The
sion points can be identified at discrete points of greatest difference is likely to be seen when
time, see Figure 27. Note the resemblance NPV is close to zero, hence referring to a project
between decision tree and decision map in Figure whose activation is questionable.
1. For a static NPV calculation, all uncertainty is
presumably captured by the discount rate, the The higher the level of uncertainty, the higher
decision tree adds more understanding of the the option value because flexibility allows for
options during the activitys course and their gains in the upside and minimise the downside
probabilities. By modelling the investment as a potential. Often the total discounted operation
decision tree, different actions in different scenar-
ios or nodes in the tree can be introduced to in-
corporate the value of flexibility.
leaf
The tree is solved backwards from the leaf nodes nodes
and the beginning of the tree by discounting the
relevant cash flows. The result is the value of the
activity with flexibility as modelled in the tree.
This approach is intuitive and simple to compre-

hend, although it may have a number of limita-
tions: i) the number of events may grow in a way
such that analysing the tree is intractable, ii) the
same discount rate is commonly used throughout
the tree. Frequently, the risk differs at different
nodes/branches inspiring for different discount
rates. Commonly the discount rate applied is for
an activity without flexibility. Figure 27 Decision tree
time
analysis
1) Operating cash flow = revenue all OAM, sales and other costs variable follow investments (line cards, etc.)
Investment option Variable Call option Option to defer: when a decision is postponed.
Present value of a projects S Stock price Combinations of options may also be identified.
operating assets to be acquired Two cases are:
Expenditure to acquire X Exercise price
the projects assets Compound option: an option is an option of
the value of another option. For example, one
Length of time the decision T Time to expiration
may be deferred may have the possibility to shrink or expand
an activity, but the amount (e.g. how many
Time value of money rf Risk-free rate of return parts) may be flexible. Another example is

Riskiness of project assets T Cumulative volatility sequential option plan (e.g. for a phased activ-
ity plan) where an option may be a potential
follow-on of another option.
Table 4 Relating investment cash flow1) is modelled by a lognormal stochas- Switching options: reflecting the possibility
option and stock call option tic variable. The uncertainty is modelled by a of switch between two operation modes. Two
volatility commonly used for financial call examples are to: i) exit and re-enter by a modi-
options. fied activity plan, ii) change from delivering
a certain result to another result. When the
Utilising real options, theory from financial switching cost is greater than zero, situations
analysis is applied to evaluate the value of physi- may occur when one is still carrying on with the
cal or real assets. It is claimed that real options activity even though the result would be better if
have been very useful in assisting top manage- changes were incorporated. This is due to the
ment of companies facing a significant manage- high cost of changing the activity plan.
rial flexibility and amount of uncertainty.
An important difference between real and finan-
There are two types of financial options: i) call cial options is that management can affect the
options, and ii) put options. The former is a right value of the underlying risky assets as the actual
to buy, while the latter is a right to sell (note: project is under the managements control. Third
these are options and not obligations). Referring parties control the financial options. This can be
to an activity and the corresponding managerial illustrated by mapping an investment question
flexibility, some relevant options are: into a call options as shown in Table 4.
Abandonment option (a put option): when the A number of observations can be made, see Fig-
complete activity is stopped. ure 28:
Option to contract (a put option): when part of An increase in the present value of the project
the activity may be stopped, outsourced, sold, will increase the NPV (without flexibility) and
etc. therefore the option value will also increase.
Figure 28 Estimating when
managerial flexibility is Option to expand (a call option): when the A higher investment cost will reduce NPV
greatest valued activity is enhanced. (without flexibility) and therefore reduce
option value.
A longer time to expiration allows learning

more about the uncertainty and therefore will
High
increase the option value.
Moderate High
flexibility flexibility An increase in the risk-free rate of return will
Room for value value
increase option value since it will increase the
managerial
flexibility (ability time value of money advantage in deferring
to respond) the investment cost.
Low
In an environment with managerial flexibility
Moderate
flexibility flexibility an increase in uncertainty will increase option
value value value.
Low In every case where the case value without flexi-

bility is close to break even, the flexibility will
have most relative value. Hence, the flexibility
Low High
value is greatest when:
Likelihood of receiving new information (uncertainty)
Great uncertainty about the future. Very likely to the binomial lattice approaches the time con-
to receive new information over time. tinuous solution of Black-Scholes as the number
of time-steps in the binomial tree gets larger.
Much room for managerial flexibility. Allows
management to respond appropriately to this The calculations are started from the leaf nodes
new information. of the tree (similar to Figure 27) where the ques-
tion of whether to exercise the option or not is a
NPV without flexibility near zero. If the activ- trivial one (being the end-points). The calcula-
ity is neither obviously good nor obviously tions are run backwards to the beginning of the
bad, flexibility to change course is more likely tree, choosing the best decision of whether to
to be used and is therefore more valuable. exercise or not at each branch/node of the tree.
At each branch/ node the market-replicating
In the financial world, the Black-Scholes for- portfolios or the risk-neutral approach can be
mula is well-applied for valuing real options. applied, both being equivalent to each other. The
This is an analytical solution to a differential former refers to finding an equivalent combina-
equation describing the value of a European call tion of risky and risk-free assets, while the latter
option. The underlying risky asset is assumed to refers to adjusting the cash flows in the NPV cal-
follow a geometric Brownian Motion with a culations.
Markov-Wiener stochastic process. When S is
the value, the percent change is given by: Naturally, one may raise the question of how to
find an equivalent portfolio when referring to an
S
= t + t activity never-seen-before for a certain com-
S
pany. On the one hand, it can be claimed that
where these are all models used as background when
making the actual decisions. On the other hand,
is a drift term or growth parameter that the static NPV of an activity could be used as a
increases at a factor of time steps t. best-guess on the unbiased estimate of the mar-
ket value of the activity. Considering the value
is the volatility parameter, growing at a rate including flexibility, one could consider this as
of the square root of time. placing one part of the money on the activity and
another part in bonds.
is simulated, usually following a normal distri-
bution with a mean of zero and variance of one. The binomial event tree describes the evolving
uncertainty of the underlying risky asset and the
The first term is the deterministic part and the stochastic process is based on a single volatility.
second term is the stochastic part. Monte Carlo simulation is commonly used to
produce the volatility for the return of activity.
A number of limitations on this formula are: This is however based on the assumption that the
there is only one source of uncertainty, no com- uncertainty is resolved continuously over time.
pound options are included, the underlying There are often different causes for uncertainties,
stochastic process is assumed to be known, the such as market and technology. These may well
variance of return is constant through time, the be resolved differently with time; an example
underlying risky asset follows a Geometric being that technological uncertainties decrease Figure 29 Potential
Brownian Motion with a Markov-Wiener with time while market uncertainties increase categorisation of
stochastic process. with time. vendors strength
One way of modelling real options is to apply

binomial lattices. Market-replicating portfolios
Breadth of vision
and risk-neutral approach could for instance be 4
used when calculating the values of the option
sets. The basic idea is to represent the evolving 3
uncertainty of the value of a risky asset by a Financial stability 2 External expression of vision
binominal tree. The risk itself may or may not
increase with time, but uncertainty increases with 1
time. This can be considered as an uncertainty 0
cone (hence growing as time passes). In the same
way as the Black-Scholes formula this relies on
a Geometric Brownian Motion following a Ability of change direction Internal expression of vision
Markov-Wiener process. As for any tree, time
is discretised, and hence it may become quite
demanding to compute every combination. As a
limit, the value of an option calculated according Ability to execute
Real Options Referring to Stocks
One model for stock behaviour is a so-called geometric Brownian motion. The time discrete
version is expressed as:
S(t)
= rf t + t
S(t)
where:
S(t) stock price at time t
S(t) change in stock price during a short time interval, t
rf drift rate (risk-free rate of return)

volatility
random number drawn from a standard normal distribution (~N(0,1)).
Now it can be shown that S(t) is log normally distributed with mean value given as

2
t = ln(S(0)) + rf t
2

and standard deviation of t = t.
The measure under consideration, here the total discounted value of the operating cash flow of a
project, T, is estimated at S(0) with the current information. Up to decision time, T, the manage-
ment receives information of relevant value and the measure value will therefore move up or
down with accumulated volatility t .
The project investment is only incurred if S(T) > X = X + (1 + rD)2V, where X is the initial
investment at time T and V is the NPV of the second phase of the project if the uncertain action
is not carried out. rD is the discount rate.
The NPV at time T is found as:

N P V = E max(S(T ) X, (1 + rD )2 V )
X
= (S(T ) X)p(S(T ))dS(T ) + (1 + rD )2 V p(S(T ))dS(T )
X 0
where
(ln(S(T ))T )2
1 2 2
p(S(T )) = e T
T 2S(T )
After some reordering and variable substitution (e.g. w = ln(S(t)) -> S(T) = ew, dS(t) = ewdw),
one gets
NPV = erfT (S N(d1)erfT X N(d2)) + V (1 N(d2))
where

2 S 2
ln S
+ rf + T ln X + rf 2 T
X 2
d1 = and d2 = = d1 T .
T T
N(.) is the cumulative standard normal distribution (~N(0,1)).
9.6 Stochastic Optimisation 9.7 Evaluating Vendors

Stochastic optimisation is an operational One element in the risk assessment is evaluating
research technique for handling uncertainty. This the situation of the current vendors. Both techni-
is applied in order to optimise simulation models cal and financial criteria have to be considered.
by defining a stochastic model and accompany-
ing decision variables. Iterations are carried out Having an assessment of the key vendors for an
between drawing input parameters from selected operator is also central. Several criteria could be
distributions and optimising the decision vari- derived, of which some are (from [Heav03]):
ables for each set of draws. In some sense this
is similar to solving the task of finding the best Breadth of vision
path in a decision tree. External expression of vision
Internal expression of vision
Ability to execute Netw F
Netw A
Ability to change direction
Financial stability Netw D IF <conditions> IF <conditions>
Netw B
This inspires for illustrative factor diagrams as IF <conditions> Netw E
IF <conditions>
shown in Figure 29. However, it is important to
be aware that the sequence of factors influences
the area of the diagram. Netw C Netw G
10 Overall Results
time
Regarding the overall objective of network strat-
egy studies, more than one target state may very
well be described within each study. A natural Figure 30 Illustration of network portfolio migration states with decision points
justification for this is that a future time frame is
commonly attached with uncertainty, leaving it
unlikely that every factor influencing the solu-
tions will be known. Hence, instead of looking
for the ultimate solution, a decision tree app- Production means
roach could be strived for. That is, paths of
migration are put together to choose which Netw A Netw B Netw C .. Figure 31 Relating
branch to follow if the tree is attached with a set products and production
of conditions. These conditions may be related Product z V V means for each time
to time and events. On a more abstract level period
Product x V V
every branching would actually be event-related,
although some indication of timing will be of Product w V
interest. This is schematically illustrated in Fig-
.
ure 30. The network naming has been chosen
simply for illustration. Moreover, more than two Year 200y
branches might appear from one state. Following
a line from a network without any merging or
splitting into branches indicates an upgrade of
network capabilities.
Naturally, any modification of an existing oper-
Following a branch in the tree both the conse- ating network implies certain risks. In particular
quences and the risks must be assessed. Conse- there are risk factors related to upgrading nodes
quences reflect costs, revenue side, product port- related to cost, functionality and capacity. This
folio, organisation and so on. Here, the total also applies to any work of integrating networks,
portfolio has to be taken into account, possibly and other steps taken in case an additional aggre-
leading to investments in a certain system result- gator function is to be introduced (as assumed
ing in overall reduced operational expenses be- for some of the cases/tracks). On the other hand,
cause a less efficient system could be phased out. there may also be risks involved when no
changes are carried out. This could come
It is essential to have captured the relevant risk because the system portfolio may gradually
factors when choosing the right implementation become less efficient than what competitors
track for core network migration. In other words manage and too slow to deploy new services.
it is not sufficient for an implementation track to
be technically feasible and having the smallest As mentioned in the beginning it is essential that
cost among the various implementation options, the network strategy elaborated is related to the
the option must also have a sufficiently low risk current situation and thereby possibly leading
to be considered for actual implementation. Risk to a change of the decisions to be made in the
factors can be grouped into technical, financial shorter term.
and business/strategic. The first refers to factors
such as whether the technical solutions are actu- The products to be supported have to be related
ally able to deliver the services as promised, to the network portfolio illustrated in Figure 30.
compatibilities between versions/vendors, etc. This may be thought of as a product produc-
Financial issues come from the income and the tion means matrix, as shown in Figure 31. Here
expense side, where none of these are fixed in the main product groups are given along one
the period. Business and strategic issues refer to axis and the production means on the other axis.
relations with regulatory bodies, organisation, Such a matrix must be defined for each period/
communication with customers/competitors, and year looked at. In a period a product may be pro-
so forth. vided in a number of ways, partly competing and
partly complementing. Doing the exercise of
Services Customers
ISDN/PSTN
Voice Residential to
Dial - up IP large enterprises CU LE TE TE LE Cu
xDSL ATM IP CU - Concentration Unit

Voice Residential, LE - Local Exchange
IP SoHo, xDSL Sw R ISPs TE - Transit Exchange
SME Sw - Switch
R - Router
HSDSL WDM HSDSL
Voice SME xDSL xDSL
xx WDM
LL
SDH
LL Large enterprise SDH
SDH
Figure 32 Example of relating relating products to production means will reveal 11 Concluding Remarks
services/customer segments to further areas of potential gains. A network strategy is needed for every sound
production means (Note: not network operation. The strategy must also be
all relevant systems are To some extent products to be supported have related to decisions and actions in near-time
included, e.g. the mobile been described earlier. A few basic trends have hence, the strategy must be made operational.
systems) also been outlined earlier. Regarding the demand A main goal of having a strategy is to be pre-
patterns, most sources indicate a growth in most pared for chances that can be revealed during
issues, leading to rather questionable aggregated the course of events. That is, the strategy is
demands. However, trends showing main growth likely to assist when detecting business opportu-
areas within mobile, IP and xDLS are repeatedly nities. The total network portfolio must be
observed. included, also seeing synergies between the dif-
ferent systems in the shorter and longer terms.
Defining how to efficiently support the different
products belongs to the network portfolio man- In elaborating the strategy a number of methods
agement. A basic challenge is to relate the ser- can be applied, including scenarios, cost/benefit
vices and market segments. An example of how calculations and risk assessment. A description
this may be carried out is depicted in Figure 32. of various aspects is outlined in this article.
Acknowledgement
Although the material in this text is the sole
responsibility of the author, fruitful discussions
with all colleagues involved in Telenor strategy
work in later years are appreciated.
References
[Cari01] Canadian Treasury Board. Integrated
Risk Management Framework. 2001. Online:
http://www.tbs-sct.gc.ca
[Heav03] Heavy Reading: Incumbents wireline

strategies, 1 (3), 2003.
Portfolio Evaluation Under Uncertainty
RALPH LORENTZEN
1 Introduction Investors have expectations about asset

Principles for evaluation of the risk associated returns that are normally distributed.
with a portfolio of investment projects within
a company are based on models that describe Investors may borrow or lend unlimited
potential changes in the factors that influence the amounts of a risk-free asset.
profit of the individual projects and the interac-
tion between these. Many different models are All projects (and parts of projects) can be
described in the literature. Some approaches are bought/sold in a perfect market.
simple and may border towards the trivial, whilst
Ralph Lorentzen (67) is daily others are more involved and rely on advanced Even stronger are the conditions that must be
leader of the company
stochastic models. During the last years it has satisfied if we want to use a risk-adjusted dis-
Lorentzen LP Service which
does consulting work within become popular to apply principles and theory count rate for projects with a lifetime of more
mathematical programming. He from the mathematics of finance (see e.g. [1] and than one period. Then it must be assumed that
retired July 1, 2003 from his job
[2]). Analogy is drawn between investment in the cash flow in one period is a deterministic
as senior scientist in Telenor
R&D. Ralph Lorentzen gradu- securities on the one hand and investment in function of the cash flow in the previous period
ated in mathematics at the Uni- projects on the other. Both situations are charac- plus a random variable that is independent of
versity of Oslo, where he
terised by the investment of an amount of money what has happened in all previous periods.
worked for some time as assis-
tant professor giving lectures in at one or several points in time with the hope of
statistics and operations receiving larger sums of money at one or more 2.2 Problems with the Use of a
research. He worked as distribu-
tion planner in Norske Esso, as
future points in time. Such analogy has its strong Risk-adjusted Discount Rate
principal scientist at Shape and weak sides. Examples
Technical Centre, as systems In many situations it is not reasonable to assume
engineer in IBM, and as chief
In this paper we describe an approach based on that the conditions mentioned in the previous
consultant in Control Data Nor-
way, before joining Telenor R&D what we call Risk-Adjusted Expected Net Pre- section are satisfied. We shall show by three
in 1985. sent Value (RAENPV) and contrast it to the more project examples what happens when these
ralph.lorentzen@tiscali.no common Net Present Value (NPV) with Risk conditions are not fulfilled. The examples are
Adjusted Discount Rate. We also propose a way stylised to visualise the effect of risk-adjusting
of describing the uncertainty associated with the the discount rate. However, their basic features
cash flow of each individual candidate project in are not unrealistic.
a portfolio and how it interacts with each of the
other candidate projects. Our guiding principle The cost of capital is assumed to be 12 %, and
has been to establish an approach, which on the we assume that the relevant risk-adjusted dis-
one hand is not too narrow but on the other hand count rate has been found (using for example
avoids relying on complicated models that can- CAPM type arguments) to be 13 % for all three
not be justified because of lack of knowledge projects. All amounts are in millions of dollars.
about the future.
Example 1:
2 Evaluation of a Single We invest 98 in year 0 in building a VDSL access
Project network. There is uncertainty as to whether the
revenue from subscribers will commence in year
2.1 Risk-adjusted Discount Rate 1 or 2. With probability 1/2 we receive a net
We shall first briefly comment in general terms income of 13 in year 1. In the years 2, 3, we
on an approach which is often used where a receive with probability 1 a yearly net income of
project is rated according to its NPV calculated 13. The uncertainty is thus whether the income
using a risk-adjusted discount rate that reflects stream starts in year 1 or 2 (see Figure 1 where
the uncertainties in the cash flow. The approach the income that occurs with probability 1/2 is
is usually based on variants of the Capital Asset hatched).
Pricing Model (CAPM). A thorough description
of CAPM may be found in [1]. A feature of this Situation A Net income 13 in year 1:
approach is that high uncertainty in the cash The NPV becomes (13/1.12) / (1 1/1.12) 98
flow leads to a high risk-adjusted discount rate. = 10.3.
The approach is not without problems, even for
projects with a lifetime of one period only. Situation B Net income 0 in year 1:
We mention some of the conditions that must The NPV becomes (13/1.122) / (1 1/1.12) 98
be satisfied: = 1.3.
Expected NPV: 1/2 10.3 + 1/2 (1.3) = 4.5 With probability 1/2 we receive a net income of 13
in year 1. In year 2 we sell the network to another
Expected NPV with risk-adjusted discount rate: company for 13 / (1 1/1.12) = 121.3 (which is
1/2 (13/1.13) / (1 1/1.13) + 1/2 (13/1.132) / the NPV at capital cost of the incomes received
(1 1/1.13) 98 = 3.8. with certainty in Example 1 from year 2 and
onwards; see Figure 3). We see that Example 3 is
The method thus indicates that the project should financially equivalent to Example 1. We have only
be rejected. replaced the income in year 2 with the NPV at
capital cost of the incomes we receive with cer-
Figure 1 Cash flow in tainty from year 2 and onwards.
Example 1 13 13 13
Situation A Net income 13 in year 1:
1 2 3 year The NPV becomes 13/1.12 + 121.3/1.122 98
= 10.3.
98
Situation B Net income 0 in year 1:

The NPV becomes 121.3/1.122 98 = 1.3.
Example 2: Expected NPV:

We invest 63 in building an ADSL access network 1/2 10.3 + 1/2 (1.3) = 4.5.
in year 0. There is uncertainty as to whether the
ADSL technology will generate revenue for 4 or 5 Expected NPV with risk-adjusted discount rate:
years. With probability 1 we receive a net income 1/2 13/1.13 + 121.3/1.132 98 = 2.8.
of 20 in the years 1, ..., 4. With probability 1/2 we
receive a net income of 20 in year 5. The uncer- The method thus indicates that Example 3 should
tainty is thus whether we receive an income in be accepted.
year 5 (see Figure 2 where again the income that
occurs with probability 1/2 is hatched).
Situation A Net income 20 in year 5:

The NPV is (20/1.12) (1 (1 1/1.12)5 / (1 121.3
1/1.12) 63 = 9.1.
13
Situation B Net income 0 in year 5: 1 2

year
The NPV is (20/1.12) (1 (1 1/1.12)4 / (1
1/1.12) 63 = 2.3. 98
Expected NPV:
1/2 9.1 + 1/2 (2.3) = 3.4
Figure 3 Cash flow in Example 3
Expected NPV with risk-adjusted discount rate:
1/2 (20/1.13) (1 (1 1/1.13)5 / (1 1/1.13)
+ 1/2 (20/1.13) (1 (1 1/1.13)4 / (1 1/1.13) These results are not reasonable. We see that the
63 = 1.9. projects in Example 1 and Example 3 are finan-
cially equivalent. The criterion for whether they
The method thus indicates that Example 2 should should be accepted or rejected should therefore
be accepted. be the same. Either should both be accepted or
both be rejected. However, the expected NPV
with risk-adjusted discount rate is negative for
Figure 2 Cash flow in Example 1 and positive for Example 3. The
20 20 20 20 20
Example 2 NPV method with risk-adjusted discount rate
thus implies that the project in Example 1 should
1 2 3 4 5 year
be rejected whilst the project in Example 3
63 should be accepted. Furthermore, when the dis-
count rate is set to capital cost, the project in
Example 1 gives a higher NPV than the project
in Example 2 for all statistical outcomes. Com-
mon sense thus implies that if the project in
Example 3: Example 2 is accepted, then the project in Exam-
We invest 98 in a VDSL access network in year 0. ple 1 should be accepted. The NPV method with
There is uncertainty as to whether the revenue risk-adjusted discount rate implies, however,
from subscribers will commence in year 1 or 2. that the project in Example 1 should be rejected

whilst the project in Example 2 should be
Prob. density
accepted.
The examples demonstrate that any method

based on risk-adjusting the discount rate is
unable to differentiate correctly between differ-
ent risk structures. By risk-adjusting the discount
rate one tries to satisfy two masters. The discount q
rate should reflect VaRq 0 NPV
Capital cost (i.e. the rate of return which can

be achieved by alternative application of the
money which are tied up/released in the pro- Therefore we choose to measure risk by using Figure 4 Value-at-Risk
ject over time). This has nothing to do with the well-known concept Value-at-Risk at q
the risk of the project under consideration; (VaRq). VaRq is defined as the q-point in the
probability distribution of the NPV (see Figure
The risk associated with the project 4).
These two requirements are fundamentally dif- q is usually chosen to be 1 % or 5 %. The sug-
ferent, and it is difficult to satisfy both by adjust- gested RAENPV then becomes
ing the discount rate. The main reason to look
for an alternative approach is to be able to differ- Expected NPV (VaRq) (2.3.2)
entiate between capital cost and risk.
where the NPV is calculated using the capital
2.3 Risk-adjusted Expected Net cost w as discount rate and where the parameter
Present Value reflects how the company (or its shareholders)
We shall now describe our suggested approach. weigh expected profit versus risk. We shall see
The expected NPV is calculated using capital later that the term (VaRq) can be inter-
cost w as discount rate. The capital cost should preted as the insurance premium the company is
reflect the average expected rate of return for willing to pay in order to be guaranteed against a
alternative projects in the company and has negative NPV.
nothing to do with the risk associated with the
project under evaluation. We shall now show how we determine . For a
perfect security market which contains a risk-
Then an amount is subtracted from the expected free asset it is established knowledge that the
NPV. This amount adjusts the Expected NPV substitution rate (price of risk) between ex-
because of the risk of the project. The suggested pected return and risk (measured by the standard
measure of utility of a project is deviation of the return) in equilibrium is the
same for all investors independent of their will-
Expected NPV risk (2.3.1) ingness to take risk (see e.g. [1] chapter 6). The
value of can be obtained from stock exchange
where is a parameter to be determined. In data and is not specific for a particular industrial
financial theory the risk is traditionally ex- sector. So for any investor an increment in
pressed through the standard deviation of the risk needs to be compensated by an increment of
NPV. This is a reasonable measure for portfolios / in expected return. This is shown in [1] for
of securities. Such portfolios consist of a number investments in a portfolio of securities where the
of different securities, and the return is the sum portfolio is sold after one period, but it seems
of the returns of the individual securities. The reasonable to use the same trade-off for multi-
central limit theorem in probability theory indi- period projects. It is reasonable to measure the
cates that this sum is approximately normally utility of an investment in a portfolio of securi-
distributed. Risk should reflect the down side of ties by E where E is expected return and
the distribution of the NPV. However, the nor- is the standard deviation of the return.
mal distribution is symmetric, so the standard
deviation expresses the down side as well as the Figure 5 shows the classical picture where the
up side of the distribution. For individual invest- dots represent securities in a diagram with the
ment projects in a company the assumption that standard deviation of the return along the horizon-
the NPV has a symmetric probability distribu- tal axis and the expected return along the vertical
tion is obviously unreasonable. The NPV will axis. The lower curve the opportunity set border
often have a skewed distribution. As the stan- curve limits all possible portfolios of securities.
dard deviation also reflects the up side of the The straight line which starts at the risk-free asset
distribution, we may have a large standard de- and is tangent to the opportunity set border curve
viation even if the risk of the project is small. is the so-called capital budget line.

a discount rate r and the expected return by the

expected NPV.
1
The rate of increase in Figure 5 has here been
transformed to the rate of increase in Figure 6.
It can be shown (see the Appendix) that is
independent of which discount rate r we
f apply. We note further that the indifference
curve for any investor has rate of increase equal
to at the point where it touches the trans-
formed capital budget line. That is, represents
the optimal weighting between expected NPV
and Value-at-Risk. As mentioned earlier, the
Figure 5 The capital budget line rate of increase can be found from capital mar-
ket data. The formula for calculating from
is derived in the Appendix and is
N
= (2.3.3)
nq
1
where nq is the (1 q)-point in the standardised
normal distribution.
Since the market portfolio can be regarded as

a yardstick, it is of interest to calculate the
Nf RAENPV for the market portfolio. In the Ap-
pendix it is shown that it can be expressed as
ir
(2.3.4)
(1 + r)(1 /nq )
-VaRq
where i is the internal rate of return for the mar-
ket portfolio.
Figure 6 Transformed capital The point where the capital budget line touches
budget line the opportunity set border curve represents the A project should then be accepted if the
market portfolio and is marked by a small square RAENPV is greater than this. That will be in the
in Figure 5. The upper curve in Figure 5 repre- interest of the shareholders. The project can then
sents an indifference curve for an investor. The be represented by a point in Figure 6 that lies
point where this curve touches the capital budget above the transformed capital market line. If the
line, and which is marked by a small circle, rep- project is considered as a new investment possi-
resents the combination of the market portfolio bility in the capital market, a larger opportunity
and the risk-free asset which is considered to be set, a new transformed capital market line, and
optimal for this particular investor. hence an improved market portfolio would arise.
In Figure 7 the new opportunity set, capital mar-
Figure 7 Project with higher Figure 5 can be transformed to Figure 6 where ket line, and market portfolio are indicated by
risk-adjusted NPV than the we have replaced the standard deviation of the broken lines. The new market portfolio would
market portfolio return by Value-at-Risk for the NPV based on necessarily include the project.
An investment possibility represented by a point

underneath the transformed capital market line
might still improve the revised market portfolio
N New market portfolio and thus be a profitable investment. This would,
however, be rather difficult to ascertain in gen-
eral. We therefore propose to be conservative
New project and leave to the planner the decision whether
a project with a negative RAENPV should be
accepted or not.
Nf 2.4 Analysis of the Examples Using

Risk-adjusted Net Present Value
We shall now apply our criterion on the three
examples described earlier where we assume
that a has been estimated to be 0.2 and that the
-VaRq RAENPV of the market portfolio is very small.

In all three examples VaRq at capital cost is the Based on the resulting probability distribution of
same whether q is 1 % or 5 % since the probabil- the NPV N one can calculate project characteris-
ity distributions of the NPV are different from tics like:
zero for two values only, and for those values
the probabilities are equal to 50 %. Expected NPV (EN)
Value at risk at q (VaRq(N))
Example 1: Standard deviation ((N))
The expected NPV and VaRq with discount rate Risk-adjusted ENPV (EN (VaRp(N)))
equal to capital cost are 4.5 and 1.3 respective-
ly. The RAENPV becomes 4.5 0.2 1.3 = 4.24. Example 4:
The project should therefore be accepted. We assume a single scenario. Based on a de-
tailed analysis of the elements contributing to the
Example 2: cash flow the pessimistic and the optimistic esti-
The expected NPV and VaRq with discount rate mates of the NPV for this scenario have been
equal to capital cost are 3.4 and 2.3 respective- evaluated to be 400 and 300 respectively, and
ly. The RAENPV becomes 3.4 0.2 2.3 = 2.94. the most probable NPV is evaluated to be 100.
The project should therefore be accepted. We The NPV is assumed to lie in the interval (-200, 0)
see that the RAENPV is less than for the project with probability 1/2. (All figures are in mill. USD.)
in Example 1, which is reasonable since the pro-
ject in Example 1 has a higher NPV at capital cost The Beta distribution fit gives an expected NPV
than the project in Example 2 under both statisti- equal to 82.5 with standard deviation 134.4. The
cal outcomes. probability that the NPV is negative becomes 0.72
and VaR5% becomes 296. If = .2, the
Example 3: RAENPV becomes 82.5 .2 296 = 141.7.
The expected NPV and VaRq with discount rate
equal to capital cost are 4.5 and 1.3 respec- 3 Evaluation of Project
tively. The RAENPV becomes 4.5 0.2 1.3 = Portfolios
4.24. The project should therefore be accepted.
The RAENPV is the same as for the project in 3.1 Project Portfolio Alternatives
Example 1, which is logical since the two projects We shall consider the situation where we have
are financially equivalent. a series of mutually exclusive portfolio alterna-
tives. If we have n projects, there are in principle
2.5 Establishing the necessary Data up to 2n portfolio alternatives. In practice the
for Evaluation of a Single Project number of alternatives will be much lower. Typ-
The cash flow of a project is considered as a ically, some projects can be started only if some
stochastic process. For every realisation of the other projects are also started. Similarly there
process the NPV with capital cost as the dis- may be projects that cannot be part of the same
count rate can be calculated. This NPV is thus a portfolio. The planner must specify such restric-
random variable, and we propose the following tions, and a planning tool must facilitate the
procedure for estimating its probability distribu- specification of such restrictions.
tion.
3.2 Scenarios
1 Establish a finite set of alternative future sce- We can specify scenarios for project portfolios
narios and assign a subjective probability of in two alternative ways:
occurrence for each scenario.
We can specify scenario alternatives across all
2 For each scenario establish a pessimistic, an projects.
optimistic and a most probable cash flow for
the project and calculate the NPV for each of We can specify scenario alternatives for each
those three realisations. individual project.
3 For each scenario specify an interval around The two ways are in principle equivalent. If we
the most probable realisation in which the for each project p specify the mutually exclusive
NPV supposedly will lie with probability 1/2. scenarios Sp1 , K, Spk(p), we can define k(p)
scenarios of the form (S1i1 S2i2 L Snin )
4 Fit a Beta distribution to the NPV for each which run across projects. We shall in the sequel
scenario. Based on the mixture of these Beta assume that the scenarios are defined across the
distributions using the scenario probabilities, projects.
calculate the RAENPV.

3.3 Evaluation of a Portfolio of and the market portfolio. Unfortunately, the nec-
Projects essary conditions are in general unrealistic, even
The evaluation of a portfolio is done by a for portfolios of securities. For project portfolios
straightforward generalisation of the approach these conditions cannot be justified at all. How-
for a single project. Based on the probability dis- ever, in this case the number of covariances to
tributions of the NPVs of the individual projects be estimated is considerably lower. In a security
and the correlations between them we can calcu- market the investors can choose between hun-
late the same characteristics as for a single pro- dreds of securities so that there will be tens of
ject such as expected NPV, value-at-risk, stan- thousands of covariances. A portfolio of invest-
dard deviation, and RAENPV. ment projects in a company will normally con-
sist of between 5 and 10 projects so that the
In general one proceeds as follows: number of covariances will be less than 50. We
shall briefly discuss two approaches to estima-
Establish a set of mutually exclusive scenario tion of the covariance between two NPVs, name-
alternatives. Which of these scenarios that ly direct estimation and the use of independent
occurs should have a significant influence on explanatory variables.
the NPV of the portfolio. Each scenario is
assigned a subjective probability of occur- However, we should make note of the fact that
rence. the RAENPV measure is approximately additive
in the sense that the RAENPV of a union of two
For each scenario establish a pessimistic, an projects is approximately equal to the sum of the
optimistic and a most probable realization of RAENPV of the individual projects. The ex-
each project in the portfolio and calculate the pected NPV is obviously additive. The same
NPV for each of these realizations. is true for VaRq when q = 0 %, and the two pro-
jects are not totally negatively correlated. So for
For each scenario and each project in the port- q = 1 %, the approximation may not be too mis-
folio specify an interval around the most prob- leading. For preliminary analyses we may thus
able realisation where it is assumed that the skip the correlation calculations altogether.
NPV will lie with probability 1/2.
3.4.2 Direct Estimation
For each scenario specify a matrix of correla- Here the planner estimates the covariances, or
tion coefficients between the NPVs of the rather the correlation coefficients, based on
individual projects in the portfolio. experience and sound judgement. In order to
facilitate this a planning tool should enable the
For each scenario calculate the expectation planner to choose from a selection of preset cor-
and the variance of the NPV of the portfolio, relation coefficients as suggested in Table 1.
and approximate the distribution of the NPV
by a suitable Beta distribution. Example 5
Suppose that we have three projects A, B and C
Based on the mix of these Beta distributions with NPVs NA, NB and NC. We estimate the cor-
over the scenarios, calculate the wanted char- relation coefficients for all pairs of NPVs as given
acteristics of the portfolios NPV. in Table 2.
3.4 Establishing Covariances Not every table set up in this manner gives rise
to valid correlation matrices. A necessary, but
3.4.1 General not sufficient, condition is that the matrix is pos-
Establishing correlation coefficients (or equiva- itive semidefinite. Based on the distributions of
lently, covariances) between the NPVs of all the NPVs it is also possible to establish lower
pairs of projects in a portfolio is not trivial. and upper bounds on the individual correlation
CAPM (see [1]) has an elegant solution to this coefficients. This, however, is computationally
for portfolios of securities which are registered demanding and probably not worth the effort.
on a stock exchange. Here it is only necessary These bounds will very seldom lie in the interval
Table 1 Correlation codes to establish the covariance between the portfolio [3/4,3/4], and total correlations (T) will only
Description Total Large Medium Small None Small Medium Large Total
negative negative negative negative positive positive positive positive
Code T L M S N S M L T
Correlation 1 3/4 1/2 1/4 0 1/4 1/2 3/4 1

coefficient

be used when the NPVs are linearly dependent. NB NC Table 2 Specification of
Then the bounds will be 1 and +1 and thus be correlations
NA M L
without value.
NB S
If the planner has set up a correlation matrix
which is not positive semidefinite, the planning
tool should be able to suggest a modified corre-
lation matrix which is positive semidefinite. Using (3.4.6) and (3.4.7) we can find an expres-
sion for cov(N1, N2). Analogously we may find
3.4.3 Use of Independent Explanatory formulas for var Np for relevant portfolios p.
Variables
We assume that we can write the NPVs N1 and Example 6
N2 for two investment projects 1 and 2 as func- N1 = 10V1 + V2
tions of F independent random variables V1, V2, N2 = 2V1 3V2
..., VF with known distributions:
where V1 is Beta distributed over (0,1) with
Np = Np(V1, V2, ..., VF) for p = 1,2. (3.4.1) parameters n = 3 and n = 6, whilst V2 is Beta
1 1
distributed over (0,1) with parameters n = 4 and
2
The V-s are explanatory variables which deter- n = 2, and where V1 and V2 are independent.
2
mine e.g. equipment costs, total markets, and
market shares. Furthermore, we assume that the Then
functions Np can be approximated by generalised
6 2 2 1
polynomials: EV1 = = , EV2 = =
3+6 3 2+4 3
Np (V1 , V2 , K, VF ) 36 1
varV1 = = ,
p (3 + 6)2 (3 + 6 + 1) 45
i1 KiF V1i1 KVFiF (3.4.2)
42 2
i1 ,K,iF varV2 = = ,
(4 + 2)2 (4 + 2 + 1) 63
where i1, ..., iF may take a finite number of inte-
ger values which are not necessarily positive.
We are then able to calculate the covariance cov(N1 , N2 ) = cov(10V1 + V2 , 2V1 3V2 )
between N1 and N2: 22
= 20varV1 3varV2 =
63
cov(N1, N2) = E(N1N2) EN1EN2 (3.4.3)
EN1 = 10EV1 + EV2 = 7,
where
1
E(N1 N2 ) = EN2 = 2EV1 3EV2 =
3

i11 KiF j21 KjF EV1i1 +j1 KEVFiF +jF (3.4.4)
142
i1 ,K,iF varN1 = 100varV1 + varV2 = ,
j1 ,K,jF 63
118
and varN2 = 4varV1 + 9varV2 = .
315
EN1 EN2 =

i11 KiF j21 KjF EV1i1 EV1j1 KEVFiF EVFjF
We may approximate the distributions of N1 and
i1 ,K,iF
j1 ,K,jF N2 by Beta distributions. The intervals are deter-
(3.4.5)
mined by:
such that
aN = 0, bN = 11, aN = 3, bN = 2.
cov(N1 , N2 ) = i11 KiF j21 KjF 1 1 2 2
i1 ,K,iF
j1 ,K,jF In order to find the parameters we normalize N1
(EV1i1 +j1 KEVFiF +jF EV1i1 EV1j1 KEVFiF EVFjF ) and N2:
(3.4.6) N1 7 142
N1 = , EN1 = , varN1 = ,
11 11 7623
We assume furthermore that Vf is normalized to N2 + 3 2 118
N2 = , EN2 = , varN2 = .
lie between 0 and 1 and is Beta distributed with 5 3 7875
parameters f and f. Then
(f + f )(f + i)
EVfi = (3.4.7) This gives
(f )(f + f + i)

(7/11)(1 7/11)2 which situation that will occur. Let n be the
N 1 = (1 7/11) = 4.15,
142/7623 RAENPV of the market portfolio. The overall
RAENPV n for the project becomes
(7/11)2 (1 7/11)
N 1 = 7/11 = 7.27
142/7623 n= m n m . (4.1.2)
m
(2/3)(1 2/3)2 4.2 Portfolios

N 2 = (1 2/3) = 4.61, We now proceed to the analysis of portfolios of
118/7875
projects where each project has a starting date
(2/3)2 (1 2/3) some time in the future and where information
N 2 = 1/3 = 9.55.
118/7875 received before the starting date may indicate
that it is not profitable to start the project.
4 Extension to Options
Again we shall limit ourselves to describing a
4.1 A Single Project discrete options model where the additional
We shall first focus on the analysis of a single information that we receive before the possible
project. We consider a Project A that we may start of a project in the portfolio is an adjustment
start at some future time T. Between current time of the probability of occurrence of the individual
T0 and time T we may receive information which scenarios. We enumerate all the projects accord-
can influence our decision whether to launch ing to increasing starting time: P1, P2, ..., and we
Project A or not. If, for example, this project want to decide whether to start project P1. We
competes with an alternative Project B with shall approach this problem using dynamic pro-
launching time T0, we must consider the value of gramming. We let Dp denote a partial portfolio
the option of not starting Project A that we will of projects amongst P1, ..., Pp, and we denote by
have at time T. If the information we receive m a situation at the starting time for project
during the time interval [T0,T] implies that the Pp+1. We assume that we for all such partial
probability that the project will be profitable is portfolios Dp and all m know the partial port-
unacceptably low we will at time T exercise the folio Dp(m) amongst the projects Pp+1, Pp+2 ...
option of not starting the project. This should in which together with Dp yield the portfolio
general give a higher utility of the project than Dp Dp(m) with highest RAENPV.
what we would get if we base the analysis
merely on estimates of the cash flow of the pro- We now assume that we have decided on a par-
ject as seen at time T0. What is required, how- tial portfolio Dp-1 amongst the projects P1, ...,
ever, is the probability distribution of the addi- Pp-1, that we are in situation m, and that we face
tional information as seen at time T0. the decision whether to start project Pp. We now
consider two cases1):
Here we shall limit ourselves to describing a dis-
crete options model where the additional infor- (i) We start project Pp. Then we form the partial
mation that we receive in [T0,T] relevant for the portfolio Dp = Dp-1 Pp. The probability of
project in question is an adjustment of the proba- being in situation m at the starting time for
bility of occurrence of the individual scenarios. project Pp+1 is assumed to be known and

We thus assume that we do not receive any addi- equal to mmp
. The portfolio Dp Dp(m)
tional information that is sufficiently significant with the highest RAENPV n(Dp, m) in situ-
to warrant a change in the Beta distributions for ation m is known, and the RAENPV
the individual scenarios. This may seem to be a becomes
severe limitation. It is, however, mitigated some-
mm
p n(Dp , m ) (4.2.1)
what by the possibility of splitting a scenario
m
where we would like to change the NPV distri-
bution into two or more subscenarios. (ii) We do not start project Pp. The portfolio
Dp-1 Dp(m) with highest RAENPV
We define m to be the probability at time T0 n(Dp-1, m) in situation m is known, and the
that we at time T are in situation m amongst M RAENPV becomes
alternative situations. Situation m is character-
ized by a probability distribution over the differ- mm
p n(Dp1 , m ) (4.2.2)
ent scenarios. The RAENPV for scenario i is ni. m
All ni are assumed to be known. So the uncer-

tainty at time T is reduced to the question as to
1) Note that in many situations we cannot choose whether we should start project P or not. It may happen that P either is not compatible with D
p p p-1
or that Pp is a necessary consequence of Dp-1. This contributes to a reduction of the number of possible portfolios and thus the time required for the
necessary calculations.

If (4.2.1) is larger than (4.2.2), we choose to S5 Figure 8 Situation transitions
start project Pp. If (4.2.1) is less than or equal
to (4.2.2) we choose not to start project Pp. For S4
every partial portfolio Dp-1 of the projects P1, ...,
Pp-1, and every situation m at the start time for S3
project Pp, we thus know the partial portfolio
S2
Dp-1(m) of the projects Pp, Pp+1 ..., which
together with Dp-1 yield the portfolio Dp-1 S1
Dp-1(m) with highest RAENPV n(Dp-1, m).
t1 t1 t1
This then forms the basis for a dynamic pro-
gramming algorithm for finding the project port-
folio with the highest RAENPV. The input we The transition probabilities are given in Table 3.
need in addition to what we have discussed ear-

lier is the Markov chain matrices p = (mm p )

where mm p
is the probability of being in situa- S2 S3 S4 S5 Table 3 Transition
tion m at the starting time of project Pp+1 given probabilities
that we are in situation m at the starting time of S1 1/2 1/2
project Pp. S2 2/3 1/3
S3 1/3 2/3
To get going we start with the project Pn with
the latest starting time and then work our way
backwards in time: The RANPVs of the three projects in the relevant
situations are given in Table 4.
For every partial portfolio Dn-1 of projects from
P1, ..., Pn-1, and every situation m at the starting
time for project Pn we consider two cases: P1 P2 P3 Table 4 RANPVs for the
individual projects
(i) We start project Pn, form the portfolio Dn = S1 1
Dn-1 Pn and calculate the RAENPV n(Dn). S2 5/4
S3 1/4
(ii) We do not start project Pn, put Dn = Dn-1 and
calculate the RAENPV n(Dn). S4 5/2
S5 1/2
If the RAENPV we calculated in (i) is greater
than the RAENPV we calculated in (ii), we
choose to start project Pn. If the RAENPV we We assume further that P3 cannot be started if P2
calculated in (i) is less than or equal to the is not started. Thus the possible portfolios are ,
RAENPV we calculated in (ii), we choose not to P1, P2, P1 P2, P2 P3, P1 P2 P3. We
start project Pn. For every partial portfolio Dn-1 want to know whether we shall start project P1 at
from the projects P1, ..., Pn-1, and every situation time t1. We also make use of the approximate
m at the starting time for project Pn we thus additivity of RAENPV mentioned in 3.4.1. We
know whether it is profitable or not to start pro- shall fill in Table 5 where the entries are to be
ject Pn and can therefore for every situation m interpreted as RAENPVs disregarding contribu-
establish the extension of the partial portfolio tions from projects started earlier in time, and
Dn-1 to a complete portfolio with the highest optimal decisions are made at later points in time.
RAENPV. The set of relevant situations may
of course vary from one project starting point From Table 4 we see that we at time t3 start pro-
to the next. ject P3 in situation S4 whilst we do not start pro-
ject P3 in situation S5. Thus the rightmost column
Example 7 in Table 5 is easily established.
We assume that we have five situations S1, ...,
S5, and three projects P1, P2 and P3 with startup If we do not start P2, the tail RAENPV becomes 0.
times t1 < t2 < t3. At t1 we are in situation S1.
We assume that the transitions between the situ- If we start P2 in S2, the expected tail RAENPV
ations are as depicted in Figure 8. becomes 5/4 + 2/3 5/2 = 5/12.
If we start P2 in S3, the expected tail RAENPV

becomes 1/4 + 1/3 5/2 = 13/12.
If we do not start P1, the expected RAENPV

becomes 1/2 5/12 + 1/2 13/12 = 3/4.

If we start P1, the expected RAENPV becomes We thus have
1 + 1/2 5/12 + 1/2 13/12 = 7/4.
EA = (1 + r)ENrM + r. (A4)
Thus we conclude that project P1 should be
started. Let aq be the q-point in the distribution of A.
Then

Ar aq r
Table 5 RANPVs for the tail P1 P2 P3 q = P(A aq ) = P
1+r 1+r
portfolios
a q r
S1 7/4 = P NrM (A5)
1+r
S2 5/12
such that
S3 13/12
aq r
S4 5/2 VaRq (NrM ) = . (A6)
1+r
S5 0
We can estimate aq based on the empirical distri-
bution of A by using the q-point in this distribu-
5 Conclusion tion. Since the return of the market portfolio is
We have argued that it makes more sense to a weighted sum of the returns of the individual
evaluate portfolios based on a utility measure securities that constitute the portfolio we may
that balances expected net present value at capi- alternatively permit ourselves to assume that A
tal cost against value-at-risk, rather than to use is normally distributed N(EA, A). Then we can
a risk-adjusted discount rate. The concepts used establish the following relation
are fairly simple, and the approach can readily
be implemented in a spreadsheet tool that calcu- q = P(A aq)
lates the total utility of all relevant portfolio can-
A EA aq EA
didates and ranks them according to decreasing =P (A7)
A A
total utility. We believe that the output of such a
tool will be useful for the planners responsible which gives
for portfolio analysis and investment recommen- aq EA
dations. = nq (A8)
A
Appendix: where nq is the q-point in the standardized nor-

Risk-Adjusted Expected NPV mal distribution. (A8) can be written as
Detailed Analysis
aq = EA nqA. (A9)
A1 The Market Portfolio
We consider a suitable market portfolio M and This gives further
operate with a fixed period length which we for EA nq A r
simplicity set equal to one year. At the begin- VaRq (NrM ) =
1+r
ning of the year we buy for one dollar worth of
n q A
the market portfolio and sell it at the end of the = ENrM (A10)
1+r
year for W dollars. The rate of return A is then
given by which gives the following expression for A:
W 1 1+r
A= = W 1. (A1) A = ENrM VaRq (NrM ) (A11)
1 nq
The NPV NrM with discount factor r then The capital market line is given by:
becomes
EA = f + A (A12)
W Ar
NrM = 1= . (A2)
1+r 1+r
where f is the risk free interest rate. reflects
with expected value how the market portfolio balances expected
return against risk2).
EA r
ENrM = . (A3)
1+r
2) A new investment possibility that lies above the capital market line will be included in the market portfolio and thus alter it. The investment will there-
fore be considered profitable in a perfect capital market. (An investment possibility below the capital market line could also be profitably included in
the market portfolio, but this is considerably more difficult to ascertain.)

From (A9) we see that References
1 Copeland, T E, Weston, J F. Financial
EA = aq + nqA (A13) Theory and Corporate Policy. Boston, Mass,
Addison Wesley, 1992.
We assume that the market portfolio is not risk
free, in other words that aq < f. (A12) og (A13) 2 Mina, J, Xiao, J Y. Return to Risk Metrics:
then imply that the coefficient of substitution The Evolution of a Standard. New York,
between return and risk is less than nq. (A12) RiskMetrics, 2001.
gives the following expression for :
EA f
= . (A14)
A
By substituting (A4) and (A11) into (A12) we
get an analogous capital market line which ties
together ENrM and VaRq(NrM):
(1 + r)ENrM + r =
1+r
f + ENrm VaRq (NrM ) (A15)
nq
or
(f r)nq
ENrM = +
(1 + r)(nq )

VaRq (NrM ) (A16)

where = . This is meaningful since
nq
< nq.
represents the balancing of expected NPV

against risk. We observe that is independent
of r.
A2 The Profitability of a Project

We propose to measure the profitability of a pro-
ject by balancing the expected NPV against the
value-at-risk where is used as the weighting
coefficient. Thus we choose to let the companys
willingness to take risk be the same as for the
market portfolio. The expected NPV and the
value at risk depend on the choice of discount
rate. We suggest the use of the companys esti-
mated average capital cost as a value for r. It
represents the average return on investment
within the company and is thus a good estimate
of what the company may obtain through an
alternative use of the money. (We are not dog-
matic here. Other considerations may cause the
choice of another discount factor.) Thus the
profitability of a project P is measured by the
risk-adjusted expected NPV (RAENPV) given
by
ENr (VaRq(Nr)). (A17)
The definition (2.3.2) together with (A16) gives

that the RAENPV of the market portfolio is
f r
. (A18)
(1 + r)(1 /nq )

Forecasting An Important Factor for
Network Planning
KJELL STORDAHL
The paper gives an overview of forecasts and forecast methodology used for network planning. Specific
attention is given to how forecasts are applied for development of strategies and planning. An extensive
list of references is annexed for more detailed studies.
1 Is Forecasting Necessary for For example, Telenor SHDSL subscription fore-

Network Planning? casts for the business market for one specific
Network planning is an important activity for the local access area are based on the following:
operators. In order to utilise the resources and
investment means in the best possible way, it is Estimation of potential DSL accesses for dif-
Kjell Stordahl (58) received his of crucial importance to have insight in future ferent market segments, especially type of
M.S. in statistics from Oslo Uni-
telecommunication demand. A professional fore- industry (SIC code) and size in the specific
versity in 1972. He worked with
Telenor R&D for 15 years and casting process will show the expected evolution area
with Telenor Networks for 15 of the telecommunication demand. Forecasts of the DSL business penetration in
years, mainly as manager of
the area
Planning Department Region
Oslo and then Market analysis. Questions to be solved in the network planning Forecasts of the evolution of demand of sym-
Since 1992 he has participated process are: metric business access demand in the specific
in various techno-economic EU
area
projects (TITAN, OPTIMUM,
TERA, TONIC) analysing rollout Choice of technology and network components Prediction of expected market share in the
of different broadband technolo- Design of network structure specific area
gies. Kjell Stordahl has been
Routing principles and redundancy in the net-
responsible for working pack-
ages on broadband demand, work There are different players in the telecommuni-
forecasting and risk analysis in Dimensioning of nodes and routes in the net- cation market. Because of open network provi-
these projects. He has pub-
work sioning additional operators have been estab-
lished more than 140 papers in
international journals and con- Timing of network expansion lished in the market. The access network is
ferences. Implementation of additional functionality available based on Local Loop UnBundling and
kjell.stordahl@telenor.com Introduction of new and enhanced services several operators hire the copper pair from the
Integration of functionality on various OSI incumbents. In addition wholesale has been
levels introduced to create a more open and competi-
Replacement strategies for old network com- tive telecommunication market with new service
ponents/old technology providers. The telecommunication market is
Long-term strategy planning for network evo- more complex and there are even more need for
lution forecast modelling taking into account the new
environments.
Important forecasts to support the planning pro-
cess are: There are always economic risks related to net-
work planning. The forecasting process should
Forecasts for service demand generate reduced risks by identifying the new
Forecasts for enhanced services demand and enhanced applications and make predictions
Identification of demand for new services of access and traffic demand generated by the
Subscription/access forecasts for the services new and traditional services in environments
Traffic volume forecasts for services and where the competition and evolution of expected
applications market share are described.
Busy hour traffic forecasts for services and
applications Of course there are significant uncertainties
Forecasts based on market segmentation related to the forecasting process. Hence, an
Forecasts taking into account competition and important part of the process is to describe the
market share uncertainty and try to incorporate the uncertainty
Forecasts separated into national level, evaluations into the network planning process.
regional level and local level
Forecasts allocated to the various networks 2 Network Evolution and the
Forecasts allocated to transport network, Forecasts
regional network and access network The circuit switched telephone network has
evolved through certain milestones from being
Specific forecasts are defined by combining the an analogue network to a network consisting of
items on the list. only digital exchanges. The next step was the

introduction of ISDN. Now, Telenor has the The market, the established services, the en-
highest ISDN penetration in the world. During hanced and new services and applications create
the last few years the IP based Internet has telecommunication demand. Market forecasts
evolved significantly. In Norway the Internet are important and necessary for taking the right
penetration is more than 60 %. Before the resi- decisions for network evolution. The forecasts
dential broadband take off, the main capacity help to get the right timing from introduction
in the transport network was made up by leased of new network platforms and services. More
lines and PSTN/ISDN traffic. A significant part detailed description of the forecasts are shown
of PSTN/ISDN traffic has been narrowband in [1, 3, 5, 7-8, 11-13, 26-27, 30-31, 34-35, 37,
Internet traffic. 51, 55, 57, 63, 65, 74, 82-83, 85-87].
Now we see conversion of narrowband Internet 3 Techno-Economic Tool

traffic to broadband ADSL traffic. New services for Strategic Evaluations
and applications are established and new demand of New Technology and
generated. The broadband platform will be en- Network Structures
hanced either through ADSL2+ or VDSL, which To be able to make the right decision for rolling
in turn include broadband entertainment services out a new network platform, a comprehensive
in the fixed network. In parallel HFC networks techno-economic analysis has to be carried out.
from the cable operators UPC and Telenor Avidi
have been deployed and Fixed broadband radio Within the European programs RACE, ACTS
systems like LMDS are established. and IST, the projects RACE 2087/TITAN, AC
226/OPTIMUM, AC364/TERA and IST-2000-
Today there is a significant conversion of voice 25172 TONIC have developed a methodology
traffic from the PSTN/ISDN to mobile networks. and a tool for calculation of the overall financial
The mobile platform is evolving from GSM to budget of any access architecture. The tool han-
GPRS, possibly to EDGE and to UMTS. In addi- dles the discounted system costs, operations,
tion WLAN hot spots are deployed. maintenance costs, life cycle costs, net present
value (NPV) and internal rate of return (IRR).
The business market has evolved through data The tool has the ability to combine low level,
line switched and data packet switched network detailed network parameters of significant strate-
without considerable success, to leased lines, gic relevance with high level, overall strategic
Internet and DSL. Now, new services like sym- parameters for performing evaluation of various
metric DSL (SHDSL), fast- and GB Ethernet are network architectures [4046, 5859, 66, 7172].
introduced.
Demand for the Telecommunications

Services
Services Architectures
Architectures
Topology
DB
Revenues First Installed

Investments
Cost
OA&M Life Cycle

Costs Cost
Economic Cash flows,

Inputs Profit & loss accounts
Year 0 Year 1 Year n ... Year m
Figure 1 TONIC techno-

NPV IRR
Payback economic tool for investment
Period projects analysis

Figure 1 shows a techno-economic tool used in The OA&M costs are divided into different
different international EU projects. components like cost of repair parts including
civil work and operations and administration
The figure shows that telecommunication de- costs. Typically the OA&M costs are being
mand is an input to the tool and to the techno- driven by services, say by number of customers
economic calculations. The TONIC tool is and number of critical network elements.
widely used to analyse economic consequences
by implementing new network platforms. Impor- Investment costs together with the OA&M costs
tant parts of the tool are: give the life-cycle cost for the selected architec-
ture scenario. Finally, by combining service rev-
Service definitions enues, investments, operating costs and general
Subscription and traffic forecasts economic inputs (e.g. discount rate, tax rate)
Service tariff predictions OPTIMUM, cash flows and other economic fac-
Revenue model tors (NPV, IRR, Payback period etc) are calcu-
A topology model mapping geographic areas lated.
with given penetrations into the tool
Network component cost data base including 4 Strategic Evaluations
more than 300 network components of New Technology and
Network component cost prediction model Network Structures
Investment model The possibility to introduce new services, en-
Operation and maintenance model hanced services, new applications, additional
Model for economic calculations traffic growth and generation of additional rev-
Risk analysis model enue is important for network operators. Strate-
gies for introduction of new technologies and
The following steps are needed in the techno- network platforms open the possibilities for gen-
economic evaluations of the network solutions: eration of additional revenue. Some examples of
introduction of new technology during the last
The services to be provided must be specified. 20 years are:
The market penetration of these services over the
study period will be defined. The services have Establishment of NMT
associated tariffs, i.e. the part of the tariff that is Deployment of optical fibre and fibre technol-
attributed to the network under study. From the ogy in the network
combination of yearly market penetration and Digitalisation of the PSTN
yearly tariff information TONIC calculates the Introduction of ISDN
revenues for each year for the selected service set. Establishment of GSM
Introduction of Internet
Next, the architecture scenarios to provide the Establishment of ATM network
selected service set must be defined. This needs Establishment of IP networks
network design expertise and is mostly outside Establishment of SDH transmission technology
of the framework of TONIC methodology. How- Introduction of ADSL
ever, TONIC includes several geometric models,
which facilitate the network design by automati- The next step is enhancement of the ADSL plat-
cally calculating lengths for cables and ducting. form to ADSL2+ or VDSL and for the mobile
These geometric models are optional parts of the network UMTS, WLAN and 4G. Important
methodology and TONIC can be used without input for making the right decisions of introduc-
them. The result of an architecture scenario defi- ing new technology has been demand forecasts
nition is a so-called shopping list. This list indi- for new and enhanced services. The forecasts
cates the volumes of all network cost elements and tariff predictions give revenue forecasts,
(equipment, cables, cabinets, ducting, installa- which are used together with techno-economic
tion etc.) for each year of the study period and analysis to support decisions of introduction of
the location of these network components in dif- new technology.
ferent flexibility points and link levels.
The traffic forecasts are also fundamental for
The costs of the network components are calcu- dimensioning the networks and establishing
lated using an integrated cost database devel- optimal network structures.
oped within the TONIC project, containing data
gathered from many European sources. Archi- Techno-economic evaluations of new network
tecture scenarios together with the cost database technology is analysed in [2, 4, 1415, 1825,
give investments for each year. 2829, 33, 36, 3839, 48, 50, 5254, 60, 68, 70,
76, 81, 84, 8889].

5 Subscriber/Access Subscription forecasts have been developed for
Forecasting (narrowband) Internet, ATM, ADSL, SHDSL,
Network planning of the access network depends SHDSL point-to-point, Fast Ethernet and GB
heavily on subscriber forecasts. Until recently, Ethernet and leased lines. The subscription fore-
substantial investments in the local telephone casts give valuable information for planning, di-
exchanges have been based on forecasts of the mensioning and expanding the network structure.
subscriber growth in the local area. The sub-
scriber forecasts indicated time for expansion However, it is important to take into account sig-
of the telephone exchange. In addition the sub- nificant substitution effects between the services.
scriber forecasts were used together with queu- The ISDN and PSTN forecasts are mutually
ing models to estimate the busy hour traffic vol- dependent. The ADSL forecasts influence sig-
ume in the exchange. When the data traffic was nificantly the narrowband Internet forecasts
limited, Erlangs dimensioning rule was used for and ISDN and PSTN forecasts. Lost PSTN and
traffic dimensioning taking into account the dis- ISDN market shares for the incumbent increase
tribution of residential and business customers. the leased line demand forecasts. Subscription
30 years ago the business customer traffic was forecasts for 2.5G and 3G increase the leased
dominating and the busy hour was around lunch. lines forecasts. Increased ADSL subscriber fore-
Now, traffic from residential customers is domi- casts and demand for LLUB increase the leased
nating on the main part of the Norwegian ex- line forecasts because of expansion of the traffic
changes. During the last few years the busy hour capacity between DSLAM and broadband access
has been moved from after the main television point, and because of additional traffic in the
news programme to the 21.0022.00 period, transport network. Subscriber demand forecasts
mainly because of the Internet traffic. for SHDSL and SHDSL point-to-point and
point-to-multipoint will lower the leased lines
ISDN was introduced in the Norwegian market forecasts, since the new services are leased lines
10 years ago. There have been significant substi- substitutes with cheaper tariffs and a degraded
tution effects between ISDN and PSTN in this service quality.
period. The ISDN forecasts consist of: 2B+D
residential, 2B+D business and 30B+D business. To be able to control the dependencies, forecast-
Also multiple ISDN line forecasts will be imple- ing models for the different services are linked
mented. together.
The ISDN forecasts have been important for plan- 6 Traffic Forecasting
ning and providing new network components. An important part of network planning is design
of network structures. The PSTN/ISDN consists
Because of competition and substitution effects of the access network including local exchanges,
between services, the penetration of PSTN and region networks including group exchanges and
ISDN have saturated the Norwegian network. the long distance network including long dis-
Now, the new PSTN and ISDN forecasts reflect tance exchanges and national exchanges. To
a smaller number of subscribers. There are two maintain a high degree of reliability, indepen-
main reasons for the decrease: dent routes are established between local ex-
changes and group exchanges on the one hand
Some residential customers substitute their and between group exchanges and long distance
main telephone subscriptions with mobile exchanges on the other hand. Between the long
subscriptions distance exchanges there is a logical mesh net-
work. However, even when there is a mesh net-
Some residential customers substitute their work there is a simpler physical network taking
PSTN/ISDN Internet connections with ADSL. into account the logical mesh network abilities.
The physical network is based on deployed fibre
These forecasts are crucial for network planning. rings and SDH transmission equipment.
The forecasts show as a function of time the
spare capacity in the network regarding traffic Network planning designs optimal structure of
capacity and number of connections. the network with the nodes (exchanges) and the
routes by minimising the costs for a given redun-
The subscriber forecasts have been important dancy. Important factors are the exchange sites,
also for restructuring the access network. The the route length and the capacity on the routes.
forecasts are used for planning and establishing Specific network planning tools have also been
service connection points in the access network developed to minimise the investments. Differ-
and for dimensioning the access lines and fibre ent tools have been developed for the access net-
rings between the service connection points. work and for the transport network. The tools
can be applied for establishing new network
structures or for expanding the network.

The traffic forecasts are important input for period is found by minimising the total costs.
designing the network and for dimensioning the The estimation of additional capacity depends on
exchanges and capacity on the different routes the planning process and the technical systems
and as input to the network planning tools. and will vary from one incumbent to another.
The busy hour traffic forecasts quantify increases/ 8 Forecasting Methodologies

decreases in the traffic. The dimensioning calcu- A variety of forecasting models are used for pre-
lations based on queuing theory add some extra dicting the evolution of the telecommunication
capacity to control random variation in the traf- market. An important task is collection and anal-
fic load during the busy hours. When the traffic ysis of statistics historical data. A lot of re-
on a route or in an exchange approaches the sources are used to maintain rather large cus-
capacity, an expansion will be carried out. The tomer based systems. An extraction of data
expansion of the capacity will be based on the from these systems is used to create subscription
forecasts for a given planning period. statistics, which is a base for subscriber fore-
casts. In addition specific measurement systems
The forecasting procedure described is used not are established to extract traffic measurements
only for the PSTN/ISDN but for other networks on specific points in the networks and also on
like PSDN, Frame relay/ATM, Digital Cross the aggregated level.
Connect, various IP networks, networks for
mobile operators and leased lines. There are dif- The forecasting models take into account histori-
ferent possibilities for making traffic forecasts. cal data. The most common models for access
One possibility is to measure the traffic and use forecasts are:
the traffic measurements as a historic database
for the forecasts. One problem is rerouting of the Diffusion models
traffic based on new network structures or Regression models
because of change of interconnection sites in the Econometric models
network. Such changes affect the traffic mea- ARIMA models
surements, which have to be adjusted before the Kalman filter models
forecasts can be developed. Another possibility Smoothing models
is to analyse the total traffic growth in the given
network and make the forecasts on the aggre- The access forecasting models have to take into
gated level. account substitution effects between services as
pointed out in chapter 5. Even the competition
A detailed description of the traffic forecasting between the incumbent and the other operators
procedure for the transport network is described influences not only the access forecasts but also
in [1]. The traffic forecasting procedure for the the BOT and leased lines forecasts. Telenor is
transport network is rather complex since the using a rather comprehensive composite model
network carries business and residential traffic where the substitution effects between PSTN,
from all the other networks. ISDN, Leased lines, ADSL, VDSL/ADSL2+,
HFC, FTTH, FWA are included. Important fac-
7 Capacity Forecasts tors in the model are the predicted technology
Models for making traffic forecasts during the coverage for the coming years, the predicted
busy hour have been described. However, the market shares etc.
capacity needed to carry the traffic is higher.
Additional capacity has to be dimensioned tak- The most common models for traffic forecasts
ing into account stochastic variations around the are:
mean traffic in the busy hour. Usually Erlangs
blocking formula is applied for voice traffic, Regression models
while Lindbergers approximation is useful for Econometric models
dimensioning the data traffic capacity [90]. The ARIMA models
traffic will probably be transported on SDH sys- Kalman filter models
tems where packet overhead is added. In addi- Smoothing models
tion the systems average load factor is less than
the maximum capacity. Finally there will in gen- In addition specific forecasting procedures can
eral be some extra dimensioning since the ex- be used. The traffic is generated by sources and
pansion of the network is planned and deployed moves in a network from source to a set of sinks.
stepwise. A planning period is defined as the To dimension the network, it is important to
time between two expansions. The duration of know the traffic behaviour from the edge routers
the planning period is found by taking into or the local exchanges. The traffic streams are
account the repeating costs by performing the mapped in a traffic matrix. Each element in the
expansion and the unused investment means matrix denotes the traffic between two ex-
during part of the period. The optimal planning changes. Suppose the number of exchanges is

N. Then an NxN matrix describes the traffic be- ple to use as much information as possible.
tween all exchanges. Specific forecasting proce- Hence, forecasts of the traffic elements, the row
dures can be used to forecast the traffic in the sums, the column sums and the total traffic are
network or the matrix. The procedures are used used. The model is described as follows:
when the network structure is stable during a
certain period. Cijt is the traffic between i and j at the time t
Ci.t is the traffic from i at time t
The best known traffic matrix forecasting proce- C.jt is the traffic from j at time t
dure is Kruithofs method [103]. The method C..t is the total traffic at time t
makes traffic matrix forecasts based on the ob-
served traffic matrix, let us say at time 0, and The forecasts at time t is given by Cijt, Cijt, Cijt
forecasts of the outgoing traffic and forecasts of and C..t respectively. The extended weighted
the incoming traffic from an exchange, let us say leased square forecasts are denoted by {E} and
at time t. The outgoing traffic corresponds to the found by solving the following minimisation
row sums in the traffic matrix and the incoming problem:
traffic corresponds to the column sums. There
will be inconsistency between the row sums at The square sum Q is defined by:
time 0 and the forecast row sums at time t and in
the same manner for the column sums. To reach Q = ij . (Eijt Cijt) + i (Ei.t Ci.t)
consistency, Kruithofs method tries to take into + j (E.jt C.jt) + (E..t C..t)
account both the traffic structure at time 0 and
the row and column forecasts in the best possible where {ij}, {i}, {j} and are given constants.
way. Hence Kruithofs method uses an iteration The square sum is minimised given that adjusted
procedure to get a compromise between the traf- forecasts {E} are consistent, i.e. satisfy the con-
fic matrix structure at time 0 and the forecasts. dition that the row sum is equal to the row sum
The iteration procedure upgrades the rows in the forecasts of each element and the column sum is
traffic matrix to correspond with the row sum equal to the column sum forecasts of each ele-
forecasts. Then the iteration procedure upgrades ment and the sum of the elements in the whole
the columns in the traffic matrix to correspond to matrix is equal to the total forecasts. A natural
the column forecasts. After some iterations the choice of the weights is the inverse of the vari-
adjusted traffic matrix elements correspond to ance of forecast uncertainty of the forecasts. One
the row and column forecasts, which gives a way to find estimates of the forecasts uncer-
traffic matrix forecast at time t. tainty is to perform ex-post forecasts and then
calculate the mean square error. The solution
An extension of Kruithofs method can be per- of the minimisation problem is found by using
formed by making point-to-point forecasts for Lagranges multiplicator method adding the con-
each element in the traffic matrix together with straints based on consistency in the adjusted
forecasts of the outgoing traffic and incoming forecasts. By using the method we get N(N + 3)
traffic from each exchange and then adjust the + 2 equations when N is the dimension of the
traffic elements in the matrix based on the traffic matrix. The solution of the system of
exchange forecasts using the same iteration equations gives the traffic matrix forecasts.
procedure.
The forecasts methodologies are described in
A further extension of Kruithofs traffic matrix more detail in [91103].
forecasting procedure has been developed. The
weighted least squares method is based on the 9 Forecasts Uncertainty and
fact that the relative uncertainties in the traffic the Risk
element forecasts are larger than the row and There are always uncertainties connected to
column sum forecasts. The method calculates forecasts. Network planning decisions have to be
the adjusted traffic element forecasts by taking made based on expectations of future evolution.
into account the relative uncertainty in the traffic The uncertainty in the evolution can be classi-
element forecasts and the row and column sum fied in the following main groups:
forecasts. The different forecasts are found by
weighting the forecasts according to their uncer- Uncertainty in market (penetration and market
tainty using weighting least square method. share) forecasts
Extended least square method takes also into Uncertainty in tariff forecasts (predictions)
account the total traffic forecasts for the whole
traffic matrix. The total traffic is defined as the Uncertainty in network component cost pre-
sum of the row sums or the sum of the column dictions.
sums in the matrix, which of course are identi-
cal. The method is based on the statistical princi-

NPV
10 000 Trials Frequency Chart 52 Outliers
Component Price
.027 269
.020 201.7
.013 134.5
.007 67.25
.000 0
0.68 0.74 0.80 0.86 0.92 -3000 -1000 1000 3000 5000
kECU
Service Penetration
Statistical Variation of the input
parameters
Using Monte Carlo Simulation
Results: probability distribution,
risk profile of the business case
Extended basis for investment
1.09 1.54 2.00 2.46 2.91
decisions
Revenue per customer
Figure 2 Risk analysis

0.55 0.78 1.00 1.23 1.45
The effect of uncertainty in the predictions should ing to their impact. This uncertainty ranking is
be quantified to examine the consequences. The based on the percentage to the contribution to
first step is to find the most critical variables in the variance of the NPV or the rank correlation
the project. The next step is to perform calcula- with the NPV.
tions based on variation in the critical variables
to identify the consequences. A commercial spreadsheet application Crystal
Ball has been integrated with the techno-eco-
One possibility is to apply sensitivity analysis. nomic tool. A graphical interface in Crystal Ball
Another and more advanced method is to use makes it possible to specify the distribution
the risk analysis. The output in an economic risk functions directly from a palette type of menu
analysis is net present value (NPV), internal rate inside the techno-economic model.
of return (IIR) and pay back period. One of the
outputs or all of them simultaneously can be The risk analysis has been applied on a set of
used in risk analysis. different cases studying strategies for rolling out
new network technology [6, 910, 1617, 32,
Instead of using the expected forecast, say at 37, 47, 49, 56, 6164, 67, 69, 73, 75, 77, 7880].
time t, a probability distribution of the forecast
at time t has to be used. Similar probability dis- 10 Conclusions
tributions have to be constructed for all critical The paper documents that forecasts and forecast
variables. Usually truncated Normal distribu- methodology are important factors in the net-
tions or Beta distributions are used. work planning process and strategy process for
rolling out new technology platforms. Since the
A Monte Carlo simulation with 1000 5000 tri- forecasts are uncertain, it is recommended to use
als is performed. In each trial, a random number risk analysis to evaluate the risk generated by
is picked from the predefined probability distri- forecasts and predictions of other critical vari-
butions; one for each of the critical variables. ables. For more detailed studies the papers in
The simulation gives as output the cumulative the reference list should be studied.
distribution of NPV, IIR and pay back period
and the ranking of the critical variables accord-

References 12 Hjelkrem, C, Stordahl, K, Be, J. Forecast-
1 Stordahl, K. Traffic forecasting models for ing residential broadband demand with lim-
the incumbent based on new drivers in the ited information. A long term supply and
market. Telektronikk, 99 (3/4), 122127, demand model. 2002 International Commu-
2003 (this issue). nications Forecasting Conference, San Fran-
cisco, USA, June 2528, 2002.
2 Elnegaard, N K, Olsen, B T, Stordahl, K.
Broadband deployment in rural and non 13 Olsen, B T, Stordahl, K. Traffic forecast
competitive areas: The European perspec- models for the transport network. In: Proc
tive. In: Proc FITCE 2003, Berlin, Germany, Networks 2002, Munich, Germany, June
September 14, 2003. 2327, 2002.
3 Stordahl, K, Kalhagen K O. Broadband fore- 14 Monath, T et al. Economics of Ethernet ver-

cast modelling. Evaluation of methodology sus ATM-based access networks for broad-
and results. International Symposium on band IP services. In: Proc Networks 2002,
Forecasting 2003, Merida, Mexico, June Munich, Germany, June 2327, 2002.
1619, 2003.
15 Varoutas D et al. Economic viability of 3G
4 Welling, I et al. Techno-Economic Evalua- Mobile Virtual Network Operators. In: Proc
tion 3G & WLAN Business Case Feasibility 3G Wireless 2002, San Francisco, USA,
Under Varying Conditions. In: Proc 10th May 2831, 2002.
International Conference on Telecommuni-
cations, Tahiti, French Polynesia, February 16 Kalhagen, K O, Elnegaard, N K. The Eco-
23 March 1, 2003. nomics and Risks of 3rd Generation Mobile
Service Deployment. In: Proc 3G 2002,
5 Stordahl, K, Kalhagen, K O. Broadband Amsterdam, Netherlands, May 2023, 2002.
Access Forecasts for the European Market.
Telektronikk, 98 (2/3), 2132, 2002. 17 Stordahl, K, Elnegaard, N K. Battle between
cable operator and incumbent: Optimal
6 Elnegaard, N, Stordahl, K. Deciding on opti- ADSL/VDSL rollout for the incumbent. In:
mal timing of VDSL and ADSL roll-outs for Proc. XIV International Symposium on Ser-
incumbents. Telektronikk, 98 (2/3), 6370, vices in the Local access, ISSLS 2002, Seoul,
2002. Korea, April 1418, 2002.
7 Kalhagen, K O, Elnegaard, N K. Assessing 18 Olsen, B T, Kalhagen, K O, Stordahl, K.

Broadband Investment Risk Trough Option Provisioning of broadband services in non-
Theory. Telektronikk, 98 (2/3), 5162, 2002. competitive areas. In: Proc. XIV Interna-
tional Symposium on Services in the Local
8 Stordahl, K. Developing demand forecasts access, ISSLS 2002, Seoul, Korea, April
and tariff predictions for different telecom 1418, 2002.
services both for fixed and mobile based on
results from the IST TONIC project. IIR 19 Elnegaard, N K, Stordahl, K. Deciding on
Telecoms Market Forecasting 2002, Prague, the right timing of VDSL rollouts: A real
Czech Republic, October 710, 2002. options approach. In: Proc. XIV Interna-
tional Symposium on Services in the Local
9 Kalhagen, K O, Elnegaard, N K. The Eco- access, ISSLS 2002, Seoul, Korea, April
nomics and Risks of 3rd Generation Mobile 1418, 2002.
Service Deployment. Launching Appealing
Mobile Services, September 1819, London, 20 Monath, T et al. Economics of Fixed Access
UK. Networks for broadband IP services (in Ger-
man). 3. ITG-Fachtagung Netze und Anwen-
10 Elnegaard, N K. How to incorporate the dungen, Duisburg, Germany, February 28
value of flexibility in broadband access net- March 1, 2002.
work rollout investment projects. 41st Euro-
pean Telecommunications Congress (FITCE), 21 Katsianis, D et al. The financial perspective
Genova, Italy, September 47, 2002. of the mobile networks in Europe. IEEE Per-
sonal Communications Magazine, 8 (6),
11 Stordahl, K, Kalhagen, K O, Olsen, B T. 5864, 2001.
Broadband technology demand in Europe.
2002 International Communications Fore-
casting Conference, San Francisco, USA,
June 2528, 2002.

22 Ims, L A et al. Building a solid business plan 33 Stordahl, K et al. Optimal roll out strategies
for developing fibre in the access network. for XDSL broadband upgrades of the access
In: Proc. Optical Access Networks, London, network. In: Proc. International Telecommu-
UK, Oct 35, 2001. nication Society 2000, Buenos Aires,
Argentina, July 25, 2000.
23 Ims, L A et al. End-to-end-solutions for Tel-
cos is it a Broadcasters World? In: Proc. 34 Stordahl, K, Ims, L A, Moe, M. Forecasts of
Interactive Convergence, London, UK, Sep broadband market evolution in Europe. In:
35, 2001. Proc. International Telecommunication
Society 2000, Buenos Aires, Argentina, July
24 Ims, L A et al. From a large scale VDSL 25, 2000.
market trial towards commercial services
key issues. In: Proc. DSLCon Asia, Hong 35 Hjelkrem, C. Forecasting access demand
Kong, China, Aug 1316, 2001. when information is scarce : A case study of
the Norwegian ISDN market. In: Proc. Inter-
25 Elnegaard, N K et al. From VDSL market national Telecommunication Society 2000,
trials to commercial launch The key issues Buenos Aires, Argentina, July 25, 2000.
on content, services, technical platform and
the business case. In: Proc. XDSL Summit, 36 Cerboni, A et al. Evaluation of demand for
Geneva, Switzerland, July 1113, 2001. next generation mobile services. In: Proc.
International Telecommunication Society
26 Hjelkrem, C. Forecasting with limited infor- 2000, Buenos Aires, Argentina, July 25,
mation: A study of the Norwegian ISDN 2000.
access market. Journal of Business Fore-
casting, 20 (3), 2001. 37 Stordahl, K et al. Broadband upgrading and
the risk of lost market shares Under Increas-
27 Stordahl, K, Elnegaard, N K, Olsen, B T. ing Competition. In: Proc. ISSLS 2000,
Broadband access rollout strategies in a Stockholm, Sweden, June 1923, 2000.
competitive environment. In: Proc. Optical
Hybrid Access Network/Full Service Access 38 Elnegaard, N K, Stordahl, K, Ims, L A. Roll-
Network workshop, Yokohama, Japan, April Out Strategies for the Copper Access Net-
35, 2001. work: Evolution towards a Full Service
Access Network. In: Proc. ISSLS 2000,
28 Ims, L A et al. Market driven deployment Stockholm, Sweden, June 1923, 2000.
of next generation networks increasing
the service footprint by hybrid broadband 39 Katsianis, D et al. The Economics of Next
access. In: Proc. Hybrid Optical Access Generation Mobile Systems. ISSLS 2000,
Network 2001 (HOAN 2001), Yokohama, Stockholm, Sweden, June 1923, 2000.
Japan, April 35, 2001.
40 Olsen, B T. Introduction to telecom invest-
29 Elnegaard, N K, Stordahl, K. Broadband ments and techno-economics. SAS session.
rollout strategies. In: Proc. 17th European Techno-economics of multimedia services
Network Planning Workshop, Les Arcs, and networks, ICC 2000, New Orleans,
France, March 1923, 2001. USA, June 1822, 2000.
30 Stordahl, K, Elnegaard, N K. Broadband 41 Stordahl, K. Broadband market evolution

market evolution in Europe and the upgrade demand forecasting and tariffs. SAS session.
risks. 2000 International Communications Techno-economic methodologies and case
Forecasting Conference, Seattle, USA, studies, ICC 2000, New Orleans, June
September 2629, 2000. 1822, 2000.
31 Stordahl, K, Moe, M, Ims, L A. Broadband 42 Ims, L A. Designing case studies. SAS ses-
market The driver for network evolution. sion. Techno-economics of multimedia ser-
In: Proc. Networks 2000, Toronto, Canada, vices and networks, ICC 2000, New Orleans,
September 1016, 2000. USA, June 1822, 2000.
32 Elnegaard, N K, Stordahl, K, Ims, L A. The 43 Olsen, B T. The TERA project main objec-
risks of DSL access network rollouts. In: tives and results. SAS session. Techno-eco-
Proc. Networks 2000, Toronto, Canada, nomics of multimedia services and networks.
September 1016, 2000. ICC 2000, New Orleans, USA, June 1822,
2000.

44 Stordahl, K. Overview of risks in multimedia 57 Istad, S, Stordahl, K. Broadband demand
networks. SAS session. Techno-economic survey in the residential and SOHO market
methodologies and case studies, ICC 2000, in Norway. Telektronikk, 95 (2/3), 4549,
New Orleans, June 1822, 2000. 1999.
45 Ims, L A. Case studies on broadband access: 58 Olsen, B T. OPTIMUM a techno-eco-

HFC, LMDS, VDSL, ADSL. SAS session. nomic tool. Telektronikk, 95 (2/3), 239250,
Techno-economics of multimedia services 1999.
and networks, ICC 2000, New Orleans,
USA, June 1822, 2000. 59 Tahkokorpi, M, Lhteenoja, M. Techno-Eco-
nomic Guidelines for Telecommunication
46 Olsen, B T. The challenge of predicting cost Networks and Services. Telektronikk, 95
evolution. SAS session. Techno-economics (2/3), 236238, 1999.
of multimedia services and networks, ICC
2000, New Orleans, USA, June 1822, 2000. 60 Ims, L A et al. Towards broadband access in
Norway the view from Telenor. Telektron-
47 Stordahl, K. Case studies on broadband ikk, 95 (2/3), 191201, 1999.
access: Market risk analysis for fiber roll-
out. SAS session. Techno-economics of mul- 61 Elnegaard, N K, Ims, L A, Stordahl, K.
timedia services and networks, ICC 2000, Techno-economic risk assessment of PNO
New Orleans, USA, June 1822, 2000. access network evolutionary paths. Telek-
tronikk, 95 (2/3), 286290, 1999.
48 Olsen, B T, Stordahl, K, Ims, L A. Evolution
of network architectures. In: Proc. 16th 62 Stordahl, K et al. Risk methodology for
European Network Planning Workshop, multimedia projects assessments. In: Proc.
Les Arcs, France, March 2024, 2000. ECMAST 1999, Madrid, Spain, May 2628,
1999.
49 Stordahl, K et al. Overview of risks in Multi-
media Broadband Upgrades. In: Proc. 63 Stordahl, K. Evaluation of key risks in
Globecom 99, Rio de Janeiro, Brazil, broadband network development taking into
December 510, 1999. account uncertainties in market forecasts and
competition. In: Proc. Future Network Plan-
50 Ims, L A et al. The economics of broadband ning Strategies, Vision in Business, London,
infrastructure upgrade investments in various March 24, 1999.
European markets : The EURESCOM view.
Telecom 99 Forum, Geneva, Switzerland, 64 Elnegaard, N K, Stordahl, K. Technology
October 917, 1999. choice risk analysis and its impact on your
business development and market position-
51 Stordahl, K. Identifying the key risks in net- ing. In: Proc. Future Proof Network Plan-
work development, taking into account ning Strategies, Vision in Business, London,
uncertainties in market forecasts and compe- UK, March 24, 1999.
tition. In: Proc. Future proof network plan-
ning strategies, Vision in Business, Berlin, 65 Hjelkrem, C. Forecasting the demand for
Germany, September 2728, 1999. ISDN accesses when the information is
scarce : A case study from the Norwegian
52 Ims, L A. Introduction to broadband access market. International Business Forecasting
networks. Telektronikk, 95 (2/3), 222, 1999. 99, Las Vegas, USA, February 26, 1999.
53 Ims, L A. Wireline broadband access net- 66 Budry, L et al. The economics of broadband
works. Telektronikk, 95 (2/3), 7387, 1999. service introduction : Area specific invest-
ments for broadband upgrade. In: Proc.
54 Ims, L A. Design of access network case TRIBAN, Bern, Switzerland, November
studies. Telektronikk, 95 (2/3), 251253, 1999. 1719, 1998.
55 Stordahl K, Rand, L. Long term forecasts for 67 Stordahl, K et al. Broadband access network
broadband demand. Telektronikk, 95 (2/3), competition analysis of technology and
3444, 1999. market risks. In: Proc. Globecom 98, Syd-
ney, November 812, 1998.
56 Stordahl, K, Ims, L A, Olsen, B T. Risk
methodology for evaluating broadband
access network architectures. Telektronikk,
95 (2/3), 273285, 1999.

68 Ims, L A et al. Economics of broadband 79 Ims, L A, Olsen, B T, Myhre, D. Economics
access network upgrade strategies: The of residential broadband access network
European perspective. In: Proc. Globecom technologies and strategies. IEEE Networks,
98 Access Networks Mini Conference, 11 (1), 5864, 1997.
Sydney, November 812, 1998.
80 Stordahl, K. Risk assessments of broadband
69 Stordahl, K et al. Evaluating broadband upgrade strategies. In: Proc. Broadband
strategies in a competitive market using risk strategies the battle for the customer, IBC
analysis. In: Proc. Networks 98, Sorrento, conference, London, December 45, 1997.
October 1823, 1998.
81 Ims, L A et al. Evolution of technologies and
70 Ims, L A, Myhre, D, Olsen, B T. Investment architectures to a full services network. In:
costs of broadband capacity upgrade strate- Proc. Broadband strategies the battle for
gies in residential areas. In: Proc. Networks the customer, IBC conference London,
98, Sorrento, Italy, October 1823, 1998. December 45, 1997.
71 Lhteenoja, M et al. Broadband access net- 82 Stordahl, K. Forecasting long term demand
works: techno-economic methodology and for broadband services and their influence
tool application results. IEEE symposium on broadband evolution. Institute of Interna-
on planning and design of broadband Net- tional Research (IIR). In: Proc. Market
works, Montebello, Quebec, October 811, Forecasting in the Telecoms Industry,
1998. London, UK, December 14, 1997.
72 Ims, L A (ed). Broadband Access networks : 83 Stordahl, K, Ims, L A, Olsen, B T. Forecasts

Introduction Strategies and Techno-eco- and risk analysis of PNO and Cable opera-
nomic Evaluation. London, Chapman & tors by introducing broadband upgrades in
Hall, 1998. (ISBN 0 412 82820 0) the access network. 1997 International
Communication Forecasting Conference,
73 Stordahl, K, Ims, L A, Olsen, B T. Risk San Francisco, June 2427, 1997.
analysis of residential broadband upgrades
based on market evolution and competition. 84 Ims, L A et al. Key factors influencing the
In: Proc. Supercom ICC 98, Atlanta, June overall costs and project values of broadband
711, 1998. access network upgrades. NOC 97,
Antwerp, Belgium, June 1720, 1997.
74 Stordahl, K. Forecasting long term demand
for broadband services. In: Proc. Market 85 Stordahl, K. Forecasting long term demand
forecasts in the telecoms industry, Institute for broadband services in the residential
of International Research (IIR), Hong Kong, market: Devising a method for developing
May 2527, 1998. forecasts for new services. Institute of Inter-
national Research (IIR). In: Proc. Market
75 Olsen, B T et al. Multimedia business cases Forecasting in the Telecoms Industry, Lon-
in a deregulated environment : Opportunities don, UK, May 1921, 1997.
and risks. ISSLS 1998, Venice, March
2227, 1998. 86 Hjelkrem, C. Forecasting models for ISDN
accesses in the Norwegian market and their
76 Ims, L A et al. Key factors influencing influences on the investments strategies.
investment strategies of broadband access Institute of International Research (IIR). In:
network upgrades. In: Proc. ISSLS 98, Proc. Market Forecasting in the Telecoms
Venice, Italy, March 2227, 1998. Industry, London, UK, May 1921, 1997.
77 Stordahl, K. Evaluating broadband upgrade 87 Stordahl, K. Techno-economic analysis,

strategies in a competitive market using risk market forecasts and risk analysis of broad-
analysis. Institute of International Research band access network upgrading. In: Proc.
(IIR). In: Proc. Evaluating Broadband Ser- 13th European Network Planning Workshop,
vice Strategies, London, UK, February Les Arcs, France, March 915, 1997.
1820, 1998.
88 Olsen, B T. Access Network Evolution. In:
78 Ims, L A, Stordahl, K, Olsen, B T. Risk Proc. 13th European Network Planning
analysis of residential broadband upgrade in Workshop, Les Arcs, France, March 915,
a competitive environment. IEEE Communi- 1997.
cation Magazine, June, 1997.

89 Olsen, B T et al. Techno-economic evalua- 96 Stordahl, K. Prognoser som grunnlag for
tion of narrowband and broadband access planlegging. (Forecasts as a base for plan-
network alternatives and evolution scenario ning). Telektronikk, 82 (3), 1986.
assessment. IEEE Journal of Selected Areas
in Communications, 14 (8), 1996. 97 Stordahl, K, Holden, L. Traffic forecasting
models based on top down and bottom up
90 Lindberger, K. Analytical methods for the procedures. In: Proc. 11th International
traffical problems with statistical multiplex- Teletrafic Congress, Kyoto, Japan, Septem-
ing in ATM network. In: Proc. 13th Interna- ber 1985.
tional Teletraffic Congress, Copenhagen,
June 1926, 1991. 98 Stordahl, K. Routing algorithms based on
traffic forecast modelling. Time series analy-
91 Stordahl, K. Methods for traffic matrix fore- ses Theory and practice. O D Anderson
casting. In: Proc. 12th International Tele- (ed.). Amsterdam, North Holland Publishing
traffic Congress, Torino, Italy, June 18, Company, 1985.
1988.
99 Stordahl, K. Forecasting models for traffic in
92 Stordahl, K. Extended weighted minimum the future broadband network. In: Proc. V
squared method for forecasting traffic matri- Nordic Teletraffic Seminar, Trondheim, Nor-
ces. In: Proc. VII Nordic Teletraffic way, June 57, 1984.
Seminar, Lund, Sweden, August 2527,
1987. 100 ITU. Forecasting of traffic. Geneva, ITU,
1984. CCITT Red Book Rec E.506.
93 Stordahl, K. Forecasting models for network
planning. In: Proc. VI Nordic Teletraffic 101 Stordahl, K. Routing algorithms based on
Seminar, Copenhagen, Denmark, August traffic forecast modelling. Time series analy-
2830, 1986. ses Theory and practice. Toronto, Canada,
August 1821, 1983.
94 Stordahl, K, Damsleth, E. Traffic forecasting
methods for network planning. In: Proc. 6th 102 Stordahl, K. Centralized routing based on
International Conference on Forecasting forecasts of the telephone traffic. In: Proc.
and analysis for business in the information 10th International Teletraffic Congress,
age, Tokyo, Japan, August 1986. Montreal, Canada, June 1983.
95 ITU. Procedures for traffic matrix forecast- 103 Kruithof, J. Telefoonverkeerrsrekening. De

ing. Geneva, ITU, 1986. CCITT Com II 50-E. Ingenieur, 52 (8), 1937.

Traffic Forecasting Models for the Incum-
bent Based on New Drivers in the Market
KJELL STORDAHL
1 Introduction Gigabit Ethernet, Lamda wavelength, Dynamic

The circuit switched voice traffic has tradition- bandwidth allocation.
ally been a significant part of the traffic in the
transport network. However, during recent years Operators: ADSL, SDSL, VDSL, LMDS,
Leased lines ordered by different operators and Leased lines, Fast Ethernet, Gigabit Ethernet IP-
also the Leased lines ordered from the business VPN, Lamda wavelength, Dynamic bandwidth
market have expanded the transport network allocation. (mobile operators, ISPs, other opera-
capacity. A new capacity wave has also started: tors).
the data traffic moving from narrowband to
Kjell Stordahl (58) received his broadband. The data traffic is increasing expo- Network Platforms
M.S. in statistics from Oslo Uni-
nentially and will for some years be the dominat- Relevant network platforms are: PSTN/ISDN,
versity in 1972. He worked with
Telenor R&D for 15 years and ing traffic in the transport network. Important PSDN, Frame Relay/ATM, Digital Cross Con-
with Telenor Networks for 15 traffic drivers are broadband applications carried nect, Various IP Networks including IP net-
years, mainly as manager of
by HFC, ADSL, VDSL, LMDS, UMTS and works for mobile operators, Leased lines.
Planning Department Region
Oslo and then Market analysis. WLAN. PSTN/ISDN is a circuit switched network.
Since 1992 he has participated Leased lines have no concentration effect at all,
in various techno-economic EU
This paper analyses the traffic and capacity evo- while the other network platforms also have
projects (TITAN, OPTIMUM,
TERA, TONIC) analysing rollout lution of the transport network of an incumbent packet switched concentration.
of different broadband technolo- operator having the possibility to integrate dif-
gies. Kjell Stordahl has been
responsible for working pack-
ferent type of traffic into the network. A traffic 3 Traffic From the Residential
ages on broadband demand, volume indicator is developed for traffic in- Market
forecasting and risk analysis in crease in the transport network. The access fore- The residential market generates different types
these projects. He has pub-
cast modelling has been developed based on of traffic: Voice traffic, Dialled Internet traffic,
lished more than 140 papers in
international journals and con- parts of the results from the projects ACT 384 ADSL traffic, VDSL/LMDS traffic.
ferences. TERA and IST-2000-25172 project TONIC
kjell.stordahl@telenor.com [116]. The voice traffic has nearly reached saturation.
During the next years, the circuit switched voice
2 Market Segments traffic will be rather stable before parts of the
circuit switched voice traffic are substituted by
Services IP voice. The dialled Internet traffic will reach
Traffic from the services is transported on differ- maximum within a few years. Then the dialled
ent network platforms or on leased lines. Impor- Internet traffic will continuously be substituted
tant services for the transport network are: by broadband traffic. A battle has already started
PSTN/ISDN, Internet, Leased lines, PSDN between broadband operators to capture parts
(packet switched data network), Frame relay, of the broadband market.
ATM, IP Virtual private network (VPN),
ADSL/SDSL, VDSL/LMDS, Fast Ethernet, Broadband access forecasts have been developed
Gigabit Ethernet, Lamda wavelength, Dynamic in the IST-2000-25172 project TONIC. Figure 1
bandwidth allocation. shows the market share evolution of ADSL,
VDSL, FWA (Fixed wireless broadband access)
Market Segments and HFC/cable modem for West European coun-
An incumbent operator leases transport capacity tries.
to other operators. In addition the incumbent
offers transport capacity to the residential and Figure 2 shows total broadband penetration. The
the business market. The incumbent operator total broadband penetration forecasts are ad-
offers transport capacity either via own service justed compared to the forecasts developed in
provider or as wholesale. A segmentation of the the TONIC projects. A combination of the fig-
market will be: ures gives the broadband penetration for each
technology.
Residential market: PSTN, ISDN, Internet,
ADSL, VDSL, LMDS, HFC. Now, the question is which traffic is carried in
the incumbents transport network. Usually the
Business market: PSTN, ISDN, IP VPN, Inter- cable TV/HFC traffic is carried outside the trans-
net, PSDN, Frame Relay, ATM, ADSL, SDSL, port network, while a part (market share) of the
VDSL, LMDS, Leased lines, Fast Ethernet,

60 ADSL Figure 1 Market share
VDSL distribution and prediction
50
FWA of ADSL, VDSL, FWA and
40 HFC (cable modem) for West
HFC
European countries
30
20
10
0
2001 2003 2005 2007 2009
60
Figure 2 Broadband
50 penetration forecasts for the
European residential market
Penetration (%)
40
30
20
10
0
1999 2001 2003 2005 2007 2009
penetration of each of the other technologies will Market Share and Access Penetration
be carried in the incumbents transport network. The factor Nt Mit pit represents a forecast of
number of households connected to the incum-
A traffic volume forecast indicator,VR(t), for the bents transport network in year t using technol-
residential busy hour traffic into the transport ogy i. Suppose the incumbent operator has 40 %
network is given by: of the ADSL market share and expects to be in
the same position the next years, then M3t = 0.40
VR(t) = Nt i=1,2,3,4,5 bit uit Ait Cit Mit pit (1) for t = 2001, 2002, ... .
where Mean Downstream Access Capacity

and the Technology Evolution
i = 1 denotes voice traffic The downstream access capacity, C1t , for voice
i = 2 denotes Dialled Internet traffic is 64 kbs and will be the same the next
i = 3 denotes ADSL years. However, the downstream access capacity
i = 4 denotes VDSL C3t for ADSL changes. Figure 3 shows how the
i = 5 denotes FWA mean downstream access capacity increases.
Nt is the number of households in year t Now, the operators offer a set of different access
bit is busy hour concentration factor for tech- capacities. There will be an evolution from low
nology i in year t access capacities to higher access capacities
uit is packet switching concentration factor for especially because of new and enhanced applica-
technology i in year t tions. In addition, new functionality such as
Ait is the access capacity utilisation for tech- bandwidth on demand will be introduced by the
nology i in year t operators.
Cit is mean downstream access capacity for
technology i in year t So far, ADSL based on reasonable low down-
Mit is incumbents access market share for stream capacity is introduced. The next step will
technology i in year t be to enhance the ADSL capacity in the range 6
pit is the access penetration forecasts (%) for 8 Mbs. However, the capacity size offered will
technology i in year t be dependent on the twisted copper pair length.
The capacity offered will decrease with the cop-
per pair length.

Figure 3 Evolution of 4.50
predicted average downstream
capacity in Mbs (Tonic 2002) 3.50
2.50
1.50
0.50
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
An extension to the VDSL platform will Access Capacity Utilisation

increase the offered capacity further. New ser- A broadband customer is not utilising the maxi-
vices like VoD and peer-to-peer applications mum capacity all the time. The access capacity
based on downloading and exchange of films utilisation factor indicates average capacity use
will increase the capacity demand. The access taking into account the proportion of time during
capacity for VDSL will be in the range 15 24 the conversation for downloading and the pro-
Mbs depending on the copper pair length. portion of time for uploading. The factor also
reflects the degree of using the specified band-
ADSL2+ is a supplement to VDSL and ADSL. width.
While VDSL covers subscribers up to 1.5 km
from the exchange or the last fibre drop, Busy Hour Concentration
ADSL2+ has the possibility to offer 10 Mbs up The busy hour concentration effect is well
to 2.0 km and probably within a few years 2.5 known. Usually about 10 % (b1t = 0.1) of the
km distance from the exchange/last fibre drop. customers make phone calls in the busy hour.
Traditionally Erlangs blocking formula (assum-
ADSL2+ and VDSL have enough capacity to ing exponential interarrival time and holding
offer Internet and digital TV while also offering time) is used for dimensioning. The busy hour
the possibility to use interactive TV. The number concentration factor is increasing because of the
of independent TV streams will be dependent on Internet. For broadband connections the busy
the selected solutions. It will also be possible to hour concentration factor is significantly higher
transfer traditional TV programs on the copper because of heavy users, longer holding times,
pairs like the ones seen in cable TV networks. flat rate and evolution of new applications.
However, only part of this capacity will be indi-
vidual, while the capacity for dedicated TV The busy hour for residential narrowband and
channels will be a common resource for all broadband traffic is in the evening.
households in the transport network.
Packet Switching Concentration
Dynamic Spectrum Management (DSM) will The circuit switched services PSTN and ISDN
enhance the copper line capacity further. The have no packet switching concentration (u1t = 1).
methodology is based on dynamic regulation of The other services have significant concentra-
the frequencies on the copper line to reduce the tions. Internet use consists of sessions based
noise and cross talk. downloading, thinking and uploading. Traffic
will be packet according to use. Traditional
The market share evolution for different ADSL Internet use gives low packet switching concen-
access capacities is described in the IST-2000- tration. Applications like music on demand and
25172 project TONIC. Mean access capacity is video on demand generate high packet switching
calculated according to the distribution of differ- concentration. The evolution of the packet
ent access capacities for each year. The results switching concentration factor is complex.
are shown in Figure 3.
Uncertainty in the Concentration
The capacity predictions are based on the Factors
assumption of flat rate principle for the first Figure 4 shows possible evolutions of combin-
years. However, there may be a delay in the ations between busy hour concentrations and
demand for capacity if and when traffic charges packet switching concentrations. There are sig-
are introduced by the operators. nificant uncertainties in the evolution. The basis

for the predictions in Figure 4 is 0.15 (15 %) 0.300 Alternative 1
busy hour concentration and 0.20 packet switch- Alternative 2
ing concentration in 2001. Four alternatives are Alternative 3
defined having a linear yearly increase: 0.200 Alternative 4
Busy hour concentration: 0.15 in 2001 to respec-

tively 0.195 0.24 0.285 0.33 in 2010. 0.100
Packet switching concentration: 0.20 in 2001 to

respectively 0.335 0.47 0.605 0.74 in 2010.
0
2001 2003 2005 2007 2009
Figures 24 show a nearly exponential evolution
of broadband penetration, capacity increase and
traffic concentration in the coming years. The
most probable prediction will be between alter-
native 2 and 3. The traffic volume indicator The mobile operators are important transport Figure 4 Description of
described in equation (1) has a much stronger network customers using leased lines in the possible evolutions of
exponential evolution because of a multiplica- transport network. In the coming years these concentrations of ADSL traffic
tive effect of the same factors. operators will generate the following traffic: as a function of busy hour and
packet switching concentration
4 Traffic from the Business 2G traffic (Digital mobile systems such as
Market GSM)
The business market generates the following 2.5G traffic (HSCSD, GPRS, EDGE)
types of traffic/capacity: Voice traffic, Dialled 3G traffic (UMTS)
Internet traffic, PSDN, ATM, Frame Relay, DSL 3.5G traffic (Ubiquitous roaming among 3G
traffic, IP Virtual Private Networks (IP VPN) and WLAN systems)
traffic, Leased lines, Fast and Gigabit Ethernet.
The access forecasts for the different systems
There are significant substitution effects between have been developed in the IST-2000-25172
DSL, IP VPN, Leased lines, Fast and Gigabit project Tonic.
Ethernet, which have to be taken into account in
the forecasting process. Leased lines are used to Subscription forecasts for the mobile systems,
establish fixed connections between sites often allowing the subscribers to have more than one
based on head office and branch offices or subscription have also been developed. The
between different enterprises. The established traffic capacity for data applications per user
network forms a local network with high service increases from 40 kbs to 1.92 Mbs from 2G to
quality. There are no busy hour concentration 3G, while WLAN offers up to 54 Mbs. The traf-
or packet switching concentration. Leased lines fic forecast model for mobile traffic is similar to
constitute a significant part of the transport net- the traffic in the fixed network. The busy hour
work capacity. Some parts of the leased lines concentration factor and the packet switching
capacity will be transferred to IP VPN or DSL concentration factor are extremely important for
because of cheaper tariffs and in spite of reduced the dimensioning. The services/applications are
service quality/SLA. defined in different service classes: conversa- Figure 5 Subscriber
tion, streaming, best effort, and the capacity is penetration forecasts for
A traffic volume forecast indicator VB(t) for the reserved according to service level agreements. different mobile systems
business market busy hour traffic is given by:
VB(t) = Nt i bit uit Ait Cit Mit pit (2)

100
where the different traffic/capacity types i are 2G
defined in the first paragraph of the chapter. 2.5G
80 3G
5 Traffic Generated by Other 3.5G
Operators 60 Sum
Different operators like mobile operators, ISPs
and other fixed network operators lease neces-
40
sary capacity in the transport network. The
capacity demand depends on type of services
offered and the market share to the operators and 20
of course the probability to use the transport net-
work of the incumbent. 0
1998 2000 2002 2004 2006 2008 2010

The traffic forecasting model VM(t) for mobile 7 Conclusions
operators is given by: A traffic volume indicator has been developed to
estimate busy hour traffic increase in the trans-
VM(t) = Nt i bit uit Ait Cit Mit pit (3) port network. The indicator estimates the traffic
entering the transport network. The traffic vol-
Here Nt is number of persons, not number of ume indicator does not include redundancy and
households. Mit denotes the market share to the protection capacity in the core network. Telenor
operator and i = 1, 2, 3, 4 the system generations uses the traffic volume indicator forecasts as
2G, 2.5G, 3G and 3.5G. The penetration fore- input to the transport network planning process
casts pit are shown in Figure 5. The capacity Cit evaluation of new network structures, expansion
is a mean capacity. A 3.5G subscriber has the of the network and introduction of new core
possibility to use a data rate up to 144 kbs using network technology.
UMTS but a significantly higher capacity in
WLAN hot spot. The number of hot spots avail- The indicator depends on the application evolu-
able and the proportion of time the subscriber tion and also the tariff regime for broadband
uses WLAN compared with UMTS give the services. So far, most countries use a flat rate
mean capacity C4t . The factor Ait indicates the for broadband traffic. However, specific applica-
real utilisation of the capacity. tions will overload the transport network heavily
if no actions are performed. There exist no in-
Let VO(t) be the busy hour traffic forecasts for citements to control the size of the traffic in the
the other operators, then the total of busy hour transport network. A probable solution will be
traffic forecasts V(t) is given by: to introduce a tariff on high capacity traffic and
on bandwidth on demand and specific applica-
V(t) = VR(t) + VB(t) + VM(t) + VO(t) (4) tions. The traffic forecasts used assume that a
new tariff regime for broadband services will
There are definitely possibilities to reduce V(t) be implemented within the next two years.
since the business traffic has busy hour before/
after lunch, while the residential traffic has busy References
hour in the evening. Since the operators use 1 Stordahl, K, Kalhagen, K O. Broadband
leased lines, day and night capacity is equal. Let Access Forecasts for the European Market.
VBL(t) be the leased line capacity and VBP(t) be Telektronikk, 98 (2/3), 2132, 2002.
the packet switched traffic in the business mar-
ket. Then VB(t) = VBL(t) + VBP(t). The residential 2 Elnegaard, N, Stordahl, K. Deciding on opti-
broadband traffic is larger than the packet mal timing of VDSL and ADSL roll-outs for
switched business traffic in the busy hour. Sup- incumbents. Telektronikk, 98 (2/3), 6370,
pose that = 0.2 (20 %) of the packet switched 2002.
business busy hour traffic is transferred during
the residential busy hour the adjusted busy hour 3 Stordahl, K, Kalhagen, K O, Olsen, B T.
traffic forecast V * (t) is: Broadband technology demand in Europe.
ICFC 2002, San Francisco, USA, June
V * (t) = VR(t) + VBP(t) + VBL(t) 2528, 2002.
+ VM(t) + VO(t) (5)
4 Stordahl, K, Elnegaard, N K. Battle between
6 Capacity Forecasts cable operator and incumbent : Optimal
Models for making traffic forecasts during busy ADSL/VDSL rollout for the incumbent. In:
hour have been described. However, the capacity Proc. XIV International Symposium on Ser-
needed to carry the traffic is higher. Additional vices in the Local Access ISSLS 2002,
capacity has to be dimensioned taking into Seoul, Korea, April 1418, 2002.
account stochastic variations around the mean
traffic in the busy hour. Usually Erlangs block- 5 Monath, T et al. Economics of Ethernet
ing formula is applied for voice traffic, while based Access Networks for broadband IP
Lindbergers approximation is useful for dimen- Services. Proc. ISSLS 2002, Seoul, Korea,
sioning the data traffic capacity [12]. The traffic April 1418, 2002.
will probably be transported on SDH systems
where packet overhead is added. In addition the 6 Stordahl, K, Elnegaard, N K. Risk analysis
systems load average factor is less than the max- of broadband access rollout strategies in a
imum capacity, and finally there will in general competitive environment. In: Proc. Optical
be some extra dimensioning since the expansion Hybrid Access Network/Full Service Access
of the network is planned and deployed stepwise. Network workshop, Yokohama, Japan, April
The estimation of additional capacity depends on 46, 2001.
the planning process and the technical systems
and will vary from one incumbent to another.

7 Stordahl, K, Elnegaard, N K. Broadband 13 Elnegaard, N K, Ims, L A, Stordahl, K. Roll-
market evolution in Europe and the upgrade out strategies for the copper access network
risks. ICFC, Seattle, September 2629, evolution towards a full service access net-
2000. work. In: Proc. ISSLS 2000, Stockholm,
Sweden, June 1923, 2000.
8 Stordahl, K et al. Broadband market, the
driver for network evolution. In: Proc. Net- 14 Stordahl, K et al. Overview of risks in Multi-
works 2000, Toronto, Canada, September media Broadband Upgrades. In: Proc.
1015, 2000. Globecom 99, Rio de Janeiro, Brazil, Dec
510, 1999.
9 Elnegaard, N K, Ims, L A, Stordahl, K. The
risks of DSL access network rollouts. In: 15 Stordahl, K et al. Evaluating broadband
Proc. Networks 2000, Toronto, Canada, strategies in a competitive market using risk
September 1015, 2000. analysis. In: Proc. Networks 98, Sorrento,
Italy, Oct 1823, 1998.
10 Stordahl, K, Ims, L A, Moe, M. Forecasts of
broadband market evolution in Europe. In: 16 Ims, L A (ed.). Broadband Access Networks
Proc. ITS 2000, Buenos Aires, Argentina, Introduction strategies and techno-eco-
July 25, 2000. nomic analysis. Chapman-Hall, 1998.
11 Stordahl, K. Broadband market evolution 17 Lindberger, K. Dimensioning and design

demand forecasting and tariffs. ICC 2000, methods for integrated ATM networks. In:
SAS session. Techno-economics of multi- Proc. 14th ITC, Antibes, France, June, 1994.
media services and networks. New Orleans,
USA, June 1822, 2000.
12 Stordahl, K et al. Broadband upgrading and

the risk of lost market share under increasing
competition. In: Proc. ISSLS 2000, Stock-
holm, Sweden, June 1923, 2000.

Planning Dependable Network for
IP/MPLS Over Optics
TERJE JENSEN
It is always challenging for an operator to find a sufficient level of functionality in the different network
layers, also in the IP and optics areas as discussed in this paper. With the growing traffic demand,
steadily more emphasis is placed on finding efficient network solutions, also considering resilience
options. A number of options are described in this article.
1 Introduction puts from sources such as the Optical Intercon-

The ongoing traffic growth, mainly related to IP- nection Forum, IETF and ITU. The Generalised
based services requires a steady improvement in Multi-Protocol Label Switching (GMPLS) has
Dr. Terje Jensen (41) is network efficiency. This is to carry the traffic while been proposed as a means for controlling the
still reducing the infrastructure cost, and, hence, resources (mainly seen from the IP layer, but
recent years he has mostly been allowing for lowering the prices to customers. also from other protocol layers). A brief descrip-
engaged in network strategy tion is given in Chapter 3, together with MPLS
Many people question to what extent the traffic recognized as a starting point for the GMPLS
Besides these activities he has load growth in the core of IP-based networks is specification. Resilience can be provided in dif-
been involved in internal and hindered by the access link capacity (e.g. dial- ferent ways and supported at different layers, as
up, ISDN, GSM). Introducing access links with reflected in a number of options treated in
ling/analyses and dimensioning. higher capacity, such as DSL and Ethernet/fibre, Chapters 410.
terje.jensen1@telenor.com the traffic loads in the core networks may grow
even more drastically. This is one argument for 2 Various Models for Inter-
investigating the use of optics in closer connec- connecting IP and Optics
tion with IP (although the general traffic growth
and price trends for optics also advocate this). 2.1 Principles
In order for the resources to be utilised effi-
One of the means to step up the traffic handling ciently, the optical networks must be survivable,
capability of the IP network is to develop routers flexible and controllable. Some have suggested
with higher throughput. Some means to be introducing more intelligence in the control
undertaken are: plane for the optical networks in order to achieve
this. Hence, utilising similar mechanisms as
Separate forwarding and route determination found for the IP-based networks is looked at as
and make the routing software leaner; IP is seen as being a common protocol for much
of the traffic carried by the optical network.
Introduce interfaces with higher transfer rates,
increased switching speed; A first issue is the adaptation and reuse of IP
control plane protocols for the control plane in
Introduce hardware adapted solutions (e.g. optical networks. These are to be used no matter
through application-specific integrated cir- which traffic flows (IP or non-IP) are carried.
cuits, ASICs). A second issue is how IP traffic can be carried
where joint control and co-ordination between
Making leaner software solutions, in some the IP and optical layer are utilised.
respect, may be contrasting the functionality for
traffic handling according to the Traffic Engi- This is illustrated in Figure 1: An optical subnet-
neering mechanisms. Reducing the number of work may consist of all-Optical Cross-Connects
layers is one step to reduce the overhead. Hence, (OXCs) or some nodes where optical-electrical-
IP direct over optics has become a subject optical conversion is applied. The switching
gaining more interest. function within an OXC is controlled by setting
the parameters of the cross-connect unit. This
Basically, there will be a number of clients to can be considered as configuring a cross-connect
an optical network sublayer. However, in this table that specifies which input port is connected
article the interplay between IP and optics is to which output port. Here, a port can indicate
looked at, expected to become the dominating fibre as well as wavelength.
traffic volume to be carried in the long run.
Two types of control interface are indicated;
Models for interconnecting the IP and the optical User-Network Interface (UNI) between the
layers are described in Chapter 2, including in- clients and the optical network, and, Network-

UNI NNI UNI
IP optical optical IP Figure 1 Schematic
network subnetwork subnetwork network illustration of optical
optical
network with client networks
UNI subnetwork
NNI UNI
other client NNI
other client
network network
optical
subnetwork
NNI = Network-Network Interface

UNI = User-Network Interface
Network Interface (NNI) between optical sub- tivity (services like light-path creation, dele-
networks. The control flow across the UNI tion, modification and status enquiry);
would naturally depend on the services offered
to the client. As the NNI control flow would be Unified service model where the IP and the
derived from IP control, similarities between optical network are treated together, as seen
NNI and UNI may well exist when an IP net- from the control plane perspective, as an inte-
work is the client. In addition to these two inter- grated network. Then the OXCs will be treat-
face types, interface within in an optical subnet- ed like any router as seen from the control
work also needs to be defined, such as the inter- plane. No distinction is then made between
face between two OXCs. Some refer to this UNI, NNI and any other router-to-router inter-
interface as an internal NNI. face. Such an interface is assumed to be based
on extensions for MPLS/GMPLS.
Between administrative domains it is essential to
consider issues like security, scalability, stability It is important to make a separation between the
and information hiding. In principle, the UNI control plane and the data plane over the UNI.
and the NNI could be implemented in the same As mentioned, the optical network basically pro-
way. However, one commonly seeks to limit the vides services to clients in the form of transport
information needed to be transferred across capacities (by light-paths). IP routers at the edge
interfaces, thereby motivating for a separation of the optical network must establish such paths
of UNI and NNI, allowing for a specialisation before the communication at the IP layer can
of their implementations. start. Therefore, the IP data plane over optical
network is done over an underlying network of
The UNI can be regarded as a client-server inter- optical paths. On the other hand, for the control
face; for example, the IP layer is the client, while plane, the IP routers and the OXCs can have
the optical layer is the server. The client roles peering relations, in particular for routing infor-
would then request a service connection and the mation exchanges. Various degrees of loose or
server role establishes the connection to meet the tight coupling between the IP and the optical
request when all admission control conditions network may be used. The coupling is given by
are fulfilled. The physical implementation of the the details of topology and routing information
UNI may vary, like exchanged, level of control that IP routers can
exercise on selecting specific paths, and policies
direct interface with an in-band or out-of-band regarding dynamic provisioning of optical paths
IP control channel. This channel is used to between routers (including access control, acc-
exchange signalling and routing messages ounting and security).
between the router and the OXC (like a peer-
ing arrangement); Three interconnection models are sketched:
indirect interface with out-of-band IP control Overlay model: The routing, topology distri-
channel. The channel may run between man- bution, signalling protocols are independent
agement systems or servers; for the IP/MPLS and the optical network.
provisioned interface involving manual opera- Augmented model: Routing instances in the IP
tions. layer and the optical network are separated by
information exchanged (e.g. IP addresses are
Two service models, in principle applicable both known to the optical routing protocols).
for UNI and NNI, are outlined in [ID_ipofw]:
Peer model: The IP/MPLS layers act as peers
Domain service model where the optical net- to the optical network. Then a single routing
work primarily offers high bandwidth connec- protocol instance can be used for the IP/MPLS
network and the optical network.

Figure 2 UNI used between
service provider and Service provider domain
user domains
User domain UNI Connection control plane UNI User domain
(signalling) (signalling)
UNI UNI
(transport) (transport)
Optical transport network
These models give a certain degree of imple- near-term deployment, although helping in the
mentation complexity; the overlay being the migration toward the peer model. This is said to
least complex one for near-term deployment and support incremental development as the inter-
the peer model the most complex one. As each connection model increases in complexity.
of the models has its advantages, an evolution
path for IP over optical network may be seen. 2.2 User Network Interface OIF
Utilising WDM as an intelligent transport service
A possible migration path is to start with the allows its clients such as IP routers to intercon-
simpler functionality, meaning the domain ser- nect and relate to optical interfaces. The increas-
vice model with overlay interconnection and no ing bandwidth need places more weight on the
routing exchange between the IP and the optical capability to manage and control optical net-
network. A provisioned interface would be ex- works. Currently, optical networks are provi-
pected. The next phase of the migration path is sioned through element management systems,
to exchange reachability information between implying that end-to-end connections across
IP and the optical network. This may allow for equipment from different manufacturers are likely
light-paths to be established in conjunction with to involve several incompatible management sys-
setting up Label Switched Paths (LSPs). The tems. This, again, will likely result in longer pro-
third phase of the migration might be supporting visioning times and manual effort required.
the peer model.
The Optical Interconnection Forum (OIF) has
Applying a common signalling framework from addressed this issue by adopting MPLS-based
the start would assist the migration. For the control intelligence for use within optical net-
Figure 3 Illustration of UNI domain service model, implementation agree- works. A standard signalling interface between
signalling used to establish ment based on Generalised MPLS (GMPLS) client and optical network has also been defined
and release bandwidth UNI signalling is being developed by the Optical to support dynamic provisioning requests.
between routers Interworking Forum (OIF). This is intended for
Version 1.0 of OIF UNI allows clients to estab-
lish optical connections dynamically by applying
OC-48s signalling procedures compatible with General-
Router ized MPLS (GMPLS), see Figure 2.
A
OC-48s
Router In addition to signalling, the IOF UNI specifica-
B
Optical network tion also describes a neighbour discovery mech-
anism and a service discovery mechanism. The
former allows nodes at both ends of a fibre link
to identify each other (e.g. reducing the manual
effort needed to build data bases of network
inventory). The service discovery mechanism
Bandwidth (Gb/s) allows clients to determine the services that are
7.5 supported by the optical network, also any new
Bandwidth connected services introduced.
5.0
An application of the capabilities is to support
dynamic traffic engineering controlled by the IP
2.5
Traffic load router traffics. An example, as shown in Figure
3, is to establish or release bandwidth depending
0
on the traffic level for example estimated by
Time traffic monitoring.

2.3 Requirements on Optical General requirements:
Network Services - Separation of networking functions group-
Some of the challenges for network operators are ing functions into control plane, data plane
efficient bandwidth management and fast service (transport plane) and management plane.
provisioning in a multi-technology and possibly
multi-vendor networking environment. ITU has - Separation of call and connection control
started work on Automatically Switched Optical call relates to end-to-end, admission control,
Network (ASON) recommendations aiming to while connection control relates to establish-
providing optical networks with intelligent net- ing and releasing the connections for a call.
work functions and capabilities in its control
plan. This enables fast connection establishment, - Network and service scalability and response
dynamic rerouting and multiplexing and switch- times to be supported, enabling several thou-
ing at different granularity levels (including sands of termination ids per OXCs, set-up
fibre, wavelength and TDM channel). The fol- time should preferably be less than a number
lowing benefits are strived for: of seconds.
Automated discovery (network inventory, - Services offered over different underlying

topology and resource) types of transport technologies including
optical and TDM.
Rapid circuit provisioning
- Well-defined service building blocks.
Enhanced protection and restoration
Service user requirements:
Service flexibility (protocol and bit-rate trans- - The common services must be supported
parent and bandwidth-on-demand) with the corresponding interfaces (SDH,
wavelength, Ethernet, Storage Area Net-
Enhanced interoperability works) with various bit rates, Service Level
Agreement (SLA) conditions, service classes
As described in [ID-iopreq] a number of addi- and routing options.
tional benefits may also be offered by ASON:
- Service invocation may be initiated by the
Reactive traffic engineering at optical layer provider (permanent and soft permanent)
that allows network resources to be dynami- or by the user side (switched). Interplay be-
cally allocated to traffic flow. tween control and management must take
place to allow for different initiation modes.
Reduce the need for service providers to
develop new operational support systems soft- Service provider requirements:
ware for the network control and new service - Supporting different modes for connecting
provisioning on the optical network, thus users; direct connection, remote access via
speeding up the deployment of the optical net- sub-networks, dual homing).
work technology and reducing the software
development and maintenance cost. - Different inter-domain connectivity schemes
must be supported. In general one seeks
Potential development of a unified control autonomy between different domain sig-
plane that can be used for different transport nalling/control, both for domain within an
technologies, including OTN, SDH, ATM and operator and between operators. Several
PDH. points of interconnection between domains
must be possible.
Services are commonly described by their
topologies; bi-directional or uni-directional and - Different name and address arrangements
point-to-(multi-)point. In addition, persistency must be allowed, depending on the clients
may be given, such as permanent, switched and (e.g. IP addresses for IP routers). Address
soft permanent. The difference between perma- resolution and translation functions should
nent and soft permanent is that only end-points be present.
in a domain are specified for the latter, leaving it
to the network to establish a connection between - Policy-based service management must be
these points. supported to enable flexible service provi-
sioning.
A number of requirements, based on the descrip-
tion in [ID-iopreq], can be categorised as:

Figure 4 Potential OXC system architecture hop, source-routing and hierarchical routing.
OXC with MPLS control plane
In the latter case, information aggregation
would be needed. Both scalability and stabil-
MPLS control plane ity aspects are to be considered. Exchange of
routing information may be trigger-based
and timeout-based.
Control adaption
- Path selection must allow for different algo-
OXC switch controller
rithms, including shortest path and con-
straint-based routing (constraints such as
cost, link utilization, diversity, service class).
OXC switch fabrc
OXC data plane - Both manual and automatic discovery shall
be supported for neighbour (physical and
logical), resource and service.
Control plane requirements (in combination Requirements on resiliency for service and
with management plane): control plane:
- Basic capabilities required are network - Different service levels are to be offered,
resource discovery, address assignment and accompanied by SLA conditions.
resolution, routing information propagation
and dissemination, path calculation and - Priorities for signalling messages should be
selection, connection management. implemented in order to allow for faster
restoration.
- A signalling network can be established
independently of the data transport plane 2.4 Optical Transport Networks
topology. Three types of relations can be A high level architectural model has been
found: worked on in IETF, grouping the modelling
i) in-band signalling: signalling messages are aspects into a horizontal dimension and a verti-
carried in a logical channel imbedded in the cal dimension. The horizontal dimension refers
data-carrying link; to special requirements for an Optical Transport
ii) in-fibre, out-of-band: signalling messages Network (OTN) including considerations like:
are carried in a dedicated communication
channel separated from the data-carrying Type of OTN state information should be
channels, but within the same fibre; discovered and disseminated to support path
iii) out-of-fibre: signalling messages are car- selection for optical channels (e.g. attenuation,
ried over a different fibre than the data-car- dispersion).
rying links.
Infrastructure used for propagating the control
- Interface to data plane to be standardised information.
allowing for the control plane to configure
switching fabrics and port functions (via Computing constrained paths fulfilling perfor-
management plane) and receive failure and mance and policy requirements.
degradation status information.
Domain specific requirements for establishing
- The control plane is considered a managed optical channels and enhancements for MPLS
entity and, hence, interacts with the manage- signalling protocols for addressing these
ment plane for configuration, status report- requirements.
ing and so forth.
The vertical dimension includes concerns when
Requirements on signalling, routing and dis- porting MPLS control plane software onto an
covery: OXC. A potential architecture of an OXC is
- Signalling must support both strict and loose illustrated in Figure 4.
routing and allow individual as well as
groups of connections. Fault notifications Looking closer into an OTN as described by
must also be supported. Crank-back and ITU-T, it should be noted that it is itself divided
rerouting shall be supported for inter-domain into layers, including
signalling.
an optical channel (OCh) layer network;
- Routing includes information on reachabil-
ity, topology/resource and path computation. an optical multiplex section (OMS) layer
Routing mechanisms to support are hop-by- network;

Box A OTNT: ITU-T Recommendations on the OTN Transport Plane
Framework for Recommendations
G.871/Y.1301 Framework for Optical Transport Network Recommendations 10/00
Architectural Aspects
G.872 Architecture of Optical Transport Networks (Revised, 11/01 pre-publ.) 2001
Structures & Mapping
G.709/Y.1331 Network node interface for the optical transport network (OTN) 02/01
G.975 Forward Error Correction 10/00
Functional characteristics
G.681 Functional characteristics of interoffice long-haul line systems using optical amplifiers,
including optical multiplexing 10/96
G.798 Characteristics of optical transport network (OTN) equipment functional blocks 11/01,
corr. 2002
G.806 Characteristics of transport equipment Description Methodology and Generic Function-
ality 10/00
G.7710/Y.1701 Common Equipment Management Requirements 11/01
Protection Switching
G.841.x Protection Switching in the OTN 2002
G.gps Generic Protection Switching 2002
Management Aspects
G.874 Management aspects of the optical transport network element 11/01
G.874.1 Optical Transport Network (OTN) Protocol-Neutral Management Information Model For
The Network Element View 01/02
G.875 Optical Transport Network (OTN) management information model for the network ele-
ment view
Data Communications network (DCN)
G.7712/Y.1703 Architecture and specification of data communication network 11/01
G.dcn living list
Error Performance
G.optperf Error and availability performance parameters and objectives for international paths
within the Optical Transport Network (OTN) 2003
M.24otn Error Performance Objectives and Procedures for Bringing-Into-Service and
Maintenance of Optical Transport Networks 2003
Jitter & Wander Performance
G.8251(G.otnjit) The control of jitter and wander within the optical transport network (OTN)
11/01, amend/corr. 2002
Physical-Layer Aspects
G.691 Optical Interfaces for single-channel SDH systems with Optical Amplifiers, and STM-64
and STM-256 systems 10/00
G.692 Optical Interfaces for Multichannel Systems with Optical Amplifiers 10/98
G.694.1 Spectral grids for WDM applications: DWDM frequency grid 2002
G.694.2 Spectral grids for WDM applications: CWDM wavelength grid 2002
G.664 General Automatic Power Shut-Down Procedures for Optical Transport Systems 06/99
G.959.1 Optical Transport Networking Physical Layer Interfaces 02/01
G.693 Optical interfaces for intra-office systems 11/01
Sup.dsn Optical System Design 2003
Fibres
G.651 Characteristics of a 50/125 m multimode graded index optical fibre cable 02/98
G.652 Characteristics of a single-mode optical fibre cable 10/00
G.653 Characteristics of a dispersion-shifted single mode optical fibre cable 10/00
G.654 Characteristics of a cut-off shifted single-mode fibre cable 10/00
G.655 Characteristics of a non-zero dispersion shifted single-mode optical fibre cable 10/00
Components & Subsystems
G.661 Definition and test methods for the relevant generic parameters of optical amplifier
devices and subsystems 10/98
G.662 Generic characteristics of optical fibre amplifier devices and subsystems 10/98
G.663 Application related aspects of optical fibre amplifier devices and sub-systems 04/00
G.671 Transmission characteristics of passive optical components 02/01

Figure 5 Boundary of an optical transport Client trail
network and client-server relationship
Client layer
Interlayer Interlayer
adaption adaption
Optical channel trail
Optical channel layer

Optical multiplex section layer
Optical transmission section layer
Optical transport network
Figure 6 Possible relationship of

MON and LHON
Long haul
optical
Metro network
Enterprise & core
residental Metro access
Access network Metro core

Collection, consolidation & Interconnect of collecter POPs
grooming of access traffic (Point of Presences) with the long haul
optical network
Box B OTNT: ITU-T Recommendations on the ASTN/ASON Control Plane

Requirements
G.807/Y.1302 Requirements for the Automatic Switched Transport Network (ASTN) 07/01
Architecture
G.8080/Y.1304 Architecture for the Automatic Switched Optical Network (ASON) 11/01
G.ason living list
Protocol Neutral Specifications for key signalling elements
G.7713/Y.1704 Generalised Distributed Connection Management (draft version 0.3, 06/01) 10/01
G.7713.1/Y.1704 Distributed Call and Connection Management PNNI Implementation
G.7713.2/Y.1704 Distributed Call and Connection Management GMPLS RSVP-TE Implementation
G.7713.3/Y.1704 Distributed Call and Connection Management GMPLS CR-LDP Implementation
G.7714/Y.1705 Generalised automatic discovery techniques 10/01
G.7715/Y.1706 Architecture and requirements for routing in automatically switched optical networks
2002
G.7716/Y.1707 [ASTN link connection status]
G.7717/Y.1708 Connection Admission Control Specific Protocols to realise the signalling elements
Data Communication Network (DCN)
G. 7712/Y.1703 Data Communication Network (Draft, 06/01) 10/01
G.dcn living list version 02/01

an optical transmission section (OTS) layer ment. This can be viewed as a two-step process,
network. ref [RFC3031]: i) The packets are classified into
a set of Forwarding Equivalence Classes
Within ITU-T the optical transport network (FECs); ii) Each FEC is mapped to a next hop.
(OTN) is defined as transmission of information
over optical media in a systematic manner. An In contrast to legacy IP networks, when a packet
OTN is composed of a set of optical network enters an MPLS network it is assigned a label.
elements connected by optical fibre links, able to At subsequent hops the label is used as an index
provide the functionality of transport, multiplex- into a table that specifies the packets next hop
ing, routing, management, supervision and sur- and new label. The old label is swapped with the
vivability of optical channels carrying client sig- new label, and the packet is forwarded to its next
nals. The OTN is able to transport any digital hop. The path the packet traverses is therefore
signal independent of client-specific aspects. An called a Label Switched Path (LSP). A set of
illustration is given in Figure 5. LSPs can be merged at a specific node if packets
from these LSPs are forwarded in the same man-
The term metropolitan optical network (MON) ner (e.g. over the same downstream path, with
has entered the stage in recent years. MONs are the same forwarding treatment). This is called
said to have a role distinct from the long-haul label merging. Figure 7 Similarities between
network as well as the enterprise and access net- MPLS and GMPLS for a
works. Due to the increasing traffic demands A key feature of MPLS is that once the labels Label Switched Router and
optical solutions are increasingly promoted in required for an LSP have been assigned by LSP an Optical Cross-Connect
the aggregation and region networks. A few
characteristics do however motivate for treating
MON differently than long-haul networks, see
Figure 6: Input Output
Interface Label Interface Label
MONs are inherently designed for short to 4 23 3 12
medium distances, say less than 200 km. 2 11 2 17
.. .. .. ..
Hence, topics like signal regeneration, ampli-
fication and error correction are of less impor-
tance.
MPLS Output
Input
control interfaces
Lower cost is more emphasized in combina- interfaces
tion with wide coverage.
Read Write
label label
Service developments and fast provisioning Forwarding
might be more pronounced in MONs, such as matrix
bandwidth-on-demand and awareness of ser- Read Write
vice classes. label label
Box A and Box B list ITU-T recommendations

related to OTN transport plane and Label switched router
ASON/ASTN control plane, respectively.
3 Generalised Multi-Protocol Input Output

Label Switching Interface Interface
The GMPLS is described in [RFC3471]. This 4 23 3 12
basically contains extensions to signalling for 2 11 2 17
.. .. .. ..
MPLS, needed to include time-division, wave-
length and spatial switched/divided systems, see
illustration in Figure 7. Hence, this chapter starts
with a description of MPLS before briefly GMPLS Output
Input
control interfaces
addressing the more generalised version. interfaces
3.1 MPLS the Starting Point Rx Tx

Interconnect
3.1.1 Principles and Format matrix
At the IP layer (layer 3) a router makes forward- Rx Tx
ing decisions for a packet based on information
in the IP header. The analysis of the packet
header is performed and an algorithm is exe-
cuted in each router to decide upon further treat- Optical cross-connect

set up or label distribution protocols, intermedi- (therefore it can have only local significance)
ate Label Switched Routers (LSRs) transmitted and the packet is mapped to the next hop.
by the LSP do not need to examine the content
of the data packets flowing on the LSP. Several An LSP can be considered as a path created by
labels can be placed on a packet to form a label concatenation of one or more hops, allowing a
stack that allows multiple LSPs to be tunnelled, packet to be forwarded by swapping labels from
one within the other. The outermost LSP there- an incoming to an outgoing side of the MPLS
fore becomes a tunnel that makes inner LSPs node. An MPLS path is frequently referred to as
transparent to the intermediate LSRs, and there- layer 2 1/2 in the OSI model. That is, it may be
fore simplifies the forwarding tables at these considered as a tunnel. In order to introduce a
LSRs. This feature is critical for the local repair tunnel, a header is attached to the IP packet as
approach as described later. An explicitly routed shown in Figure 8 for the Point to Point Protocol
LSP is an LSP whose path is established by (PPP) case. The MPLS architecture is described
means other than normal IP routing. This re- in [RFC3031].
quires i.a. a management system representation
of the LSPs. Referring to Figure 8, the fields in the MPLS
header can be used as follows:
The following advantages of MPLS are listed in
[RFC3031]: Label contains a 20 bit tag identifying an
LSP.
MPLS forwarding can be done by nodes not
capable of analysing the IP packet headers, or Exp contains 3 bits (originally not allocated,
not capable of analysing these headers with intended for experimentation) which can refer
sufficiently high speed. to a certain service class, e.g. in analogy to the
DiffServ classes.
Assigning a packet to an FEC, the ingress
router may use information about the packet S 1 bit indicates end of label stacking as
that goes beyond the content of the packet several labels may be stacked.
header, like the interface. Hence, assignment
to FECs can be a more involved process, with- TTL 8 bit giving the Time To Live informa-
out impacting all routers in the network. tion.
Forwarding decisions within a network may When an MPLS packet enters a Label Switching
be made depending on which ingress router a Router (LSR) a table containing information
packet used. Then a packet may be forced to Label Information Base (LIB) on further treat-
follow a particular route explicitly chosen, ment of the packet is looked up. This base may
circumventing the ordinary routing. also be referred to as the Next Hop Label For-
warding Entry (NHLFE), typically, containing
To some extent, the use of LSPs can be consid- the following information (ref. [RFC3031]):
ered as introducing tunnelling as seen from the
IP layer. That is, when an LSP is introduced an next hop of the packet;
intermediate node would not examine the IP
header information in order to decide upon the operation to perform on the packets label
proper handling of the packets arriving in that stack (replace the label at the top with another
LSP. That is, with MPLS the classification of label, pop the label stack; or, replace the label
packets into FECs is only performed at the at the top of the stack with a new label, at the
ingress to the MPLS domain. The packet is then same time push one or more new labels onto
mapped to an LSP by encapsulation of an MPLS the stack);
header. The LSP is identified locally by the
header, see Figure 8. In successive routers data link encapsulation to use when transmit-
within the MPLS domain the label is swapped ting the packet;
Label (20) Exp (3) S (1) TTL (8)
Figure 8 MPLS header and

e.g. PPP MPLS header IP header
placement in layer 2 1/2

way of encoding the label stack when trans- In several sources, the terms traffic trunk and
mitting the packet; LSP are used synonymously. A fundamental
difference between traffic trunk and LSP can be
other information relevant for forwarding observed, though. That is, a traffic trunk is an
treatment. abstract representation of traffic to which spe-
cific characteristics can be associated. An LSP
In a given LSR, the next hop LSR may be the is a description of a path in the network through
same LSR, implying that the top level label which the traffic traverses.
should be popped and the packet forwarded to
itself, allowing for more forwarding decisions. Trunks having the same egress point may be
merged into a common tree towards the egress.
At the ingress of an MPLS domain an FEC-to- This may reduce the number of trees signifi-
NHLFE mapping is needed; that is when packets cantly. Trunks can also be aggregated by adding
arrive without an MPLS label. a new label to the stack for each trunk (that is,
bundling the trunks into a single path/tunnel).
Within an MPLS domain an incoming label
mapping is executed, mapping the packet onto Designing an MPLS network on top of a phys-
a set of NHLFEs. ical network could be looked upon as relating
two graphs to each other:
MPLS can operate on a label stack. Operations
on this stack are push, pop and swap. This can Physical graph, G = (V, E, c) is a capacitated
be used to merge and split traffic streams. The graph depicting the physical topology of the
push operation adds a new label at the top of the network. V is the set of nodes in the network
stack and the pop operation removes one label and E is the set of links. For v to w in V, (v, w)
from the stack. The MPLS stack functionality represents the link in E when v and w are
can be used to aggregate traffic trunks. A com- directly connected under G. c indicates the set
mon label is added to the stack of labels. The of capacity and other constraints associated
result is an aggregated trunk. When this MPLS with E and V.
path is terminated the result will be a splitting
(de-aggregation) of the aggregated trunk into MPLS graph, H = (U, F, d), where U is a sub-
its individual components. Two trunks can be set of V representing the LSRs; that is the set
aggregated in this way if they share a portion of of LSRs that are endpoint of at least one LSP.
their path. Hence, MPLS can provide hierarchi- F is the set of LSPs. For x and y in U, (x, y) is
cal forwarding, potentially becoming an impor- in F if there is an LSP going from x to y. d
tant feature. A consequence may be that the tran- represents the set of demands and restrictions
sit provider need not carry global routing infor- associated with F.
mation, thus making the MPLS network more
stable and scalable than a full-blown routed net- The fundamental problem of designing an MPLS
work. network is to relate the two graphs such that an
objective function is optimised.
3.1.2 Traffic Engineering and MPLS
When utilising MPLS with Traffic Engineering, One of the requirements from Traffic Engineer-
a number of mapping relations is asked for: ing is to be able to reroute an LSP under a num-
ber of conditions (failure, better route available,
Mapping packets onto FECs. An FEC com- etc.). It is desirable that this is done without dis-
poses a group of packets to be forwarded over turbing the traffic flows. This could be done by
the same path with the same forwarding treat- establishing the new LSP before the old/existing
ment. In order to carry out this mapping fields LSP is released, which is called make-before-
in the IP packet are examined. break. In case the existing and new LSP compete
for the same resources, particular concerns have
Mapping FECs onto traffic trunks. A traffic to be made, also considered by the admission
trunk is an aggregation of traffic flows of the control.
same class. A traffic trunk can again be routed
(placed inside an LSP; i.e. a traffic trunk is In addition to attributes related to traffic trunks,
only given for one LSP and not a sequence of see Box C, some attributes are also related to
LSPs). resources (frequently thought of as the links).
These attributes are ([RFC2702]):
Mapping traffic trunks onto LSPs.
Maximum allocation multiplier attribute. The
Mapping LSPs onto links in the physical net- value of this attribute tells what proportion of
work. the link and buffer capacity is available. Then,

Basic operations on traffic trunks are
Box C Attributes characterising traffic trunks are ([RFC2702]): ([RFC2702]):
Traffic parameter attributes. These are used to describe the traffic flows (the
FECs) transported in the traffic trunk. Relevant parameters include peak rates, Establish a traffic trunk
average rates, maximal burst size, etc. Possibly, equivalent measures could be Activate a traffic trunk, to start passing packets
applied, like the effective bandwidth. Deactivate a traffic trunk
Modify attributes for a traffic trunk
Explicit path specification attribute. An explicit path assignment for a traffic
trunk is a path that is specified through operator action (e.g. management Reroute a traffic trunk
procedures). Such a path can be completely or partially specified. Path prefer- Remove a traffic trunk.
ence rules may be associated with explicit paths, telling whether the explicit
path is mandatory (has to be followed) or optional (other paths could be In addition to these basic operations, a few more,
selected in case sufficient resources are not available on the preferred path). like policing and shaping could also be defined.
Resource class affinity attribute. This attribute can be used to specify which
resource types can be explicitly included or excluded from the path through A traffic trunk is defined as unidirectional. As
which the traffic trunk is routed. If no affinity attribute is given a dont care a bidirectional transfer capability is commonly
condition is assumed. Routing traffic trunks onto resources has to take these asked for, two traffic trunks having the same
attributes into account, matching the requirements. end-points but passing packets in opposite direc-
tions can be defined. In case these are always
Adaptivity attribute. As network state and traffic state change over time, more
optimal routes of traffic trunks could appear. Setting this attribute tells whether handled as a unit, it is called a bidirectional traf-
or not the route can be re-optimised for the traffic trunk. However, appropriate fic trunk (they are established as an atomic oper-
thresholds should be given avoiding too frequent changes of routing. ation and one may not exist without the other).
If a trunk is routed through a different physical
Load distribution attribute. In case several traffic trunks are used between the
path from the corresponding trunk in the oppo-
pair of nodes, the load distribution attribute can tell whether or not the load
site direction, the bidirectional traffic trunk is
(traffic trunk) can be distributed on these trunks. In general the packet order
should be maintained, implying that packets belonging to the same traffic flow called topological asymmetric. Otherwise, it is
are transferred on the same traffic trunk. called topological symmetric.
Priority attribute. This attribute gives the relative importance of the traffic trunk.
3.2 GMPLS Multiplexing Hierarchy
The value can be used to determine the order in which trunks are assigned to
MPLS uses labels to support forwarding of
paths under establishment and failure situations. Priorities will also be used
together with preemption.
packets. Label Switching Routers (LSRs) have
a forwarding table recognising the cells/frames
Preemption attribute. The value of this attribute tells whether or not a traffic with the labels, or the IP packet headers (at the
trunk can preempt another traffic trunk, and whether or not another traffic trunk
border of the MPLS domain). This is extended
can preempt a specific traffic trunk. This will assist to ensure that high priority
in GMPLS where the following interfaces are
traffic trunks are routed through even though the capacity is not sufficient to
given for an LSR:
handle all traffic trunks.
Resilience attribute. The resilience attribute gives the behaviour of a traffic Interfaces that recognise packet/cell/frame
trunk when faults occur along the path followed by a traffic trunk. In case of boundaries and forward the data based on the
fault the traffic trunk could be rerouted or not depending on the value of this
content in the packet or label/cell header. This
attribute. For rerouting, the constraints given (e.g. by affinity) could be observed
is referred to as Packet-Switch Capable. Ex-
or not.
amples are MPLS-capable routers and ATM
Policing attribute. The value of this attribute tells which actions to take when the switches.
traffic on the trunk is not complying (traffic is exceeding). Examples of actions
are packet dropping (rate limiting), packet tagging and packet shaping. Interfaces that forward data based on time slot
in a periodic cycle. This is referred to as Time-
Division Multiplex Capable. SDH cross-con-
nect is one example.
over-allocation could be achieved (as well Interfaces that forward data based on the
as under-allocation). wavelength. Such interfaces are referred to
as Lambda Switch Capable. An optical cross-
Resource class attributes. The attributes connect is an example.
express the resource type (e.g. thought of as
colours). These are matched with the traffic Interfaces that forward data based on physical
Figure 9 Organising labels trunk affinity attribute when finding paths space position of data. This is referred to as
hierarchically onto which the traffic trunks are routed. Fibre-Switch Capable. An optical cross-con-
nect operating on a level of single or multiple
fibres is an example.
time-division lambda fibre These can be organised in a hierarchical manner

IP packet packet-switch multiplex switch switch as shown in Figure 9 and corresponding labels
and Label Switched Paths (LSPs) defined. Then,

an LSP that starts and ends on a packet-switch (close-by) lambdas to be switched along the
capable interface can be grouped together with same route a waveband label has been devised,
other similar LSPs into a common LSP that giving the start and end label (and thereby the
starts and ends on a time-division multiplex range of wavelengths to be routed together).
interface. This LSP can be grouped together with
other similar LSPs into an LSP that starts and GMPLS allows for bi-directional LSPs to be
ends on a lambda switch capable interface, and established as an atomic activity (compared with
so forth. This is similar to a multiplexing hierar- having to set up two uni-directional LSPs for
chy. Note that all these levels may not be present MPLS). This allows for reduced set-up times,
in all cases. although one label each has to be assigned in the
different directions.
Compared with MPLS, the GMPLS introduces
additional interface types. The formats of the Link-related protection can also be carried by the
labels on the interfaces are given in [RFC3471]. signalling field called Protection Information.
The use of this field is optional. The protection
Motivations for combining solutions for MPLS, capabilities of a link may be advertised by the
in particular related to Traffic Engineering, and routing scheme. The Protection Information field
mechanisms for control plane in OXCs are: also indicates whether the LSP is primary or sec-
ondary (backup). The resources of the secondary
to provide a framework for real-time provi- LSP may be used by other LSPs until the pri-
sioning of optical channels in automatically mary LSP fails over to the secondary LSP.
switched optical networks;
The Protection Information field consists of
to foster the expedited development and three units:
deployment of a new class of versatile OXCs;
1 bit secondary flag set to 1 for a secondary
to allow the use of uniform semantics for net- LSP
work management and operation control and
hybrid networks consisting of OXCs and label 25 bit reserved (set to zero)
switching routers (LSRs).
6 bit link flags indicates protection type (e.g.
A particular emphasis may be placed on support 1 + 1, 1 : 1, shared, unprotected)
of various protection and restoration schemes.
4 Single-Layer Survivability
3.3 GMPLS Signalling Functional Four options exist for a two-layer structure if
Description each of the layers is considered in isolation:
Extensions to MPLS signalling to support i) no survivability mechanism; ii) survivability
GMPLS are described in [RFC3471]. A general- mechanisms in the lower layer; iii) survivability
ized label contains sufficient information for a mechanisms in the upper layer; and iv) surviv-
receiving node to program its cross connect ability mechanisms both in the lower and upper
fabric. Between nodes generalized labels are layer, but without any coordination. The first
exchanged in order to allow these nodes to know option is of little relevance for this discussion.
how to handle the information attached with Factors to be considered for the others are
these labels in the next phase. Generalized label (adapted from [Coll02]):
requests are used consisting of
Lower layer survivability only: A basic advan-
8 bit LSP encoding type: gives the LSP types, tage for having recovery mechanisms in the
such as 1 = packet, 2 = Ethernet, 5 = SDH, lower layer is that simple root failure has to
8 = lambda, 9 = fibre be treated and recovery actions are performed
on the coarsest granularity. This results in the
8 bit switching type: gives the type of switch- lowest number of required recovery actions.
ing (packet switch, layer 2 switch, TDM Moreover, failures do not need to propagate
switch, lambda switch, fibre switch) through other layers before triggering any
recovery actions. A drawback, however, is
16 bit Generalized payload identifier: gives that failure in upper layers may not be re-
the type of client layer to the LSP (IP, MPLS, solved. In addition, when a failure occurs in a
Ethernet, SDH, lambda, fibre) lower layer node, this layer recovers affected
traffic where this node is involved, possibly
The label itself depends on the link type it is leaving upper layer nodes in isolation (see
referring to. For fibre and lambda a 32 bit label Figure 10).
has been described being significant between
two neighbour nodes. Allowing for a range of

Figure 10 Only recovery d
b f
mechanisms in the lower layer
may result in isolated upper e
layer nodes a c
B
D F
E
C
A
Figure 11 Single failure at

b d f
lower layer may result in
several failure indications at
upper layer a c e
B
D F
E
C
A
Upper layer survivability only: An advantage only layer taking any recovery actions. Hence,
of upper layer survivability is that it can han- a failure in a higher layer node will then be
dle node or higher layer failures more easily. detected and dealt with in that layer. A situa-
A disadvantage is that several recovery action depicted in Figure 10 may still not be
tions may be needed because of finer granular- handled. Another variant is survivability at the
ity of the traffic flows in the upper layer, see highest possible layer. This addresses the situ-
Figure 11. On the other hand, this finer granu- ations where traffic flows are injected at dif-
larity allows for a differentiation of traffic ferent layers (e.g. combination of optical
flows, both in speed of recovery and decision channels, SDH-based leased lines and IP traf-
on whether or not recovery actions should be fic flows). A recovery strategy could be to
activated. In some cases finer granularity deal with failures at the layers where the dif-
might also lead to more efficient capacity ferent traffic flows are injected.
usage. One cause of this is that aggregated
flows (at lower layer), poorly filled with 5 Multi-Layer Survivability
working traffic flows, may have quite a lot of A common design goal for a network with multi-
spare resources. A second cause is that finer ple technological layers is to provide the desired
granularity allows distributing flows on more level of service in the most cost-effective man-
alternative paths. ner. Multi-layer survivability may allow the
optimisation of spare resources through the
Single layer survivability combinations: improvement of resource utilization by sharing
A number of variants considering recovery spare capacity across different layers. Coordina-
mechanisms in different layers can also be tion during recovery among different network
included under the single layer survivability. layers might necessitate the development of a
The argument is that for each failure scenario, vertical hierarchy. The benefits of proving sur-
the responsibility to recover traffic is the situ- vivability mechanisms at multiple layers and the
ation on only one layer. A variant to the above optimisation to the overall approach must be
is survivability at the lowest detecting layer. weighed with the associated cost and service
That is in a multi-layer configuration, the impacts.
(single) layer detecting the (root) failure is the

A default coordination mechanism for inter-layer install recovery mechanisms in several layers,
interaction could be the use of nested timers and but without any coordination between them.
current SDH fault monitoring, as has been done An advantage is that it is simple to install and
traditionally for backward compatibility. Thus, operate. A main disadvantage is that each
when lower-layer recovery happens in a longer recovery mechanism requires spare resources.
time period than higher-layer recovery, a hold- This implies that spare resources are seen for
off timer is utilized to avoid contention between each layer (and the overall situation could be
the different single-layer survivability schemes. fairly low resource utilisation). Moreover,
In other words, multiplayer interaction is more extra traffic (that is pre-emptable and
addressed by having successively higher multi- unprotected) could be disrupted during a fail-
plexing levels operate at a protection/restoration ure. An example (Figure 12) is where the
time scale greater than the net lowest layer. This lower layer switches traffic from a working
can impact the overall time to restore service. link to a recovery path and hence pre-empts
For example, if SDH protection switching is extra traffic on that link. However, the upper
used, MPLS recovery timers must wait until layer may gradually route the traffic on
SDH has had time to switch. Setting such timers another path (possibly also pre-empting addi-
involves a trade-off between rapid recovery and tional traffic along that path). For the example
creation of a race condition where multiple lay- in Figure 12 link A D fails and the lower
ers are responding to the same fault, potentially layer moves traffic to links A B D. How-
allocating resources in an inefficient manner. ever, the upper layer may find that another
path is better so the traffic is effectively
In other configurations where the lower layer moved onto links A C E (potentially re-
does not have a restoration capability or is not sulting in poor utilisation of links A B D).
expected to protect, say an unprotected SDH lin-
ear circuit, then there must be a mechanism for Sequential approach: Realising that some
the lower layer to trigger the higher layer to take coordination would likely result in improved
recovery action immediately. This difference utilisation, different options are investigated.
in network configuration means that implemen- The next level of complexity is seen where the
tations must allow for adjustment of hold-off responsibility for recovery is handed over to
timer values and/or a means for the lower layer the next layer when it is clear that the current
to immediately indicate to a higher layer that the layer is not able to fulfil the recovery task. In
fault has occurred so that the higher layer can principle any sequence of layers can be
take restoration or protection actions. thought of, however, two obvious ones are
bottom-up and top-down. For the former the
Moreover, faults at higher layers should not lowest detecting layer starts the recovery and
trigger restoration or protection actions at lower traffic that cannot be resolved by this layer
layers. (e.g. due to capacity shortage) will be restored
by a higher layer. This has the advantage that
The fact that there are a number of layers present recovery actions are taken at the appropriate
raises the question of whether these could be granularity and more complex secondary fail-
utilised in a coordinated manner to improve ures are only treated when needed. For the
the recovery schemes. Three main classes of top-down approach the upper layer starts the
schemes are: recovery action and only if that layer is unable
to restore all traffic are actions at lower layers
Uncoordinated approach: This may be con- initiated. An advantage of this is that higher
sidered as the simplest approach, that is to layers may differentiate traffic due to require-
b d
e
a c
B
D
E Figure 12 Illustration of
uncoordinated recovery; lower
C layer routes on neighbour link,
A while upper layer may find that
another path is better

Table 1 Overview of some Survivability strategy
recovery strategies (from Factor Preferred
value
[Coll02]) Bottom layer Bottom-up Top layer Integrated
approach
Switching Coarse Coarse Fine Coarse Coarse

granularity
Failure Simple Simple Complex Simple Simple

scenario
Recovery Yes Yes No Yes Yes

close to root
Capabilities, Low High High High High

flexibility
Failure Low High High High High

coverage
Coordination Low High Low Low Low

management
Resources Low High Low Low/High Low
ments and hence restore high-priority traffic shaping spare capacity among pre-established
first. When coordinating multiple layers a few backup paths. Then low-priority traffic can be
mechanisms have to be in place to avoid that dropped, e.g. by applying DiffServ.
the different layers take steps destroying for
each other. One proposal is to introduce a In the case of peer models becoming a reality
hold-off timer. This timer is set when a layer in the longer term, an integrated survivability
starts to restore traffic. If the timer expires and approach may be likely. This is due to the single
the traffic is not adequately restored, the next integrated control plane in the peer model.
layer initiates its actions. A disadvantage by
this is that recovery actions at the next layer Automation of wavelength establishment/release
would always be delayed independent of the does not require a fixed logical IP/MPLS net-
failure. To try to compensate for this an ex- work design. Hence, the logical network design
plicit recovery token signal can be exchanged can be re-optimised during a failure situation.
between layers. Such a signal has to be in- Then, if a router fails, an automatic reconfigura-
cluded in the standards covering the interfaces tion of the logical IP/MPLS network could be
between these layers. undertaken to restore the traffic handling capa-
bility (instead of traditional rerouting or ensuring
Integrated approach: This is based on a com- bi-connected IP/MPLS network designs).
mon integrated recovery scheme for the stack
of layers. Hence a full overview of all the lay- A disadvantage of current IP/MPLS networks
ers is needed in order to decide which layer is that failure detection is based on the periodic
and which actions to take. In principle, this exchange of Hello messages between adjacent
approach is the more flexible one, however routers. If no Hellos are received through an
the algorithms have to be devised and imple- interface, the conclusion is that the opposite side
mented. of the interface in unreachable. This does not
allow for a separation of route failures and link
Table 1 summarises a number of recovery strate- failures. Another concern with Hello messages is
gies by a number of selected factors. The col- the detection time. Currently this may be in the
umn called preferred value indicates a typical range of 10 s and a failure is declared after miss-
(but not necessarily needed) value. ing 4 Hello messages. This would allow the opti-
cal/SDH to spend plenty of time restoring the
The aspects discussed above can be applied on traffic handling capability without even inform-
the configuration of IP/MPLS over optical/SDH ing the IP/MPLS layer. As more capacity is
networks. In principle MPLS is able to provide introduced on the links (say 2.5 Gbit/s and
fairly fast protection switching. Therefore, one 10 Gbit/s), more frequent Hello messages can
may choose to promote recovery at the IP/MPLS be exchanged without stealing too much of the
layer (hence promoting survivability at the top link capacity.
layer). An argument backing this is that IP/MPLS
may allow for less spare resources. This may When discussing gains and drawbacks by utilis-
come from packet switching being suitable for ing IP/MPLS and optics survivability mecha-

nisms, the observations would naturally depend switchover congestion may lead to longer queue-
on the network configuration looked at. Basi- ing delays and therefore back-off by the TCP
cally, using cost of links and nodes, it is ex- sources. Depending on the duration of the
pected that dedicated protection is more expen- switchover time, a number of the TCP sources
sive than shared protection. Moreover, this dif- on the failed MPLS path may have backed off.
ference may be bigger for local protection. This This may lead to a more gentle traffic handling
could be caused by the fact that sharing two (or of the TCP sources running on the restoring link.
more) backup paths using the same resource is An observation in [Coll02] is that when the
only possible if the two working segments do switchover delays are in the range 0 250 ms,
not overlap (here a segment is a path in the case the differences in the goodput do not differ all
of path protection or a local loop-back of a link that much, especially when the number of TCP
in the case of link protection and two links in flows is large.
case of node protection). For local protection,
the working segments are in general shorter than Typically, a restoration time requirement of 50
for path protection or local loop-back implying 60 ms is seen referring to SDH. This is advo-
lower probability of working segments overlap- cated by the fact that voice calls (circuit switched)
ping and hence a higher probability that sharing may potentially be dropped if restoration takes
between the two backup paths is indeed allowed. longer than this. As SDH may detect a transmis-
This is comparing one or two links versus a sion failure in the range of 2.3 100 s, there
complete path. Hence, the relative difference may be little assistance in the optical layer sup-
between dedicated and shared protection in porting restoration faster than 1 ms. The SDH
terms of capacity requirements will be more sub- layer would already potentially detect the failure
stantial for local protection than for path protec- and any further actions by the SDH could be
tion or local loop-back. held back until the optical layer has finished its
actions (as described for sequential and inte-
It is also expected that the network topology has grated approach above).
a significant impact, in particular for the local
protection. That is, the topology with the small- 6 Optical Layer Survivability
est nodal degrees suffers the most from dedi- A protection taxonomy referring to the optical
cated protection (node degree = number of links layer is given in Table 2. The terms refer to
divided by number of nodes). This can be caused SONET. A protection scheme either operates at
by the fact that as a topology becomes sparser, i) an aggregate WDM level, called line layer or
backup paths tend to become longer and more optical multiplexer section (OMS) layer; or, ii)
overlapping (the extreme is a ring topology). at individual wavelength level, called path layer
This explains why dedicated protection is rela- or optical channel (OCh). As shown, protection
tively more costly in sparse networks than in is either dedicated or shared. When rings are
dense networks. used, these are called dedicated protection rings
(DPRings) and shared protection rings
An overall observation, from a capacity point (SPRings), respectively. Again in SONET,
of view, is that MPLS recovery may allow for DPRings are called optical unidirectional path-
fewer resources. On the other hand, the switch- switched rings (OUPSRs) or optical unidirec-
over delay would likely be longer. This may tional line-switched rings (OULSRs), depending
have a severe impact on the traffic flows. An on whether they operate at the path layer or the
example of this is when a number of TCP flows line layer. Shared protection rings are called
are carried by an MPLS path that is switched optical bidirectional line/path switched rings
over and share underlying resources with (OBLSRs or OBPSRs, respectively). Table 2 Protection taxonomy
another MPLS path. Just before the failure, the for optical layer survivability,
TCP sources on both MPLS paths may be trans- Ring-based survivability involves the use of bi- referring to SONET terms
mitting at high rates. Then immediately after the directional line switched rings (BLSRs) or unidi- (from [Gers00])
Line layer (OMS) Path layer (OCh)

Network topology
Dedicated protection Shared protection Dedicated protection Shared protection
Point-to-point 1+1 linear APS 1:1/1:N linear APS 1+1 linear APS 1:1/1:N linear APS
(OMS-DP) (OMS-SP) (OCh-DP) (OCh-SP)
Ring OULSR OBLSR OUPSR OBPSR

(OMS-DPRing) (OMS-SPRing) (OCh-DPRing) (OCh-SPRing)
Mesh (no standard term) (no standard term) 1+1 OSNCP 1:N mesh protection

rectional path-switched rings (UPSRs) as self- for fast restoration. Moreover, a few cases (e.g.
protecting transmission system overlaid on the ref. [Stam00]) show that p-cycles can be about
network topology. More recent variants of as capacity-efficient as mesh networks. Com-
UPSR- and BLSR-based transport networks are pared to rings, p-cycles also de-couple the rout-
optical path protection ring (OPPR) and optical ing of working and protecting paths by the fact
shared protection ring (OSPR) variants in a that the working paths may be routed along the
WDM context. An important factor is that rings shortest paths, while protecting paths follow the
may use simple switching mechanisms, which cycle.
permits restoration in about 50 60 ms, although
by nature they require 100 % redundancy. In The reason for p-cycles being as fast as rings is
particular, the BLSR uses a working-to-protec- that only two traffic-substituting connections are
tion loop-back switching mechanisms at the two made for any working path failure to be restored.
nodes adjacent to a failure. In essence, this is The two end nodes perform the switching only
identical to the switching mechanism employed by transmitting a control signal as they know in
for p-cycles (in WDM or SDH). According to advance which working-to-protection functions
[Stam00], in conventional multi-ring network are needed for any given failure. p-cycles can be
designs, where the working fibres of channel introduced, in principle, on optical, SDH, and
groups are not fully utilizable, effective protec- MPLS layers.
tion-to-working capacity ratios (capacity redun-
dancy) can be 200 300 %. This may show that It is quite common to find mesh physical topolo-
rings allow fast restoration but may not be gies in the backbone networks. The SDH layer
capacity-efficient. placed on that, however, is in most cases pro-
tected in the form of rings. These rings are inter-
Mesh-based survivability is more capacity-effi- connected to provide an overall network connec-
cient due to the fact that each unit of spare tivity and protection. It is a question whether
capacity can be used in several manners, i.e. for these schemes are less bandwidth efficient to
several failures. Hence, overall performance protect a physical meshed network. Efficiency
close to the ideal maximum-flow routing-effi- improvements in the area 20 60 % have been
ciency can be realised. Mesh restoration has tra- demonstrated by introducing mesh protection
ditionally been based on cross-connect systems schemes, although these observations naturally
embedded in a mesh-like set of point-to-point depend on the topology, traffic patterns and pro-
transmission systems. This is one factor for tection scheme. Taxonomy of mesh protection
slower restoration compared with ring systems. variations including trade-offs between them as
In addition comes the more general nature of depicted in Figure 13.
solving a capacitated multiple-path rerouting
problem for mesh networks compared with Deploying a robust mesh protection scheme, a
rings. number of challenges have to be met (adapted
from [Gers00]):
Hence, mesh and ring both have their strengths
and weaknesses. Mesh networks tend to be app- A fast signalling mechanism is needed to
lied in the long-haul networks where capacity on propagate information about failure events to
transmission links is expensive. Rings tend to be relevant nodes so they can reconfigure the
more cost-efficient in the metro areas where cost switches to accommodate the recovery path.
is dominated by terminal equipment and not so
much with the link and through-connect (due to The recovery routing tables become large if
short links). A method named p-cycle has been pre-computed routes are used to expedite the
promoted trying to merge the advantages of ring protection and full flexibility in the choice of
and mesh. p-cycles are based on the formation of routes for protection is requested (to maximise
closed paths (elementary cycles in graph theo- the bandwidth efficiency). In particular main-
retic terms), called p-cycles, in the protection taining large sets of routing tables is a non-
capacity of a mesh-restorable network. They are trivial task, e.g. with regard to ensuring con-
formed in advance of any failure, out of the pre- sistency.
viously unconnected protection capacity units of
a restorable network. A p-cycle protects both on- Failure propagation may be easier in a mesh
cycle and straddling failures (the latter refers to network than in a ring topology. Larger por-
paths not following the cycle during normal tions of a network may therefore be affected.
working situation, but may follow the cycle dur- This is essential to consider as the protection
ing failure of any link not being part of the scheme in itself may be inadequate, e.g. not
cycle). This implies that protection capacity on a considering all failure configurations that can
p-cycle is more widely accessible, i.e. more occur (such as multiple failures), operator
highly shared for restoration compared with tra- errors, software bugs, and so forth.
ditional rings. Still, the cycle formation allows

Protection or a mesh network Figure 13 Taxonomy and
trade-offs for mesh protection
Decompose into protection domains variants (from [Gers00])
Pure rings
p-cycles
Dual protection trees
Protection mesh as a whole
On-the-fly route computation

Signalling
complexity vs. Distributed
Protection processing Centralised route computation and coordination
routing table level vs.
management Route Route computation and coordination at ends of failed
flexibility line
Distrbuted route computation and coordination at
ends of path
Pre-computed routes
Bandwidth
efficiency vs. 1+1 path protection
Size of
protection Protection route per wavelength
routing tables
Protection route per failure
Making full use of the bandwidth efficiency ered as the IP/MPLS combined layer. However,
may require sophisticated design algorithms this description can also be further generalised.
and network planning tools, themselves add-
ing to the complexity. Referring to configurations with clients over
optical networks, a few protection schemes are
Different variants of mesh protection schemes depicted in Figure 14. In part a) clients have
may imply that standardisation and interoper- connections to duplicated transponder/optical
ability are harder to achieve. systems. Introducing a spare client as shown in
part b) allows for fewer transponders, although
Still comparing mesh and ring topologies, a clients have to coordinate any switchover. The
mesh scheme allows the entire network to be taxonomy given in Figure 14 does not concern
managed in an integrated manner. Moreover, itself with internal node implementation. Hence,
interconnecting rings avoiding the single point there could be varying cost levels for imple-
of failure adds complexity to this scheme. menting the different protection schemes. As
transponders may be relatively expensive (most
Introducing wavelength conversion may consid- of the WDM terminal costs come from the
erably improve the utilisation of available wave- transponders), there may be a benefit by having
lengths. Avoiding full conversion, limited con- an arrangement with fewer transponders. This
version capability is proposed to reduce the cost may favour a post-split arrangement versus a
of optical nodes. However, this may increase the pre-split arrangement as shown in part d) and c),
protection complexity. For example, without respectively.
conversion in a ring, the same wavelength could
simply be looped around the other way, while One approach to achieve fewer transponders is
when wavelength conversion is introduced one to introduce spare arrangements. Three cases are
has to check that the wavelength has not been depicted in Figure 15. These are mostly for pro-
used for other purposes on any hops. tecting failures in transponders. In part a) a
transponder on a designated wavelength is used
7 Survivability Issues for as spare. In case of failure of a working trans-
Optical/WDM and Client ponder, the signal link is routed to the spare
Networks transponder through its switch. The same opera-
When discussing options for multi-layered net- tion takes place at the receiver end, where the
works, one may apply an approach of treating signal is routed to the corresponding client
the neighbouring layers iteratively. Hence, one equipment. When a tuneable laser/detector is
may start by examining the options for two lay- used in the spare transponder no designated
ers. Here the lower layer can be considered as wavelength for spare is needed. For a failure, the
the optics, while the higher layer can be consid- signal from the client equipment is routed to the

Client equipment Transponder Transponder Client equipment
Client equipment Client equipment
Demux
Transponder Transponder
Mux
Demux
Mux
a) Protection using 1 + 1 client scheme
Demux
Mux
Spare client Transponder Transponder Spare client
b) Protection by 1: N client protection scheme and 1 + 1 optical line protection
Splitter
Splitter

Demux
Mux
Demux
Mux
c) Protection by transponder pre-split

Splitter
Splitter

Demux
Mux
Demux
Mux
d) Protection by transponder post-split
Figure 14 Illustrating a few protection arrangements for client and optical layers

Splitter
Splitter
Demux
Mux
Spare transp. Spare transp.
a) Fixed spare
Splitter
Splitter
Demux
Transponder Mux Transponder
Tunable Tunable
b) Tunable spare

Demux
Mux
c) Provisionable spare without additional transponder
tuneable spare, which is then tuned to the wave- tection Switching (APS) in SDH. In a mesh net- Figure 15 Three transponder
length of the previously used transponder. A work the coordination grows in complexity and configurations
similar operation may take place at the receiver may require a few additional wavelengths for
end, although this operation is not mandatory, protection only.
and may also allow for additional protection. In
part c) a series of small optical switches is intro- Tuneable spares do not need the coordination
duced. Under normal situation the switches con- between the two ends because as the same wave-
nect the signal straight through, although in case length is used, the receiver may see no changes.
of failure, the signal is routed to the next switch- A downside is that tuneable lasers are consid-
ing element until a spare transponder is found. ered expensive. Moreover, the scheme cannot
Hence, no dedicated transponder is spare. How- handle other failures in the network related to
ever, for this part the overall traffic handling wavelengths, such as failures in wavelength
capability may be reduced during a failure situa- filters.
tion.
For the provisionable case, flexibility is allowed,
The three schemes may be compared as follows: e.g. by defining protection groups. Multiple
The fixed spare scheme is the least demanding transponder failures may also be supported,
scheme from the hardware and software point of although then with further decrease in traffic
view. A complicating factor is that when a fail- handling capability or introducing more spare
ure occurs, both ends have to coordinate the transponders. The small optical switches must
switch to the protect wavelength using a fast sig- be controlled by a signalling protocol, which
nalling protocol, e.g. similar to Automatic Pro- may be adding complexity.

Figure 16 Qualitative
illustration of different Cost
protection schemes (adapted Client 1 + 1 over protected wavelengths
from [Gers00])
F(cost of protected wavelengths)
Client 1 + 1 over unprotected wavelengths
F(client protection cost)
Optical uni-directional path switched rings/ 1 + 1 pre-split
x2
Optical uni-directional path switched rings/ 1 + 1 post-split
f(traffic)
Optical uni-directional path switched rings/ 1 + 1 post-split

f(mux cost)
Optical bi-directional line switched rings

f(traffic)
Optical bi-directional line switched rings
Unprotected client over unprotected wavelengths
Availability (including probability of traffic affecting failure)
Introducing more advanced protection schemes Timing bounds for service restoration to
will also add to the cost. However, the cost of support voice call cut-off (140 ms to 2 s),
the implementations within a node has to be bal- protocol timer requirements in premium
anced with the transport cost, and the achieved data services, and missing critical applica-
availability. A qualitative comparison is shown tions
in Figure 16.
Use of restoration priority for service dif-
A set of requirements on survivability and hier- ferentiation
archy, both current and near-term, are described
in [RFC3386]. These are summarised as: B. Hierarchy requirements
A. Survivability requirements: B.1 Horizontally oriented hierarchy (intra-

Need to define a small set of interoperable domain)
survivability approaches in packet and Ability to set up many LSPs in a service
non-packet networks provider network with hierarchical IGP,
for the support of layer 2 and layer 2 VPN
Suggested survivability mechanisms services
including
Requirements for multi-area traffic engi-
- 1:1 path protection with pre-established neering need to be set up to provide guid-
backup capacity (non-shared) ance for any necessary protocol extensions.
- 1:1 path protection with pre-planned B.2 Vertically oriented hierarchy; the following
backup capacity (shared) functionality for survivability is common on
most routing equipment today:
- local restoration with repairs in proxim-
ity to the network fault Near-term need in some loose form of
coordination and communication based on
- path restoration through source-based the use of nested hold-off timers, instead
rerouting of direct exchange of signalling and rout-
ing between vertical layers

Means for an upper layer to immediately is, if no packets are sent on to a protection path,
begin recovery actions in the event that a no bandwidth is consumed. This is not the case
lower layer is not configured to perform in traditional transmission networks like most
recovery optical or TDM networks, where path establish-
ment and resource reservation cannot be de-
C. Survivability requirements in horizontal coupled.
hierarchy:
Protection of end-to-end connection is In the 1:n protection architecture, traffic is nor-
based on a concatenated set of connec- mally sent on the working entities. When multi-
tions, each protected within their area. ple working entities have failed simultaneously,
only one of them can be restored by the common
Mechanisms for connection routing may protection entity. This contention could be re-
include: i) a network element that partici- solved by assigning a different pre-emptive pri-
pates on both sides of boundary; ii) a route ority to each working entity. As in the 1:1 case,
server. the protection entity can optionally be suited to
carry pre-emptable traffic in normal operation.
Need for inter-area signalling of surviv-
ability information i) to enable a least While the m:n architecture can improve system
common denominator survivability availability with a small cost increase, it has
mechanism at the boundary; ii) to convey rarely been implemented or standardised.
the success of failure of the service
restoration action, e.g. if a part of a con- When compared with protection mechanisms,
nection is down on one side of a bound- restoration mechanisms are generally more fru-
ary, there is no need for the other side gal as no resources are committed until after the
to recover from failures. fault occurs and the location of the failure is
known. However, restoration mechanisms are
In a survivable network design, spare capacity inherently slower, since more must be done fol-
and diversity must be built into the network from lowing the detection of a fault. Also, the time it
the beginning to support some degree of self- takes for the dynamic selection and establish-
healing whenever failures occur. A common ment of alternate paths may vary, depending on
strategy is to associate each working entity with the amount of traffic and connections to be
a protection entity having either dedicated restored. It is also influenced by the network
resources or shared resources that are pre- topology, technology employed, and the severity
reserved or reserved-on-demand. According to of the fault. As a result, restoration time tends to
the methods of setting up a protection entity, dif- be more variable than the protection switch time
ferent approaches to providing survivability can needed with pre-selected protection entities.
be classified. Generally, protection techniques Hence, in using restoration mechanisms, it is
are based on having a dedicated protection entity essential to use restoration priority to ensure that
set up prior to failure. Such is not the case in service objectives are met cost-effectively.
restoration techniques, which mainly rely on the
use of spare capacity in the network. Hence, in Once the network routing algorithms have con-
terms of trade-offs, protection techniques usually verged after fault, it may be preferable in some
offer fast recovery from failure with enhanced cases to re-optimise the network by performing a
availability, while restoration techniques usually reroute based on the current state of the network
achieve better resource utilization. and network policies.
A 1+1 protection architecture is rather expensive [RFC3386] states some typical requirements on
since resource duplication is required for the protection switch time or restoration time:
working and protection entities. It is generally
used for specific services that need a very high Best effort data: recovery of network connec-
availability. tivity by rerouting at the IP layer would be
sufficient
A 1:1 architecture is inherently slower in recov-
ering from failure than a 1+1 architecture since Premium data service: need to meet TCP
communication between both ends of the protec- timeout or application protocol timer require-
tion domain are required to perform the switch- ments
over operation. An advantage is that the protec-
tion entity can optionally be used to carry low- Voice: call cut-off is in the range of 140 ms to
priority extra traffic in normal operation, if traf- 2 s (the time that a person waits after interrup-
fic pre-emption is allowed. Packet networks can tion of the speech path before hanging up or
pre-establish a protection path for later use with the time that a telephone switch will discon-
pre-planned but not pre-reserved capacity. That nect a call)

Other real-time services (e.g. streaming, fax) SDH/WDM multi-layer configuration, as
where an interruption would cause the session depicted in Figure 17. It should be noted that
to terminate typically there is a range of clients to each of
the layers. Considering the arrangement as seen
Mission-critical applications that cannot toler- from the IP point of view, [Coll02] states two
ate even brief interruptions, for example real- main drawbacks by carrying IP over ATM:
time financial transactions.
The ATM cell size; introducing the so-called
It then goes on to propose some timing bound cell tax (overhead of ATM cell header and
for different survivability mechanisms (times adaptation layer), and that many ATM cells
excluding signal propagation): must be transported and processed to carry a
single IP packet.
1:1 path protection with
pre-established capacity 100 500 ms The additional (ATM) layer to be maintained
and managed.
1:1 path protection with
pre-planned capacity 100 750 ms Naturally, ATM also has a number of benefits;
such as its connection orientation (when needed)
local restoration 50 ms providing extra support for traffic engineering,
and decoupling of control plane (e.g. routing)
path restoration 15s and user plane (e.g. forwarding).
In order to ensure that requirements for different However, capabilities such as Traffic Engineer-
applications can be met, restoration priority ing defined for IP imply that several of the argu-
should be implemented. This determines the ments in favour of using ATM can be achieved
order in which connections are restored. by other means. Introducing MPLS has been
promoted to cover several of the objectives.
8 IP over Optics
The continuing growth of IP-based traffic (both Furthermore, as traffic intensities and router
Internet-related and for other IP-based services), interface rates increase, dedicated wavelengths
advocates more streamlining of the network lay- could be likely, allowing the IP/MPLS traffic
ers. Hence, IP-over-optics with few intermediate flows to bypass both ATM and SDH networks.
network layers has been promoted by several (A distinction should be made between the net-
parties. The fact that available capacity on a works and the framing protocols, e.g. SDH
fibre cable increases, a single cable cut may framing may still be applied for a wavelength
affect huge traffic volumes. Therefore special between two IP routers.) This also means that
attention is paid to the survivability of such net- cross-connection functionality offered by the
works. SDH network may not be utilised for such
Figure 17 Illustration of an IP/MPLS traffic flows. On the other hand, opti-
IP/ATM/SDH/WDM As claimed in [Coll02] a starting point for sev- cal network elements might be introduced to
configuration eral of the incumbent operators is an IP/ATM/ provide such functionality if needed.
Another factor is that most transport network

configurations so far tend to be rather static.
That is, the connections are commonly estab-
lished/changed by manual operations through a
IP packets
IP IP
management system. In case the IP/MPLS traffic
flows show highly dynamic behaviour, more
automatic procedures are asked for in order to
adapt the transport capacity in the different
ATM cells
directions to the variations in the IP/MPLS net-
ATM ATM
work. Much research effort is currently put into
solving how this provisioning process can be
SDH STM-N automated. This would expectedly require some
control plane functions.
SDH SDH
Broadly speaking, two main (opposite) models
have emerged for an Automatic Switched Opti-
wavelengths cal Network (ASON). One is an overlay of an
WDM WDM Automatic Switched (optical) Transport Net-
work (ASTN) (ASTN can be seen as a generali-
fibre
sation of ASON). In the overlay model both the

transport network and its client networks have a A consequence of this is that links between IP
separated and independent control plane. This is routers should be disjoint in order to handle fail-
described in ITU-T recommendations. ures. In this perspective, one can consider the IP
network as consisting of a set of logical links
The IETF seems to target a peer model with the that are mapped onto a set of physical links in
Generalised MPLS (GMPLS) concept. This con- the fibre/WDM network. In the WDM network
cept stems from MPS, where the underlying the logical links are assigned a wavelength, or
idea was that a wavelength is a label like any possibly a sequence of wavelengths if wave-
other layer and therefore the MPLS concept can length conversion is present. In this way several
be adopted in the optical domain to serve the logical links could go on the same fibre (or con-
need for fast and automatic provisioning of duit), which is not a preferred situation as a sin-
wavelengths. GMPLS is more generic in the gle cable cut could affect several links.
sense that it considers any type of label: a header
string for packet switch capable LSR (PSC- When a failure occurs in the optical equipment,
LSR), a time slot for a TDM switch capable LSR one should still have a connected topology at the
(TSC-LSR, e.g. by SDH), a wavelength for a IP layer. Hence, for every possible failure, alter-
lambda-switch capable LSR (LSC-LSR, e.g. an native operational logical links (or combinations
optical cross-connect), or a fibre in a fibre- of logical links) must exist so that alternative
switch capable LSR (FSC-LSR). These are routes can be found by the IP routing protocol.
described in Section 3.2.
A requirement may then be that the alternative
Although both client and transport networks link offers about the same delay as the normal
may have their own separate and independent operating link. For example, a voice-over-IP
(G)MPLS control planes, an integration of these application may suffer if the end-to-end delay
controls planes into a singe one seems tempting. increases by more than 50 ms after switching
This then results in the so-called peer model. over to the alternative link. This comes in addi-
The following advantages are seen for the peer tion to the fact that the primary operating link
model: usually follows the shortest path and thereby has
the shortest delay.
Avoiding duplication of control plane func-
tionality in distinct layers; Hence, designing the robust IP network, both
disjoint links as well as delay concerns have to
Avoiding the requirements of standardisation be included. The latter would naturally be
of a User-Network Interface between related to the link loads. Having two different
IP/MPLS routers and optical network compo- types of requirements commonly asks for a
nents. trade-off (or combination) to be included. The
basic problem can be formulated as an Integer
A basic drawback, however, is the fact that inte- Programming problem (integer refers to the fact
gration and compatibility among multiple client that one must choose from certain capacities and
type networks seem hard, as well as the require- that logical links must be mapped onto physical
ment that all necessary information of the optical links). Generally this is known as an NP-com-
network (e.g. the topology) will be accessible in plete problem, where delay constraints, limited
the client domain. number of wavelengths and fibre/WDM network
structure have to be considered.
9 All Protection at IP Layer
Some people claim there is a trend that most of 10 MPLS-Based Protection
the protection is located at the IP/MPLS layer. Several of todays networks are firstly built for
This is backed by arguments such as: connectivity, and secondly to support one ser-
vice class. The migration of real-time and
Protection at SDH is considered as expensive higher-priority traffic to IP networks means that
as seen by the IP/MPLS layer (one pays for a modern IP networks increasingly carry mission-
back-up capacity that is rarely used). critical business data and must therefore provide
reliable transmission. Current routing algo-
Supporting differentiated protection schemes rithms, despite being robust and survivable, can
implies that such mechanisms must be located take a substantial amount of time to recover
at the IP/MPLS layer. from a failure, which can be in the order of sev-
eral seconds to minutes and can cause serious
Several people would remove the SDH net- disruption of service in the interim. This is unac-
work as such for carrying the IP/MPLS traffic ceptable for many applications that require a
(although SDH framing may still be applied). highly reliable service, and has motivated net-
work providers to give serious consideration to
the issues of network survivability.

Multi Protocol Label Switching (MPLS) can Terms
be utilised to integrate forwarding based on The intent of protection is to protect against any
label-swapping of a local label and hence allows link or node fault on a path or a segment of a
more advanced forwarding mechanisms to be path. The protected path is commonly referred to
applied. One area of these mechanisms is QoS, as a working path. A set of paths used for carry-
where MPLS can be used to transport data reli- ing traffic on the working path in case of failure
ably and efficiently and also to support a range along it, is commonly referred to as recovery
of service classes. One of the drawbacks with paths. An LSR that can select to use the working
current IP routing algorithms is the relatively path or the recovery path is called a Path Switch
long time to recover from a fault, which is in LSR (PSL). The LSR where the working and
the order of seconds or minutes. This may well recovery paths meet has to decide which packets
cause a disruption of service as observed by the that should be forwarded. This LSR is com-
applications/users. As more business-critical monly called a Path Merge LSR (PML).
applications/usages and real-time demanding
applications are placed on IP networks, such As LSPs are unidirectional and recovery may
long recovery times may be unacceptable. It is well require notification of faults being carried
often seen that applications require recovery to the LSR responsible for switch-over to the
times down to tens of milliseconds; examples recovery path, a mechanism must be provided
being virtual leased lines, voice traffic, real for propagating fault and repair indications from
video transfers, stock exchange data. For several the point of occurrence of the fault back to that
of these applications SLA conditions have been LSR. These are called Fault Indication Signal
stated which do not allow for long recovery (FIS) and Fault Repair Signal (FRS).
times.
10.1 Recovery Principles
Generally, operators would provide the fastest A nomenclature for recovery principles is also
and best protection mechanisms that can be pro- drafted in [Shar02]. Firstly it is noted that MPLS-
vided at a reasonable cost. However, the higher based recovery refers to the ability to quickly
level of protection, the more resources con- and completely restore capabilities for serving
sumed. Hence, it is naturally expected that oper- traffic affected by a fault in an MPLS network.
ators will offer a range of service levels. MPLS- The fault may be detected at the IP layer or in
based recovery should give the flexibility to lower layers. To some extent the different terms
select the recovery mechanism, choose the gran- can be arranged in a nomenclature tree as
ularity of which traffic to protect and also to depicted in Figure 18.
choose the specific types of traffic that are pro-
tected in order to give operators more control The issues and corresponding options are:
over that trade-off.
1. Configuration of recovery:
Based on the above, one motivation for MPLS is Default recovery: no MPLS-based re-
the notion that the current IP routing algorithms covery is enabled; hence traffic must be
have limited potential for improving the recov- recovered using IP layer or lower layer
ery times. Introducing recovery mechanisms on mechanisms.
the MPLS level, or partly enhance the current
routing algorithms with MPLS aspects may alle- Recoverable: MPLS-based recovery is
viate the underlying restriction. enabled; hence one or more MPLS re-
covery paths are identified for a working
Another motivation for introducing recovery MPLS path.
mechanisms operating on the MPLS level is the
presence of MPLS in several of the IP networks. 2. Initiation of path set-up:
Moreover, MPLS is commonly promoted as a Pre-established: recovery paths are estab-
base solution for IP-based transport networks. lished prior to any failure of the working
Hence, it is natural that MPLS provides protec- path. This is also called protection
tion and restoration of the traffic that has such switching.
requirements.
Pre-qualified: a path created for other
MPLS may facilitate the convergence of net- purposes is also designated as a recovery
work functionality on a common control and path for a working path.
management plane. Further, a protection priority
could be used as a differentiating mechanism for Established on-demand: a recovery path
premium services that require high reliability. is established after a failure on its work-
ing path has been detected and notified
to the Path Switched LSR (PSL). This is
also called re-routing option.

Recovery Figure 18 Recovery terms
Default (adapted from [Shar02])
1. Configuration Recoverable
Pre-established
2. Initiation of path setup Pre-qualified
Established on-demand
Equivalent recovery path

3. Recovery equivalence
Limited recovery path
Pre-reserved
4. Initiation of resource allocation
Reserved on-demand
5. Scope of recovery Local repair

Global repair
i) Topology Alternate egress repair
Multi-layer repair
Concatenated protection domain
N:M
ii) Path mapping Split path protection
Yes
iii) Bypass tunnels No
Selected traffic recovery

iv) Recovery granularity Bundling
Dedicated
v) Recovery path resource use Extra traffic allowed
Shared resources
Path failure
Path degraded
6. Fault detection Link failure
Link degraded
Control plane messaging

7. Fault notification User plane messaging
8. Switch-over operation
Automatic
i) Recovery trigger External commands
Re-route
ii) Recovery type Protection switching
9. Post-recovery operation
Revertive mode
i) Fixed protection counterpart Non-revertive mode
Yes
ii) Dynamic protection counterpart No
iii) Restoration and notification By PSL
iv) Reverting to preferred path Yes

No
10. Performance attribute values
3. Recovery equivalence: detected and notified to the PSL and

Equivalent recovery path: recovery paths before the traffic is switched over to
are capable of replacing the working path the recovery path.
without degrading the service levels.
5. Scope of recovery:
Limited recovery path: recovery paths i) Topology:
do not replace the working path without - Local repair: node immediately
degrading the service levels. upstream of the fault is the one to initi-
ate recovery. This is either a) Link re-
4. Initiation of resource allocation: covery/restoration/recovery path is con-
Pre-reserved (applies only to protection figured to route around a certain link
switching): recovery path has reserved (working and recovery paths disjoint
resources on all hops along its route. only at a certain link); or, b) Node re-
covery/restoration: recovery path routed
Reserved on-demand: recovery path around a neighbour node (working and
reserved the required resources after recovery path disjoint at a certain node).
a failure on the working path has been

- Global repair: Point Of Repair (POR) is path, referred to a protection traffic por-
at the ingress of a path (segment), hence tion (PTP).
also covering end-to-end path recovery.
In case the recovery path is completely - Bundling: grouping several working
disjoint from the working path, this pro- paths allowing for simultaneous recov-
tects against any link and node faults. ery. The logical bundling of these work-
ing paths routing between the same PSL
- Alternate egress repair: the recovery and PML is called a Protected Path
path may have a different egress point Group (PPG). When a fault occurs on
from the working path. the working path carrying the PPG, the
PPG as a whole can be protected.
- Multi-layer repair: several network
layers are managed together in order to v) Recovery path resource use (referring to
achieve adequate protection of the traf- optional use of pre-reserved recovery
fic flows (ref. Chapter 5). paths when that path is not in use no
fault on the working path):
- Concatenated protection domains: re- - Dedicated resource: the recovery path
covery provided for traffic flows cross- resources may not be used by other traf-
ing a number of network domains. As fic flows.
these domains may apply different
recovery mechanisms it is still impor- - Extra traffic allowed: the recovery path
tant that the end-to-end protection is may carry other traffic flows when that
adequate. capacity is not needed by traffic on the
working path, e.g. when there is no fault
ii) Path mapping (mapping of traffic from on the working path. Extra traffic can be
working path to set of recovering paths): displaced whenever the recovery path
- N:M protection: up to N working paths resources are needed to carry the work-
are protected using M recovery paths. ing path traffic.
Common configurations are 1:1 and N:1
protection. - Shared resource: recovery path re-
sources are used by several working
- Split path protection: multiple recovery paths (where all of the working paths
paths carry the traffic of a working path are not assumed to fail simultaneously).
according to configurable splitting
ratios. Typically, this is used when no 6. Fault detection
single recovery paths can carry the Path Failure (PF): an indication to the
entire traffic load of the working path. MPLS mechanisms that the connectivity
This may require the Path Merge LSR, of the path is lost. This could, for exam-
PML(s) to correlate the traffic arriving ple, be identified by path continuity test
on multiple recovery paths with the or liveness messages/probes.
working path. In principle a 1+1 config-
uration could be seen as a special case Path Degraded (PD): an indication to the
where the traffic is carried on two paths MPLS mechanisms that the service level
between the PSL and PML, and the on the path is unacceptable. This may
PML correlates the packets arriving typically be done local to an LSR, e.g. by
along the two paths. receiving many packets with incorrect
labels or TTL.
iii)Bypass tunnels: creation of a tunnel
(MPLS path) for a Path Protection Group Link Failure (LF): an indication from a
(PPG) between a PSL and PML. Then lower layer to MPLS recovery mecha-
one MPLS path the tunnel carries a nisms that the link has failed. This may
number of paths by label stacking, there- for example be Loss Of Signal (LOS)
by making all the inner paths transpar- from SDH.
ent to intermediate LSRs.
Link Degraded (LD): an indication from
iv) Recover granularity (refers to the ratio of a lower layer to MPLS recovery mecha-
traffic requiring protection (e.g. ranging nisms that the link gives an unacceptable
from a fraction of a path to a bundle of service level.
paths):
- Selective traffic recovery: protection 7. Fault notification (when a fault is severe
of a fraction of traffic within the same enough to require path recovery, an inform-
ed node must take appropriate actions. In

case the node is not capable of initiated i) resource class attributes (in case of equiva-
direct actions, notification of the faults lence these are the same as the working
should be sent to the POR): path), ii) path priority attributes, iii) path
Notification by control plane messaging: pre-emption attributes.
relaying hop-by-hop along the path until
a POR is reached. When several MPLS-based recovery schemes
are compared, a set of criteria has to be defined.
Notification by user plane messaging: [Shar02] proposes the following criteria:
typically sent downstream to the PML
which then takes corrective action (e.g. as Recovery time: this time can be defined as the
POR for 1+1) or communicate with a interval from the fault occurred until a recov-
POR (by control plane messaging or user ery path is carrying the relevant traffic. Hence,
plane messaging for a bi-directional it is the sum of the fault detection time, hold-
LSP). off time, notification time, recovery operation
time and traffic restoration time.
8. Switch-over operation:
i) Recovery trigger: MPLS protection Full restoration time: this time is defined as
switching may be initiated because of the interval from the fault occurred until a per-
automatic inputs of external commands. manent restoration is reached. A permanent
Automatic inputs include fault notifica- restoration reflects the state when traffic flows
tions received (when several notifications on links capable of or having been engineered
are received for the same fault case, these to handle traffic as expected. This time may
should be correlated). differ from the recovery time depending on
when limited recovery paths are used (as
ii) Recovery action: a POR executes the opposed to equivalent paths).
recovery actions, either re-routing or pro-
tection switching as described earlier. Set up vulnerability: the amount of time a set
of working paths is left unprotected during
9. Post-recovery operation (refers to decisions recovery (such as path computation and set-up).
made after traffic flows on the recovery path):
i) Fixed protection counterparts: Back-up capacity: capacity needed during
- Revertive mode: when the fault on the recovery (also depending on traffic character-
(original) working path is cleared, the istics as well as protection plan selection algo-
traffic is switched back to this path, rithms and signalling and re-routing methods.
from the (original) recovery path.
Additive latency: additional delays along the
- Non-revertive mode: a switch back to recovery path compared with the working
the (original) working path is not imme- path.
diately carried out even after the fault
has been cleared. When the fault in the Quality of protection: includes relative and
(original) working path is cleared, that absolute survivability. Relative means that the
path may be used for other purposes, or traffic flow in question has the same service
be defined as a recovery path for the one class as other flows (contending for the same
where traffic has been switched on to. resources). Absolute means that explicit guar-
antees may be given to the protected traffic.
ii) Dynamic protection counterparts: when
the network reaches a stable state after the Re-ordering: packets may arrive in a different
fault, traffic on the recovery path may be order than sent during recovery (the same may
switched over to a different preferred path happen during restoration)
(not the original working path).
State overhead: the amount of information
iii)Restoration and notification: reversion is needed to maintain the paths
performed by the PSL by receiving notifi-
cation that the working path is repaired or Loss: a certain amount of packet loss may be
that a new working path is established. introduced during switch-overs.
iv) Reverting to preferred path (or controlled Coverage: representing various types of fail-
rearrangement): typically this involves a overs: i) Fault types (link faults, node faults,
make before break switching. degraded service level, etc.), ii) Number of
concurrent faults, iii) Number of recovery
10. Performance: the recovery paths will typi- paths, iv) Percentage of coverage (certain per-
cally have a number of attributes, including centage of fault types), and v) Number of pro-

tected paths compared with the total number Recovery techniques may be applicable
of paths (e.g. given as n/N where n is the num- for an entire end-to-end path or for seg-
ber of protected paths and N is the total num- ments thereof.
ber of paths).
MPLS-based recovery mechanisms should
10.2 MPLS Recovery Objectives take into account recovery actions for
As stated in [Shar02], MPLS-based recovery is lower layers. MPLS-based mechanisms
motivated by a number of arguments: should not trigger lower layer protection
switching.
Ability to increase network availability by en-
abling faster response to faults than possible Minimise the state overhead incurred for
with mechanisms only related to the IP layer. each recovery path maintained.
IP rerouting may often be too slow for a core Preserve the constraints on traffic after
MPLS network that needs to support recovery switch-over. That is, if desired, the recov-
times smaller than convergence times of IP ery path should meet the resource require-
routing protocols. ments of, and achieve the same perfor-
mance characteristics as the working path.
Utilising lower layers (e.g. optical channels
sometimes referred to as layer 0, and SDH B. Traffic-related:
sometimes referred to as layer 1) may result in Facilitate restoration times that are suffi-
wasteful use of resources. That is, the granu- ciently fast for the end-user application,
larity of the resources may be too coarse for i.e. matching the applications require-
the group of traffic flows that should be pro- ments.
tected.
Enhance the reliability of the protected
IP-traffic can be put directly over WDM chan- traffic while minimally or predictably
nels to provide recovery options without an degrading the traffic carried by the
intervening SDH layer (supports building IP- diverted resources.
over-WDM networks). Commonly mecha-
nisms on optical layer and SDH layer do not Protection of traffic at various granulari-
have visibility into higher layer operations. ties. For example to specify MPLS-based
Hence, they may provide e.g. link protection recovery for a portion of the traffic on an
by not easily providing node protection or individual path, for all traffic on an indi-
protection of traffic transported at the IP layer. vidual path, or for all traffic on a group of
Moreover, this may prevent the lower layers paths.
from providing restoration based on the needs
of the traffic flows. An example is that fast Minimise loss of data and packet reorder-
restoration could be provided only to traffic ing during recovery operations.
flows that need this, while slower restoration
could be used for traffic flows that do not Several of these goals may be in conflict with
need fast restoration (slower restoration may each other. Moreover, engineering compromises
also allow for time to find more optimal use of would be done based on a range of factors such
the resources). For cases where the latter traf- as cost, application requirements, network effi-
fic type is dominant, providing fast restoration ciency, and revenue considerations.
to all traffic types may not be most effective
from an operators point of view. Features
From an operational point of view, [Shar02] lists
MPLS Recovery Goals the following desirable features:
[Shar02] also states a number of goals for
MPLS-based recovery. These may be broadly MPLS recovery provides an option to identify
divided into: protection groups (PPGs) and protection por-
tions (PTPs).
A. Resource-related:
Allow for better overall use of the network Each PSL is capable of performing MPLS
resources, e.g. subject to traffic engineer- recovery upon detection of impairments or
ing goals receipt of notification of impairments.
Maximise network reliability and avail- Manual protection switching commands are
ability, e.g. by minimising number of sin- not precluded.
gle points of failure in the MPLS-pro-
tected domain

A PSL is capable of performing a switch back Basically, two shared protection schemes are:
to the original working path when the fault is
cleared or a switch-over to a new working path. Path-based: the source node of a working path
computes a protection path by ensuring that
Path merging at intermediate LSRs must be the protection path is diversely routed from
possible. If a fault affects the merged segment the working path. If a fault occurs on the
all the paths sharing that merged segment working path, the terminating node in its con-
should be able to recover. If a fault affects trol plane detects the fault and sends a notifi-
a non-merged segment only the path that is cation to the first node in the path to activate a
affected by the fault should be recovered. switch-over. The source node switches over
the traffic from the working path to the protec-
10.3 Recovery and Protection tion path. As the path length influences the
Schemes restoration time, a short failure recovery time
Basically, there are two models for path recov- may be hard to reach.
ery: rerouting and protection switching. Recov-
ery by rerouting is defined as establishing new Link-based: fault localisation and restoration
paths or path segments on demand for restoring is done at upstream neighbour node of the
traffic after the occurrence of a fault. For protec- failed link or node and the traffic merged back
tion switching mechanisms a recovery path (or to the original working path at the down-
path segment) is pre-established. The pre-estab- stream neighbour node. This may allow fast
lished path may or may not be link and node dis- restoration due to rapid fault localisation.
joint with the working path. If it is not disjoint,
the overall reliability is degraded. In addition to the short restoration times, fine
service granularity and high network throughput
As traffic dynamics grows, requirements on are sought.
network survivability are not becoming more
relaxed. During recent years, much work and Protection domains are commonly introduced.
interest have been attached to finding efficient Subsequently, a diversely routed protection path
ways of allocating spare capacity in survivable segment is sought for each working path seg-
networks. Several approaches still utilise NP-hard ment within a protection domain. A domain
optimisation process based on static working traf- diameter may then be introduced, defined as the
fic demands [Ho02]. Most of the static schemes shortest path (hop count) between a source path
can be used for reallocating the spare capacity node and termination path node. Referring to
while the network is running. However, this is this measure, link-based would have a diameter
mostly after a time-consuming optimisation pro- equal to one, while path-based would have a
cess. Hence the derived solution can be far from diameter equal to the working path length. Adja-
optimal as traffic rapidly changes. A conclusion cent protection domains must overlap in order to
in [Ho02] is that the static schemes are more protect against every single link/node failure
suited for use in designing small-sized networks, along the working path (e.g. the node on the
or networks where demands are less dynamic. edge of a domain), see Figure 19.
One attempt to overcome the computational When a fault occurs on a working path, the first
complexity is to introduce heuristic algorithms, node in the protection domain where the fault
introducing a compromise between performance occurs is notified and activates a traffic
(efficient resource utilisation) and computational switchover. For example, a fault on link C D
efficiency. [Ho02] names this process a surviv- is detected by node D and signalled to node A
able routing. A survivable routing algorithm is that switches the traffic to the protection path.
used to dynamically allocate the current connec- A fault on link G H is detected by node H and
tion request into a network with protection ser- signalled to node E that switches over the traffic.
vice, while maximising the probability of suc-
cessfully allocating subsequent connection As argued in [Ho02], introducing protection
requests in the network. domains allows for shorter restoration times
(only within a domain), less computational
In contrast to dedicated protection such as 1+1, complexity (needs to find restoration within the
when a shared protection scheme is used, several domain only), and potentially higher resource
working paths may use the same protection utilisation (protection paths could be shared out-
resource. However, when two working paths side the domain). From a network management
share the same risk of failure (e.g. caused by the perspective, it allows a trade-off between restor-
same fault), they may not share the same protec- ation time and amount of protection resources
tion resources. Obeying this, it is possible to consumed. A major drawback is the increase of
guarantee the level of impact that a single fault signalling.
(link or node) would have.

Figure 19 Dividing the paths Protection domain 1 Protection domain 2
into protection domains
A X
B
D G H
C E F
Working path
Protection path
Survivability mechanisms are available at sev- offered. An example of such classes is shown in
eral network layers, e.g. SDH, MPLS and IP. Table 3.
Moreover, these mechanisms may be in opera-
tion in multiple layers at the same time. Another As seen by the values, class 1 has the strictest
issue is that a single failure at the physical layer requirements, while class 4 has no requirements.
would likely result in several failures detected at Hence, traffic of class 4 may be pre-empted if a
the IP layers (several routes unavailable as seen failure occurs, even when this traffic is not
by the IP router). directly affected by the failure itself. This is to
release network resources for recovery of other
Generally, recovery at lower layers has advan- classes.
tages in short recovery time. The recovery at IP
or MPLS layer allows for better resource effi- A number of recovery options, matching the
ciency and recovery granularity. The latter is resilience schemes in Table 3 are given in Table 4.
important when recovery/resilience classes are
Table 3 Example of service Service class 1 2 3 4

classes and resilience options
(adapted from [Aute02]) Resilience High Medium Low None
requirements
Recover time 10 100 ms 100 ms 1 s 1 s 10 s n.a.
Resilience Protection Restoration Rerouting Pre-emption

scheme
Recovery Pre- On-demand On-demand None

path set-up established immediate delayed
Resource Pre- On-demand On-demand None

allocation reserved (assured) (if available)
Service level Equivalent May be May have None

after failure temporarily reduced
reduced service level
Recovery Protection Restoration (MPLS rerouting) (IP) rerouting

model switching
Resource Pre-reserved Reserved on-demand

allocation
Resource use Dedicated Shared resources Extra traffic allowed

resources
Path setup Pre-established Pre-qualified Established on-demand
Recovery Local Global Alternate Multi-layer repair Conc. prot.

scope repair repair egress pair domain
Recovery Automatic inputs (internal signals) External commands (OAM signalling)

Table 4 Examples of recovery
trigger
options (from [Aute02])

The characteristics of the different recovery 10.4 Recovery Cycle and
models are: Time Description
Restoration time, TR, can be decomposed as:
Protection switching: the alternative path is pre-
established and pre-reserved (pre-provisioned). TR = Tdetection + Tsignalling + Tconfig
Hence, the shortest traffic disruption is
achieved. Two main groups are 1+1 and 1:1. In Where Tdetection is the time for failure detection
the former group packets are forwarded simul- and localisation, Tsignalling is the time for signal
taneously on working and protection paths. propagation and node processing, and Tconfig is
When the working path fails, the downstream the time for configuring the network elements/
node simply selects packets from the alterna- resources along the protection path.
tive path. For 1:1, packets are forwarded on a
predefined path in case of failure on the work- Figure 20 [Shar02] describes three recovery
ing path. When no failure is present, the alter- cycles containing a set of events. During the first
native path may carry other traffic flows. How- cycle, the fault is detected and traffic restored
ever, these flows must be preemptible as they onto the MPLS recovery path. In case the recov-
should be dropped if the alternative path is ery path is non-optimal, it may be followed by
needed for the protected traffic. any of the two latter cycles to achieve an opti-
mised network again. The reversing cycle
Restoration (MPLS rerouting): the recovery applies for explicitly routed traffic that does not
path is established on-demand after detecting rely on any dynamic routing protocols to be con-
a failure. As it takes a while to calculate new verged. The dynamic re-routing cycle applies for
routes, signal them and configure the relevant traffic that is forwarded based on hop-by-hop
mechanisms, this takes considerably longer routing.
than protection switching.
The different time intervals are defined as
IP rerouting: ordinary routing protocol and follows:
exchange principles are utilised to identify
alternative paths. A. MPLS recovery cycle model:
T1 Fault detection time: the time
A number of recovery scopes may also apply. between the occurrence of an impairment
A path is either switched at the ingress and and the moment the fault is detected by
egress router or locally at the router adjacent the MPLS-based recovery mechanisms.
to the failure. A protection switching scheme The length of this interval may be heavily
where an alternative path is pre-established for dependent on lower layer protocols.
each link is commonly called MPLS fast reroute.
In this case no end-to-end failure notification T2 Hold-off time: the waiting time
and signalling are required. As described above, between the detection of a fault and taking
another method is to set up an alternative path to MPLS-based recovery actions. This may
handle fast rerouting. One approach is to set up be configured to allow time for lower
an alternative path for each working path as in- layer protection to take effect (however, it
dicated from the last hop router in the reverse may also be zero). The hold-off time may
direction to the source of the working path and also occur after the notification time inter-
along a node-disjoint path to the destination val if the node responsible for the switch-
router. When a failure is detected, the adjacent over, the Path Switch LSR (PSL), rather
upstream router immediately switches the work- than the detecting LRS, is configured to
ing path to the recovery path. Therefore, only a wait.
single protection path must be set up, and the
rerouting may still be triggered based on a local T3 Notification time: the time between
decision in the router directly upstream of the initiating a Fault Indication Signal (FIS)
failure. Hence, no recovery signalling is needed. by the LSR detecting the fault and the
instant at which the PSL begins the recov-
Protection switching schemes with global repair ery operation. This is zero if the PSL
are commonly called path protection. For each detected the fault itself or infers a fault
protected path a protection path is established from such events as an adjacency failure.
either between the ingress and egress router or As noted above, if the PSL detected the
between designated recovery-switching points fault itself, there may be a hold-off time
(e.g. for segment protection/domain). Then the period between detecting and starting the
switching node must be notified in case a path recovery operation.
fails in order to switch that path to the protection
path. T4 Recovery operation time: the time
between the first and the last recovery

Figure 20 Recovery cycle model (adapted from [Shar02])
MPLS recoveery cycle model
Network
impairment
Fault detected Start of
notification
Start of recovery
operation
Recovery operation
complete Path traffic
restored
T1 T2 T3 T4 T5
MPLS reversion cycle model
Network
impairment
repaired
Fault cleared
Path available
Start of reversion
operation
Reversion operation
complete
Traffic restored
on preferred path
T7 T8 T9 T10 T11
Network enters a
semi-stable state
after an impairment
Dynamic re-routing cycle model
Dynamic routing
protocols convergence
Initiate setup of new
working path between PSL
Switch-over
operation complete
Traffic moved to
new working path
T12 T13 T14 T15
action. This can include message ex- network that experienced disrupted or
changes between the PSL and PML to degraded service due to the occurrence
coordinate recovery actions. of the fault (the length of this interval
depends on the location of the fault, the
T5 Traffic restoration time: the time be- recovery mechanisms, and the propagation
tween the last recovery action and the delay along the recovery paths.
instant when the traffic is completely
recovered. This interval is intended to T6 (not indicated): the interval when the
account for the time required for traffic impairment has not been repaired and traf-
to once again arrive at the point of the fic is sent on the recovery path.

B. MPLS reversion cycle model (executed to T12 Network route convergence time: the
switch the traffic back to the preferred path time taken for the routing protocols to con-
after the fault on that path has been cleared): verge and the network to reach a stable state.
T7 Fault clearing time: the time between T13 Hold-down time: the time for which
the repair of an impairment and instant a recovery path must be used. This is to
when the MPLS-based mechanisms learn prevent flapping of traffic between a
that the fault has been cleared. This will working and a recovery path.
likely be heavily dependent on lower layer
protocols. T14 Switch-over operation time: the
time between the first and the last switch-
T8 Wait-to-restore time: the waiting over actions. This may include message
time between the clearing of a fault and exchanges between the PSL and PML to
MPLS-based recovery actions. The wait- coordinate the switch-over actions.
ing time may be configured to ensure that
the path is stable and to avoid flapping in T15 Traffic restoration time: the time
cases where a fault is intermitted. This when all traffic has been moved to work-
time may also be set to zero. The waiting ing paths and a new stable network situa-
time may occur after the notification time tion reached.
interval if the PSL is configured to wait.
[Huan02] presents a protection mechanism
T9 Notification time: the time between that is built on the following principles:
initiating a Fault Recovery Signal (FRS)
by the LSR clearing the fault and the time Tree structure for efficient distribution of
at which the PSL begins the reversion fault and/or recovery information (called
operation. This is zero if the PSL clears reverse notification tree).
the fault itself. As noted above, there may
still be a Wait-to-restore time period be- Hello packets to detect faults as comple-
tween fault clearing and the start of the mentary fault detection (in addition to any
reversion operation. lower layer mechanism).
T10 Reversion operation time: the time Notification transport protocol, e.g. utilis-
between the first and last reversion actions. ing UDP.
This may include message exchanges
between the PSL and PML to coordinate As pointed out in that article, one of the major
reversion actions. considerations in a path protection mechanism is
to control the delay that must be met by the noti-
T11 Traffic restoration time: the time fication message travelling from the fault detec-
between the last reversion action and tion node to the protection switching node, i.e.
instant when traffic is completely restored POR. This delay may cause additional packet
on the preferred path. This interval is loss and misordering.
expected to be small since both paths are
working and care may be taken to limit the A key aspect considered in that article is that
traffic disruption. LSPs may be merged; several working paths
may converge to form a multipoint-to-point tree
In practise, the most interesting times are the with the PSLs as leaves. The fault indications
Wait-to-restore time and the Traffic restora- and repair notifications should then be sent
tion time. As this cycle is completed when along a reverse path of the working path to all
paths are in operation, there would likely be the PSLs affected by the fault. This is in case an
little need for a rapid operation. Hence, a end-to-end recovery is to be initiated. If a single
well-controlled operation is sought with segment is to be recovered, the notification
minimal disruption. would be stopped in the PSL at the beginning
of that segment.
C. Dynamic re-routing cycle model (aims to
bring the network to a stable state after im- Each LSR must be able to detect certain fault
pairment has occurred. Hence, a re-opti- types, such as path failure, path degraded, link
mised network is sought after the routing failure and link degraded, and send correspond-
protocols have converged and the traffic has ing FIS message to the PSL. Hence, a node
been moved from a recovery path to a (pos- upstream of the fault must be able to detect or
sibly) new working path. learn about the fault. This motivates for use of a
hello protocol commonly using a much shorter
timer than applied for most routing protocols.

11 Concluding Remarks [Huan02] Huang, C et al. Building Reliable
Considering the steadily increasing growth of MPLS Networks Using a Path Protection Mech-
traffic, in particular related to IP-based services, anism. IEEE Communications Magazine, 40
it is essential for an operator to configure an effi- (3),156162, 2002.
ciently operating network. The interplay of IP
and optics has been promoted as a strong candi- [ID_ipofw] Rajagopalan, B et al. IP over Optical
date in the core network. This is also the main Networks: A Framework. IETF draft. Draft-ietf-
topic in this article, with particular emphasis on ipo-framework-05.txt. Oct.2003. Work in
how to achieve resilience for the traffic flows progress.
that have such requirements.
[ID-iporeq] Xue, Y. Optical Network Service
Several studies and international projects have Requirements. IETF draft. Draft-ietf-ipo-carrier-
been carried out in this area, only a few of the requirements-05.txt. Dec 2002. Work in
results and questions are addressed in this arti- progress.
cle. Currently it can be observed that several of
the vendors are taking positions in both the IP [Gers00] Gerstel, O, Ramaswami, R. Optical
and the optics area, also proposing networking Layer Survivability An Implementation Per-
solutions to manage the topics raised. However, spective. IEE Journal on Selected Areas in Com-
a final answer on how to take care of the munications, 18 (10), 18851899, 2000.
resilience seems not to be found yet. Neither
is the general model for IP and optics interplay [RFC3386] Lai, W, McDysan, D. Network Hier-
settled, advocating for continuing effort. arch and Multilayer Survivability. IETF RFC
3386, Nov. 2002.
References
[Aute02] Autenrieth, A, Kirstdter, A. Engineer- [RFC3471] Berger, L. Generalized Multi-Proto-
ing End-to-End IP Resilience Using Resilience- col Label Switching (GMPLS) Signaling Func-
Differentiated QoS. IEEE Communications tional Description. IETF RFC 3471, Jan. 2003.
Magazine, 4 (1), 5057, 2002.
[Shar02] Sharma, V, Hellstrand, F. Framework
[Coll02] Colle, D et al. Data-Centric Optical for MPLS-based Recovery. IETF draft Draft-ietf-
Networks and Their Survivability. IEEE Journal mpls-recovery-frmwrk-07.txt. Mar. 2002. Work
on Selected Areas in Communications, 20 (1), in progress
620, 2002.
[Stam00] Stamatelakis, D, Grover, W D. IP
[Ho02] Ho, P-H, Mouftah, H T. A Framework Layer Restoration and Network Planning Based
for Service-Guaranteed Shared Protection in on Virtual Protection Cycles. IEEE Journal on
WDM Mesh Networks. IEEE Communications Selected Areas in Communications, 18 (10),
Magazine, 40 (2), 97103, 2000. 19381949, 2000.

Terms and Acronyms
Note: Several of the acronyms have been described in articles of this issue.
2.5 G ASON
Two and a half Commonly used term for enhancements in data Automatically
Generation capabilities and bit rates of second generation systems Switched Optical
(mobile system) (GSM, TDMA, etc) like EDGE. Network
2B+D ASP
ISDN basic ISDN term: 2 channels 64 kbit/s (user traffic), 1 channel Application
access 16 kbit/s (signalling, maintenance and user traffic). Service Provider
30B + D ASTN
ISDN enhanced ISDN term: 30 channels 64 kbit/s (user traffic), 1 channel Automatically
access 64 kbit/s (signalling, maintenance and user traffic). Switched Trans-
port Network
3G
3rd generation The generation of mobile systems which is a ATM
(mobile system) descendant of todays widespread and voice centric Asynchronous
second generation mobile systems like GSM. 3G Transfer Mode
systems are characterised by multimedia capability
and data rates up to 2 Mbit/s. Ref. UMTS AUC
Authentication
4G Centre
4th generation Term often used to denote future broadband mobile com-
(mobile system) munication systems or standards to follow third gene- Authentication The process of determining who the user is. It can take
ration. Often referred to as systems beyond 3G (B3G). the form of ensuring that data has come from its
claimed source, or of corroborating the claimed identity
AAA of a communication party.
Authentication, Key functions to intelligently control access, enforce
Authorization policies, audit usage and provide the necessary inform- Authentication The server that is responsible for authenticating the
and Accounting ation for billing of Internet services. server content user {source [BCD01a]}.
Accounting Tracking which services are used, by whom, when and Authorization The process of enforcing policies of determining the
for how long. Accounting is carried out by the logging types or qualities of activities, resource, or services a
of session statistics and usage information and is used user is permitted. Usually, authorization is in the con-
for authorization control billing, trend analysis, text of authentication; once you have authenticated a
resource utilization, and capacity planning activities. user, (s)he may be authorized for different types of
access or activities.
ACTS
Advanced Thematic Programme of the Fourth Framework Authorization Determines if an authenticated user is authorized to
Communications Research Programme funded by the European Union. It server view content with a particular set of delivery para-
Technologies ran from 1994 to 1998 (the last projects ended in 2000) meters. May also bill the user for use and generate a
and Services with a total funding from the EU of 681 million Euro. billing record that is sent to the content provider. The
http://www.cordis.lu/acts content provider or a third-party could own this server
{source [BCD01a]}.
ADM
Add-drop B-ISDN
multiplexer Broadband
Integrated
Aggregator/ An entity providing access to several underlying net- Services Digital
Aggregation works/systems. This is commonly done by trading ser- Network
broker vices (e.g. bandwidth) between member providers. Bro-
kers typically provide centralized authentication services BA
in order to compute and validate the broadband traffic. Behaviour A collection of packets with the same (DiffServ) code
Aggregate point crossing a link in a particular direction.
AP {source [RFC3564]}
Access Point A point where users access the system/network,
e.g. a base station in a wireless network. Baseline A study conducted to serve as a baseline for com-
analysis paring to the actual behaviour of the network.
APS {source [RFC3272]}
Automatic Pro-
tection Switching Billing cycle The period between the creation of billing files. Typi-
cally, this is done once a day at a specific time. The
ARIMA billing file contains the traffic of all end users per roam-
Autoregressive Time series modelling ing partner for one billing cycle. The shorter the billing
Integrated cycle, the better the risk of fraud can be monitored in
Moving Average roaming environments.
ASIC BLSR
Application- Bi-directional
Specific Inte- line switched
grated Circuit ring

Bluetooth A short-range wireless specification that allows radio CAPM
connection between devices within a 10-metre range Capital Asset
of each other. Bluetooth is designed as a Personal Area Pricing Model
Network (PAN) technology with a wide variety of
theoretical uses. CaTV
Cable TV
BGP
Border Gateway CDMA
Protocol Code Division Multiple access method used e.g. in UMTS. Allowing
Multiple Access for multiple transmissions to be carried simultaneously
BOT on a single wireless channel (same time slot and carrier
Build, operate Telenor product offering to cover deployment, frequency). This is done by spreading the signal across
and Telenor operation and maintenance of ICT equipment in the frequency band using one of a set of spreading
collocation addition to collocation of technical equipment. codes such that one users signal appears as noise/
interference for other (non-intentional) receivers.
Bottleneck A network element whose input traffic rate tends to be
greater than its output rate. {source [RFC3272]} CDN
Content A collection of network elements arranged for more
Branch and A tree of linear programs is formed in which each Distribution/ effective delivery of content to clients. Typically a
bound problem has the same constraints and objectives except Delivery CDN consists of a request-routing system, surrogates
for some additional bounds on certain variables. At the Network (a content server other than the origin server), a distri-
root of this tree is the original problem without addi- bution network, and an accounting system. {source
tional requirements. The solution to this root problem [BCD01a] from IETF content peering group}
will not, in general, have all integer components. We
now choose some non-integer solution component and Clearing house A provider of roaming billing data services. A clearing
define two variants by fixing one of the variables. This house typically provides post-connection services by
gives rise to two sub-problems. The left-child and handling the billing data, offering validation and rating
right-child problems separate the solution space. This services as well as settling the financial liabilities. A
branching process can be carried out recursively; each distinction should be made between data and financial
of the two new problems will give rise to two more clearing house. The latter is also referred to as settle-
problems when we branch on one of the non-integer ment house. The former only delivers the data process-
components of their solution. It follows from this ing service whereas the latter also provides invoice and
construction that the tree is binary. payment services.
Eventually, after enough new bounds are placed on the
variables, integer solutions are obtained. The value for Congestion A state of a network resource in which the traffic inci-
the best integer solution found so far is retained and dent on the resource exceeds its output capacity over a
used as a basis for pruning the tree. If the continuous time interval. {source [RFC3272]}
problem at one of the tree nodes has a final objective
value different from the value obtained in a node, so Congestion An approach to congestion management that attempts
have all of the nodes descendants, since they have avoidance to obviate the occurrence of congestion.
smaller feasible regions and hence even larger optimal {source [RFC3272]}
objective values. The branch emanating from such a
Congestion An approach to congestion management that attempts
node cannot give rise to a better integer solution than
control to remedy congestion problems that have already
the one obtained so far, so we consider it no further;
occurred. {source [RFC3272]}
that is, we prune it. Pruning also occurs when we have
added so many new bounds to some continuous prob-
Constraint A class of routing protocols that take specified traffic
lem that its feasible region disappears altogether.
based attributes, network constraints, and policy constraints
Two strategies for solving the node problems are
routing into account when making routing decisions. Constraint-
depth-first and width-first, of which the former is
based routing is applicable to traffic aggregates as well
frequently preferred.
as flows. It is a generalization of QoS routing. {source
[RFC3272]}
BRAS
Broadband
Content A conceptual identifier of a media object and its con-
Remote Access
stituent digital bit streams. An identifier is a movie
Server
title, a song name, and in the case of a live event an
event identifier. Note that content could be comprised
BSC
of several different bit streams. For example, a particu-
Base Station Network node in the GSM network controlling a
lar movie could be encoded in three different qualities
Controller number of BTSs. http://www.etsi.org
and have several language tracks. A live event could also
BTS be encoded in different qualities. {source: [BCD01a]}
Base Trans- The radio base station of a GSM network. It consists of
Content server The server that is responsible for sending the stream to
ceiver Station one or more transmitter-receiver units, each serving
a target device over some network. This may be the
one carrier frequency. http://www.etsi.org
original server or a surrogate server owned by the con-
Busy hour A one hour period within a specified interval of time tent network.
(typically 24 hours) in which the traffic load in a net-
CPU
work or sub-network is greatest. {source [RFC3272]}
Central Pro-
BW cessing Unit
Bandwidth
CT
Bypass tunnel A path that serves to back up a set of working paths Class Type The set of Traffic Trunks crossing a link that is
using the label stacking approach. The working paths governed by a specific set of bandwidth constraints.
and the bypass tunnel must all share the same PSL and CT is used for the purposes of the link bandwidth allo-
PML {source [Shar02]}. cation, constraint based routing and admission control.
A given Traffic Trunk belongs to the same CT on all
CAPEX links. {source [RFC3564]}
Capital Expenses

CWDM can be seen as an E-LAN service that is centred on the
Coarse wave- A relatively low-cost WDM technology that uses wide use of MPLS labels in the forwarding plane as opposed
length division spacing (about 10 nm) between wavelengths and to the use of CLAN tags in the Metro Ethernet Forum
multiplexing un-cooled optics. (MEF) definitions. Both E-LAN and VPLS focus on
enabling carriers to deliver true VPN services to users.
DACS {source [MEFsite]}
Digital access
cross-connect E-Line service Ethernet line service providing point-to-point Ethernet
system Virtual Connection (EVC) between two subscribers.
The E-line service is used for Ethernet point-to-point
DB connectivity. {source [MEFsite]}
Database
EBIT
DCF Earnings Before
Discounted Interests and
Cash Flow Income Taxes
DCME EBITDA
Digital Circuit Earnings Before
Multiplication Income Taxes,
Equipment Depreciation and
Amortisation
Demand side A congestion management scheme that addresses
congestion congestion problems by regulating or conditioning EDGE
management offered load. {source [RFC3272]} Enhanced Data A modulation method for GSM and IS-136 TDMA
for GSM networks that allows for wireless data transfer up to
Dijkstras An algorithm to find the shortest paths from a single Evolution 384 kbit/s. Standardized by ETSI. http://www.etsi.org
algorithm source node to all other nodes in a weighted, directed graph.
Edge server A network element close to the edge of a backbone net-
DPP work that is able to perform content services. Service
Discounted examples include caching, video/audio stream serving
Payback Period and splitting, insertion, transcoding, content listing, etc.
{source [BCD01a]}
DPRings
Dedicated pro- Effective The minimum amount of bandwidth that can be
tection rings bandwidth assigned to a flow or traffic aggregate in order to
deliver acceptable service quality to the flow or
DRM
traffic aggregate.
Digital Rights
Management Egress traffic Traffic exiting a network or network element.
DS EIR
DiffServ Equipment
Differentiated Identity Register
Services
EPG
DSL Electronic Pro-
Digital Sub- gramme Guide
scriber Line
Erlangs Well-applied formula for calculating the blocking
DSLAM blocking probability (used for decades in telephony networks).
Digital sub- formula
scriber line
access module Ethernet International standard networking technology for
wired implementations. Basic, traditional networks
DSM offer a bandwidth of about 10 Mbit/s. Fast Ethernet
Dynamic (100 Mbit/s) and Gigabit Ethernet (1000 Mbit/s), as
spectrum well as 10 Gigabit, are also available.
management
ETSI
DTH European Tele- A non-profit membership organisation founded in
Direct to home Satellite access communications 1988. The aim is to produce telecommunications
Standards standards to be used throughout Europe. The efforts are
DWDM
Institute coordinated with the ITU. Membership is open to any
Dense A carrier-class WDM technology that uses expensive,
European organisation proving an interest in promoting
wavelength cooled optics and tight spacing between wavelengths
European standards. It was responsible for the making
division of less than a nanometre based on the specifications of
of the GSM standard, among others. The headquarters
multiplexing the International Telecommunications Union (ITU)
are in Sophia Antipolis, France. http://www.etsi.org
DWDM wavelength grid. http//www.itu.int
Extra traffic Traffic carried over the protection entity while the
E-LAN service Ethernet LAN service provides multipoint connectivity
(also referred to working entity is active. Extra traffic is not protected,
by connecting two or more users. One or more of the
as pre-emptable i.e. when the protection entity is required to protect the
connected users can receive data. At each site the user
traffic) traffic that is being carried over the working entity, the
is connected to a multipoint Ethernet Virtual Connec-
extra traffic is pre-empted. {source [RFC3386]}
tion (EVC). As new sites are added they are connected
to the same multipoint EVC. From a user point of view FEC
the E-LAN service makes the carrier network (metro Forward Equi-
and wide areas) look like a switched LAN. For an easy valence Class
comparison the Virtual Private Line Service (VPLS)

FIS in North America. GSM defines the entire cellular sys-
Fault Indication A signal that indicates that a fault along a path has tem, from the air interface to the network nodes and
Signal occurred. It is relayed by each intermediate LSR to its protocols. As of 2000 there were more than 250 million
upstream or downstream neighbour until it reaches an GSM users, which is more than half the worlds mobile
LSR that is set up to perform MPLS recovery (the phone population. http://www.etsi.org
POR). The FIS is transmitted periodically by the node
(closest to the point of failure) for some configurable GSM Alliance Consortium of North American GSM 1900 operators.
length of time. {source [Shar02]}
GSM Worlds leading wireless industry representative body,
Forecasting Statement in advance of traffic characteristics, includ- Association consisting of more than 660 second- and third-genera-
traffic ing volume (e.g. measured in Mbit/s) and number of tion wireless network operators and key manufacturers
customers. and suppliers to the wireless industry.
www.gsmworld.com
Flow The term flow is commonly used to signify the smallest
non-separable stream of data from the point of view of HB@
an end-point or termination point (source or destination Hybrid Broad- Telenor project running from 20002002 with the aim
points). {based on [ID-ipofw]} band Access of investigating and testing how different access tech-
nologies could be combined in order to develop a com-
Footprint A network of access points referring to the overall plete broadband network able to provide comparable
coverage of the network. broadband services both in urban, suburban and rural
parts. The aim was to develop Telenors superior strat-
FRS egy and specific solutions on the implementation and
Fault Recovery A signal that indicates that a fault along a working path roll-out of a hybrid access infrastructure suitable for
Signal has been repaired. Again, like the FIS, it is relayed by television and personal computer centric broadband
each intermediate LSR to its upstream or downstream services in the period 20012005.
neighbour until it reaches the LSR that performs recov- Four different technical solutions were tested in
ery of the original path. The FRS is transmitted periodi- Stavanger, Svolvr, Beito and Oslo providing a selec-
cally by the node/nodes closest to the point of failure tion of next generation interactive broadband services.
for some configurable length of time. {source [Shar02]} http://www.telenor.no/prosjekt/hba/indeks.shtml (in
Norwegian)
FTP
File Transfer HDTV
Protocol High Definition
TV
FTTH
Fibre to the home HFC
Hybrid Fibre
FWA Coax
Fixed Wireless Fixed Wireless Access consists of a radio link to the
Access home or the office from a cell site or base station. It Hierarchy A technique used to build scalable complex systems.
replaces the traditional wireless local loop, either if the It is based on an abstraction, at each level, of what is
wire based infrastructure is sparse or to gain rapid most significant regarding the details and internal
expansion in denser urban and suburban areas. structures of the levels further away. This approach
makes use of a general property of all hierarchical sys-
GB tems composed of relevant subsystems that interactions
Gigabit One billion bits. Term used to denote the number between subsystems decrease as the level of communi-
1 073 741 824 bits (230). cation between subsystems decreases.
GMPLS HLR
Generalized Home Location A database that holds subscription information about
Multi-Protocol Register every subscriber in a mobile network. An HLR is a per-
Label Switching manent SS7 database used in cellular networks. The
HLR is located at the SCP (Signal Control Point) of the
GoS
cellular provider of record, and is used to identify/ver-
Grade of Service
ify a subscriber. It also contains subscriber data related
to features and services. http://www.etsi.org
GPRS
General Packet An enhancement to the GSM mobile communications
Home operator The provider that owns the relationship with the user
Radio Service system that supports data packets. GPRS enables con-
of the service by way of a subscription agreement
tinuous flows of IP data packets over the system for
allowing the provider to bill the user for services.
such applications as web browsing and file transfer.
Supports up to 160 kbit/s gross transfer rate. Practical Horizontal The abstraction that allows a network at one tech-
rates are from 1248 kbit/s. http://www.etsi.org hierarchy nology layer, for instance a packet network, to scale.
Examples of horizontal hierarchy include BGP confed-
Greenfield A service provider that introduces itself to a new
erations, separate autonomous systems and multi-area
provider market/region without prior engagements.
OSPF. In the horizontal hierarchy, a large network is
partitioned into multiple smaller, non-overlapping sub-
GSM
networks. The partitioning criteria can be based on
Global System A digital cellular phone technology that is the pre-
topology, network function, administrative policy, or
for Mobile dominant system in Europe, but is also used around
service domain demarcation. Two networks at the
communication the world. Development started in 1982 by CEPT and
same hierarchical level, e.g. two autonomous sys-
was transferred to the new organisation ETSI in 1989.
tems in BGP, may share a peer relation with each other
Originally, the acronym was the group in charge,
through some loose form of coupling. On the other
Groupe Special Mobile, but later the group changed
hand, for routing in large networks using multi-area
its name to SMG, and GSM became the name of the
OSPF, abstraction through the aggregation of routing
system. GSM was first deployed in seven European
information is achieved through a hierarchical parti-
countries in 1992. It operates in the 900 MHz and
tioning of the network. {source [RFC3386]}
1.8 GHz bands in Europe and the 1.9 GHz PCS band

Hot spot A venue where a commercial public WLAN service IEEE 802.3 The permanent CSMA/CD (Ethernet) working group
can be accessed. Hotspots are typically found in Working Group of the IEEE 802 project. http://www.ieee802.org/3/
hotels, airports and cafs.
A network element or subsystem, which is in a state IEEE 802.3ae The Task Force chartered by the IEEE 802.3 working
of congestion. {source [RFC3272]} Task Force group with writing the specification for 10 Gigabit
Ethernet. IEEE 802.3ae is also the name of the docu-
HSCSD ment produced by the Task Force.
High speed An addition to GSM for adding faster data trans- http://www.ieee802.org/3/ae/index.html
circuit switched mission. While GSM originally supports 9.6 kbit/s,
data HSCSD supports from 14.4 up to 57.6 kbit/s circuit IETF
switched data connections. http://www.etsi.org The Internet The Internet Engineering Task Force is a large open
Engineering international community of network designers,
HSSG Task Force operators, vendors, and researchers concerned with the
Higher speed The group commissioned by the IEEE 802.3 Ethernet evolution of the Internet architecture and the smooth
study group working group with drafting the technical scope and operation of the Internet. It is open to any interested
objectives of the standards effort. individual. The actual technical work of the IETF is
done in its working groups, which are organized by
Hub A multi-port device used to connect PCs to a network, topic into several areas (e.g. routing, transport, security,
typically via Ethernet cabling or via wireless access etc.). The IETF working groups are grouped into areas,
(WLAN). Wired hubs can have numerous ports and and managed by Area Directors (AD). The ADs are
can transmit data at speeds ranging from 10 Mbit/s to members of the Internet Engineering Steering Group
multi-gigabyte rates per second. A hub transmits the (IESG). Providing architectural oversight is the Internet
packets it receives to all the connected ports. A small Architecture Board, (IAB). The IAB also adjudicates
wired hub may only connect four computers; a large appeals when someone complains that the IESG has
hub can connect 48 or more, whereas a wireless hub failed. The IAB and IESG are chartered by the Internet
can connect hundreds. Society (ISOC) for these purposes. The General Area
Director also serves as the chair of the IESG and of the
ICT IETF, and is an ex-officio member of the IAB.
Information and http://www.ietf.org
communication
technologies IGP
Internal Gate-
IEC way Protocol
International An organization that sets international electrical and
Electrotechnical electronics standards founded in 1906. It is made up of IM
Commission national committees from over 40 countries. Headquar- Instant Message
ters in Geneva, Switzerland. http://www.iec.ch
IN
IEC/ISO 11801 The international and European standard for building Intelligent
cabling. Network
IEEE Industry A non-profit organisation that helps to create common

The Institute USA based organisation open to engineers and organisation technical or commercial recommendations and specifi-
of Electrical researchers in the fields of electricity, electronics, cations that streamline the marketplace.
and Electronics computer science and telecommunications. Established
Engineers in 1884. The aim is to promote research through jour- Ingress traffic Traffic entering a network or network element.
nals and conferences and to produce standards in {source [RFC3272]}
telecommunications and computer science. IEEE has
produced more than 900 active standards and has more Integer Optimisation formulation with integer constraints on
than 700 standards under development. Divided into programming some variables.
different branches, or Societies. Has daughter organi-
sations, or chapters in more than 175 countries world- Integrator A company that offers consulting services to help put
wide. Headquarters in Piscataway, New Jersey, USA. everything together.
http://www.ieee.org
Inter-domain Traffic that originates in one autonomous system and
IEEE 802.11 Refers to a family of specifications developed by the traffic terminates in another. {source [RFC3272]}
IEEE for wireless local area networks. It also refers to
Intermediate An LSR on a working or recovery path that is neither a
the Wireless LAN working group of the IEEE 802
LSR PSL nor a PML for that path. {source [Shar02]}
project. 802.11 specifies an over-the-air interface
between a wireless client and a base station, or between
IP
two wireless clients. The IEEE accepted the specifica-
Internet Protocol A protocol for communication between computers,
tion in 1997. There are several specifications in the
used as a standard for transmitting data over networks
802.11 family, including i) 802.11 provides 1 or 2
and as the basis for standard Internet protocols.
Mbit/s transmission in the 2.4 GHz band; ii) 802.11a
http://www.ietf.org
an extension that provides up to 54 Mbit/s in the 5 GHz
band (it uses an orthogonal frequency division multi- IOT
plexing encoding scheme rather than FHSS or DSSS); Inter Operator Rate that the settlement price between Wireless ISPs
iii) 802.11b provides 11 Mbit/s transmission in the Tariff for roaming traffic is based on. The roaming agreement
2.4 GHz band and was ratified in 1999 allowing wire- between the home and the visited WISPs regulates the
less functionality comparable to Ethernet; iv) 802.11g rebate (if any) that the home operator is entitled to relative
provides 20+ Mbit/s in the 2.4 GHz band; v) 802.11z to the IOT. The IOT is often roughly equal to the normal
is a method for transporting an authentication protocol per minute price for users in the operators network.
between the client and access point, and the Transport
Layer Security (TLS) protocol. More variants are also IRR
under preparation, including support of 100 Mbit/s Internal rate of
traffic flows. http://www.ieee802.org/11/ return

ISDN LLUB
Integrated A digital telecommunication network that provides Local Loop Option to rent only the local loop (e.g. copper access
Services Digital end-to-end digital connectivity to support a wide range Unbundling line to a customer) by a non-incumbent operator.
Network of services, including voice and non-voice services, to
which users have access by a limited set of standard LMDS
multi-purpose user-network interfaces. The user is Local Multipoint System for Wireless Local Loop applications which
offered one or more 64 kbit/s channels. http://www.itu.int Distribution can replace cable based access to private and corporate
System customers. Typically operates in the millimetre wave
ISO band. Can provide bandwidth of several Mbit/s to the
International The International Organization for Standardization end customer.
Organization for (ISO) is a world-wide federation of national standards
Standardization bodies from more than 140 countries, one from each Loss network A network that does not provide adequate buffering for
country. ISO is a non-governmental organization estab- traffic, so that traffic entering a busy resource within
lished in 1947. The mission of ISO is to promote the the network will be dropped rather than queued.
development of standardization and related activities in {source [RFC3272]}
the world with a view to facilitating the international
exchange of goods and services, and to developing LP
cooperation in the spheres of intellectual, scientific, Location Point
technological and economic activity. http://www.iso.org/
LS
ISP Local switch
Internet Service A vendor who provides access for customers to the
LSP
Provider Internet and the World Wide Web. The ISP also
Label Switched
typically provides a core group of Internet utilities
Path
and services like e-mail and news group readers.
LSR
IST
Label Switching
Information Thematic Programmes of the 5th and 6th Framework
Router
Society Research Programmes funded by the European Union.
Technologies The first IST programme runs from 2000 to 2004, while
MAC
the IST programme of the 6th Framework Research
Media access
Programme started in 2003 and will run until 2006.
control protocol
http://www.cordis.lu/ist/
MAC Address
ITU
Media Access A hardwire address that uniquely identifies each node
International The ITU is a United Nations (UN) body with approx.
Control of a network.
Telecommuni- 200 member countries. It is the top forum for dis-
cation Union cussion and management of technical and administra- Management The process by which managers assure that resources
tive aspects of international telecommunications. Its control/tactical are obtained and used effectively and efficiently in the
main objective is to promote connectivity and inter- planning accomplishment of the organisations objective.
operability between its member countries and to foster {source [Anth65]}
the development of all kinds of telecommunications
worldwide. It was established in 1865 as an intergov- Mbit/s
ernmental organization with the aim of streamlining Megabit per One million bits per second. Term used to denote a
interconnection of the national telegraph networks. second data signalling rate of 1 048 576 b/s (220).
In 1947 it became a specialized agency of the UN.
http://www.itu.int MD
Main distribu-
ITV tion point
Interactive TV
Measurement Measurements can be classified on the basis of where
LD basis and at which level of aggregation the traffic data are
Link Degraded A lower layer indication to MPLS-based recovery gathered. This refers to a set of packets, possibly rela-
mechanisms that the link is performing below an tive to a particular pair of source and destination, for
acceptable level. {source [Shar02]} the purposes of defining performance parameters.
Typical measurement bases are; flow-based, interface-
LEX based, link-based, node-based, node-pair-based, path-
Local Exchange based. {source [ID_Temea03]}
LF Measurement A measurement entity defines what is measured; it is a
Link Failure A lower layer fault indicating that link continuity is lost. entity quantity for which data collection must be performed
This may be communicated to the MPLS-based recov- with a certain measurement. A measurement type can
ery mechanisms by the lower layer. {source [Shar02]} be specified by a (meaningful) combination of a mea-
surement entity with the measurement basis. {source
LIB
[ID_Temea03]}
Label Inform-
ation Base Measurement A measurement interval is the time interval over which
interval measurements are taken. Some traffic data are collected
Link-disjoint Two connections are link-disjoint if they do not share
continuously, while other data are collected by sam-
any link along the path.
pling, or on a scheduled basis. A measurement interval
consists of a sequence of consecutive read-out periods.
Liveness A message exchanged periodically between two
Summarisation is usually done by integrating the raw
message adjacent LSRs that serves as a link proving mechanism.
data over a pre-specified read-out period. The period
It provides an integrity check of the forward and back-
must be short enough to capture with acceptable
ward directions of the link between the two LSRs as well
accuracy the bursty nature of the traffic, while not
as a check of neighbour aliveness. {source [Shar02]}
being so short that the volume of data produced gets
LL too voluminous.
Leased Line

Measurement A repeatable measurement technique used to derive MPLS label The hop between two MPLS nodes, on which forward-
methodology one or more metrics of interest. {source [RFC3272]} switch hop ing is done by use of labels. {from [RFC3031]}
Metric A parameter defined in terms of standard units of MPLS label The path through one of more Label Switched Routers
measurement. {source [RFC3272]} switched path (LSRs) at one level of the hierarchy followed by a
packet of a particular FEC. {from [RFC3031]}
Mesh optical A topologically connected collection of optical sub-
network networks whose node degree may exceed 2. Such an MPLS merge A node at which label merging is done. {from
optical network is assumed to be under purview of a point [RFC3031]}
single administrative entity. It is also possible to con-
ceive a large scale global mesh optical network consist- MPLS node A node which runs MPLS. An MPLS node will be
ing of the voluntary interconnection of autonomous aware of MPLS control protocols, will operate one or
optical networks, each of which is owned and adminis- more layer 3 routing protocols, and will be capable of
tered by an independent entity. In such an environment, forwarding packets based on labels. {from [RFC3031]}
abstraction can be used to hide the internal details of
each autonomous optical cloud from external clouds. MPLS The set of LSRs over which a working path and its
{based on [ID-ipofw]} protection corresponding recovery path are routed. {source
domain [Shar02]}
MON
Metropolitan MPLS
Optical Network protection plan The set of all LSP protection paths and the mapping
from working to protection paths deployed in an MPLS
MPEG protection domain at a given time. {source [Shar02]}
Moving Pictures MPEG is a committee of ISO/IEC that is open to
Expert Group experts duly accredited by an appropriate National MRA
Standards Body. It is in charge of the development of Multilateral Cooperation agreement between a wireless ISP and a
standards for coded representation of digital audio and Roaming central legal entity where the former guarantees the
video. Established in 1988, the group has produced Agreement provision of services to the need users of the members
MPEG-1, the standard on which such products as of the latter. The central legal entity guarantees the
Video CD and MP3 are based, MPEG-2, the standard setting of minimum standards of service for existing
on which such products as Digital Television set top and future members. The result of the cooperation
boxes and DVD are based, MPEG-4, the standard for agreement is that end users can roam in the network
multimedia for the fixed and mobile web and MPEG-7, of the members of the platform.
the standard for description and search of audio and
MSC
visual content. Work on the new standard MPEG-21
Mobile Switch- The switching node in a GSM network.
Multimedia Framework has started in June 2000.
ing Centre http://www.etsi.org
http://www.chiariglione.org/mpeg/
MTX
MPLS
Mobile The switching node in an NMT network.
Multi Protocol An IETF standard intended for Internet application.
Telephone
Label Switching A widely supported method of speeding up IP-based
Exchange
data communication. {RFC 2702}
MVNO
MPLS domain A continuous set of nodes which operate MPLS routing
Mobile Virtual An actor offering mobile services not having its own
and forwarding and which are also in one routing of
Network mobile network, but relying on other actors to provide
administrative domain. {from [RFC3031]}
Operator the network resources. (Ref. Telektronikk, 97 (4), 2001)
MPLS edge An MPLS node that connects an MPLS domain with a
NB
node node which is outside of the domain, either because it
Narrowband
does not run MPLS and/or because it is in a different
domain. Note that if an LSR has a neighbouring host
NCF
which is not running MPLS, that LSR is an MPLS edge
Net Cash Flow
node. {from [RFC3031]}
NE
MPLS egress An MPLS edge node in its role in handling traffic as it
Network
node leaves an MPLS domain. {from [RFC3031]}
Element
MPLS ingress An MPLS edge node in its role in handling traffic as it
Network Handover of an uninterrupted session between different
node enters an MPLS domain. {from [RFC3031]}
handover networks or access points.
MPLS label A label which is carried in a packet header and which
Network An abstraction of part of a networks topology, routing
represents the packets FEC. A short fixed length phys-
hierarchy and signalling mechanism. The abstraction may be
ically continuous identifier used to identify an FEC of
used as a mechanism to build large networks or as a
local significance. {from [RFC3031]}
technique to enforce administrative, topological, or
geographic boundaries. For example, network hierar-
MPLS label The replacement of multiple incoming labels for a partic-
chy might be used to separate the metropolitan and
merging ular FEC with a single outgoing label. {from [RFC3031]}
long-haul regions of a network, or to separate the
MPLS label An ordered set of labels. {from [RFC3031]} regional and backbone sections of a network, or to
stack interconnect service provider networks (with BGP
which reduces a network to an autonomous system).
MPLS label The basic forwarding operation consisting of looking Two perspectives may be taken on:
swap up an incoming label to determine the outgoing label, Vertically oriented: between two network technology
encapsulating, port, and other data handling informa- layers;
tion. {from [RFC3031]} Horizontally oriented: between two areas or adminis-
trative subdivisions within the same network tech-
MPLS label A forwarding paradigm allowing streamlined forward- nology layer. {source [RFC3386]}
swapping ing of data by using labels to identify classes of data
packets which are treated indistinguishably when
forwarding. {from [RFC3031]}

Network The capability to provide a prescribed level of QoS for NPV
survivability existing services after a given number of failures occur Net present
within the network. {source [RFC3272]} value
Next-gen SDH Including the Ethernet functionality in SDH and NSP

SONET equipment using GFP, LCAS and virtual Network Service
concatenation. Next-gen SDH may also support Provider
virtual switching, including multipoint connections.
NVoD
NHLFE Near Video
Next Hop Label on Demand
Forwarding Entry
OA
NMT Ordered A set of BAs that share an ordering constraint. The set
Nordic mobile Automatic mobile telephone system based on analogue Aggregate of PHBs that are applied to this set of Behaviour
telephony transmission technology. NMT was developed by the Aggregates constitutes a PHB scheduling class.
Nordic public telephone administrations (Norway, {source [RFC3564]}
Sweden, Denmark, Finland and Iceland) in the period
1969 to 1980. The first version operated in the 450 MHz OA&M
band (NMT-450) and was launched in 1981/1982. Operation,
Later the system was enhanced to operate in the Administration
900 MHz band (NMT-900, 1986). The system offered & Maintenance
voice telephony with international roaming. The tech-
nology used was narrowband frequency modulation OBLSR
(FM) with 25 kHz user channels. The NMT-900 ser- Optical bi-
vice was discontinued to release the frequency directional line
resources in 2001 when the GSM coverage had reached switched ring
a sufficiently high level. In Norway the NMT-450 sys-
OBPSR
tem will be discontinued at the end of 2004. (Ref.
Optical bi-
Telektronikk, 91 (4), 1995)
directional path
NNI switched ring
Network-
OCh
Network
Optical Channel
Interface
Offline traffic A traffic engineering system that exists outside of the
Node-disjoint Two connections are node-disjoint if they do not share
engineering network. {source [RFC3272]}
any node along the path except the ingress and egress
nodes.
OIF
Optical Inter-
Non-revertive Case where there is no preferred path or it may be
working Forum
mode desirable to minimize further disruption of the service
brought on by a revertive switching operation. A
OMS
switch-back to the original working path is not desired
Optical The optical multiples section layer provides transport
or not possible since the original path may no longer
Multiplex for the optical channels. The information contained in
exist after the occurrence of a fault on that path. {source
Section this layer is a data stream comprising a set of optical
[RFC3386]}
layer network channels, which may have a defined aggregate band-
A recovery mode in which traffic is not automatically
width.
switched back to the original working path after this
path is restored to a fault-free condition. {source Online traffic A traffic engineering system that exists within the
[Shar02]} engineering network, typically implemented on or as adjuncts to
operational network elements. {source [RFC3272]}
NOC
Network The physical space from which a typically large Opaque vs. A transparent optical network is an optical network in
Operations telecommunication network is managed, monitored and transparent which optical signals are transported from transmitter
Centre supervised. The NOC coordinates network troubles, optical networks to receiver entirely in the optical domain without OEO
provides problem management and router configura- conversion. Generally, intermediate switching nodes in
tion services, manages network changes, allocates and a transparent optical network do not have access to the
manages domain names and IP addresses, monitors payload carried by the optical signals. Note that ampli-
routers, switches, hubs and UPS that keep the network fication of signals at transit nodes is permitted in trans-
operating smoothly, manages the distribution and parent optical networks.
updating of software and coordinates with affiliated On the other hand, in opaque optical networks transit
networks. nodes may manipulate optical signals traversing
through them. An example of such manipulation would
Normalization Sequence of events and actions taken by a network that
be OEO conversion which may involve 3R operations
returns the network to the preferred state upon complet-
(reshaping, retiming, regeneration, and perhaps ampli-
ing repair of a failure. This could include the switching
fication). {based on [ID-ipofw]}
or rerouting of affected traffic to the original repaired
working entities or new routes. Revertive mode refers Operational The process of assuring that specific tasks are carried
to the case where traffic is automatically returned to a control out effectively and efficiently. {source [Anth65]}
repaired working entity (also called switch-back).
{source [RFC3386]} OPEX
Operations
NP Expenses
Network
Performance OPPR
Optical path
protection ring

Optical channel An optical channel trail is a point-to-point optical layer signalling and routing protocols necessary for comput-
trail of lightpath connection between two access points in an optical ing and instantiating optical channel connectivity in the
network. {based on [ID-ipofw]} optical domain. {based on [ID-ipofw]}
Optical An optical internetwork is a mesh-connected collection P2P

internetwork of optical networks. Each of these networks may be Peer-to-Peer
under a different administration. {based on [ID-ipofw]}
Path continuity A test that verifies the integrity and continuity of a path
Optical mesh An optical sub-network is a network of OXCs that test or path segment. {source [Shar02]}
sub-network supports end-to-end networking of optical channel
trails providing functionality like routing, monitoring, PDH
grooming, and protection and restoration of optical Plesiochronous
channels. The interconnection of OXCs in this network Digital Hierarchy
can be based on a general mesh topology. The follow-
ing sub-layers may be associated with this network. Peering A collection of network elements supporting some
form of interconnected operation among two or more
OPTIMUM entities owned by separate organizations. Examples
Optimised Research project in the European Unions 4th Frame- include accounting peering, content list peering and
Network Archi- work Research Programme ACTS. The objective of the distribution peering. {source [BCD01a]}
tectures for project was to establish guidelines for the introduction
Multimedia of advanced communications networks and multimedia Performance A systematic approach to improving effectiveness in
Services services in a competitive multiservice environment. management the accomplishment of specific networking goals
The project used techno-economic evaluation of differ- related to performance improvement. {source
ent case studies derived from field trials and projects [RFC3272]}
within and outside ACTS. Project duration was from
Performance Metrics that provide quantitative or qualitative
1996 to 1998.
measures measures of the performance of systems or subsystems
http://www.cordis.lu/infowin/acts/rus/projects/ac226.htm,
of interest. {source [RFC3272]}
http://www.telenor.no/fou/prosjekter/optimum/
Performance A performance parameter defined in terms of standard
OSI
Metric units of measurement. {source [RFC3272]}
Open System Reference model from ISO for data communication
Interconnection divided into 7 main layers; 1 physical, 2 data link,
PD
3 network, 4 transport, 5 session, 6 presenta-
Path Fault detected by MPLS-based recovery mechanisms
tion, 7 application.
Degradation indicating that the quality of the path is unacceptable.
{source [Shar02]}
OSPF
Open Shortest
PDB
Path First
Per-Domain The expected treatment that an identifiable or target
Behaviour group of packets will receive from edge-to-edge of a
OSPR
DS domain. A particular PHB (or if applicable list
Optical shared
of PHBs) and traffic conditioning requirements are
protection ring
associated with each PDB. {source [RFC3564]}
OSS
PF
Operations
Path Failure Fault detected by MPLS-based recovery mechanisms
Support System
that is defined as the failure of the liveness message
OTN test or a path continuity test, which indicates that the
Optical Trans- path connectivity is lost. {source [Shar02]}
port Network
PG
OTNT Path group A logical bundling of multiple working path, each of
Optical Trans- which is routed identically between a PSL and a PML.
port Networks & {source [Shar02]}
Technologies
PHB
OTS Per-Hop- The externally observable forwarding behaviour
Optical Trans- This layer provides functionality for transmission of Behaviour applied at a DiffServ-compliant node to a DiffServ
mission Section optical signals through different types of optical media. behaviour aggregate. {source [RFC3564]}
layer network {based on [ID-ipofw]}
PHY
OULSR Physical layer The Ethernet PHY at Layer 1 of the OSI model defines
Optical unidi- device the electrical and optical signalling, line states, clock-
rectional line- ing guidelines, data encoding, and circuitry needed for
switching ring data transmission and reception. Contained within the
PHY are several sub-layers that perform these func-
OUPSR tions including the physical coding sub-layer (PCS)
Optical unidi- and the optical transceiver or physical media dependent
rectional path- (PMD) sub-layer for fibre media. The Ethernet PHY
switched ring connects the media to the MAC (Layer 2).
OXC PML
Optical A space-division switch that can switch an optical data Path Merge LSR An LSR that is responsible for receiving the recovery
Cross-Connect stream from an input port to an output port. Such a path traffic and either merges the traffic back on to the
switch may utilise optical-electrical conversion at the working path or, if it is itself the destination, passes the
input port and electrical-optical conversion at the out- traffic on to the higher layer protocols. {source
put port, or it may be all-optical. An OXC is assumed [Shar02]}
to have a control plane processor that implements the

PNO Protection The other path when discussing pre-planned pro-
Public Network counterpart tection switching schemes. The protection counterpart
Operator for the working path is the recovery path and vice versa.
{source [Shar02]}
PoP
Point of Presence Protection entity Entity that is used to carry protected traffic in recovery
(also called operation mode, i.e. when the working entity is in error
POR backup entity or has failed. {source [RF3386]}
Point Of Repair An LSR that is set up to perform MPLS recovery. In or recovery
other words, an LSR that is responsible for carrying out entity)
the repair of an LSP. The POR can be a PSL or a PML
depending on the recovery scheme employed. {source Protection A recovery mechanism in which the recovery path or
[Shar02]} switching path segments are created prior to the detection of a
fault on the working path. Hence, the recovery path is
PPG pre-established. {source [Shar02]}
Protected path A path group that requires protection. {source
group [Shar02]} Protection Time interval from the occurrence of a network fault
switch time until the completion of the protection-switching opera-
Provisioning The process of assigning or configuring network tions. It includes the detection time necessary to initiate
resources to meet certain requests. the protection switch, a hold-off time to allow for the
interworking of protection schemes, and the switch
PSC completion time. {source [RFC3386]}
PHB Scheduling A PHB group for which a common constraint is that
Class ordering of at least those packets belonging to the same PSL
micro-flow must be preserved. {source [RFC3564]} Path Switch LSR An LSR that is responsible for switching or replicating
the traffic between the working path and the recovery
PSDN path. {source [Shar02]}
Public Switched
Data Network PTP
Protected Traffic The portion of the traffic on an individual path that
PSTN Portion requires protection. {source [Shar02]}
Public Switched
Telephony PVR
Network Personal Video
Recorder
Pre-emption A method of determining which traffic can be dis-
priority connected in the event that not all traffic with higher QoS routing Class of routing systems that select paths to be used
restoration priority is restored after the occurrence of by a flow based on the QoS requirements of the flow
a failure. {source [RFC3386]} {source [RFC3272]}.
Protection Survivability technique based on predetermined failure RACE

(also called recovery; as the working entity is established, a Research in Name of the 2nd and 3rd Thematic Programmes in
protection protection entity is also established. Protection tech- Advanced Information and Communication Technologies funded
switching) niques can be implemented by several architectures; Communi- by the European Union. In the 1st Framework Research
1+1, 1:1, 1:n and m:n. In the context of SDH they are cations in Programme (FP) the Community Action RACE 0
referred to as Automatic Protection Switching (APS). Europe was the definition phase from 19851986 and formu-
{source [RFC3386]} lated a reference model for Integrated Broadband
Communications (IBC). The first RACE programme
1+1 protection A protection entity is dedicated to each working entity. (RACE 1) ran from 1987 to 1992 as a Community
architecture The dual-feed mechanisms is used whereby the work- programme in the field of telecommunications tech-
ing entity is permanently bridged onto the protection nologies. The 2nd RACE programme (RACE 2) ran
entity at the source of the protected domain. In normal from 1990 to1994. In the 4th FP it was followed by the
operation mode, identical traffic is transmitted simulta- ACTS Thematic Programme. http://www.cordis.lu
neously on both the working and protection entities.
At the other end (sink) of the protection domain, both RADIUS
feeds are monitored for alarms and maintenance sig- Remote An authentication and accounting system used by many
nals. A selection between the working and protection Authentication (W)ISPs. When logging into a public Internet service
entity is made based on some predetermined criteria, Dial-In User you must enter your user name and password. This
such as the transmission performance requirements or Service information is passed to a RADIUS service, which
defect indication. {source [RFC3386]} checks that the information is correct, and then autho-
rises access to the WISP. Though not an official stan-
1:1 protection A protection entity is also dedicated to each working dard, the RADIUS specification is maintained by a
architecture entity. The protection traffic is normally transmitted by working group of the IETF. http://www.ietf.org/
the working entity. When the working entity fails, the
protected traffic is switched to the protection entity. RAENPV
The two ends of the protected domain must signal Risk-adjusted
detection of the fault and initiate the switchover. expected net
{source [RFC3386]} present value
1:n protection A dedicated protection entity is shared by n working Recovery Sequence of events and actions taken by a network
architecture entities. In this case, not all of the affected traffic may after the detection of a failure to maintain the required
be protected. {source [RFC3386]} performance level for existing services (e.g. according
to service level agreements) and to allow normalisation
m:n protection A generalisation of the 1:n architecture. Typically of the network. The actions include notification of the
architecture m < n where m dedicated protection entities are shared failure followed by two parallel processes: i) a repair
by n working entities. {source [RFC3386]} process with fault isolation and repair of the failed
components; and ii) a reconfigurations process using
survivability mechanisms to maintain service continu-

ity. In protection, reconfiguration involves switching RSU
the affected traffic from a working entity to a protec- Remote sub-
tion entity. In restoration, reconfiguration involves path scriber unit
selection and rerouting of the affected traffic. {source
[RFC3386]} SAP
Service access
Recovery path The path by which traffic is restored after the occur- point
rence of a fault. Hence, the path onto which the traffic
is directed by the recovery mechanisms. Back-up path, SCP
alternative path and protection path are also seen as Service Control
synonyms to recovery path. {source [Shar02]} Point
Request The system that responds to the users request to view a SDH
manager particular piece of content. This is usually determined Synchronous
by the URI of the content. It could also be a fixed Digital Hierarchy
proxy. This system could be owned by the content
provider or content network. {source [BCD01a]} SIM
Subscriber
Rerouting A recovery mechanism where the recovery path or path Identity Module
segments are created dynamically after the detection of
a fault on the working path. Hence, the recovery path is SLA
not pre-established. {source [Shar02]} Service Level A contract between a provider and a customer that
Agreement guarantees specific levels of performance and reliabil-
Restoration A survivability technique that establishes new paths or ity at a certain cost. {source [RFC3272]}
(also called path segments on demand, for restoring affected traffic
recovery by after the occurrence of a fault. The resources in these SMATV
rerouting) alternate paths are the currently unassigned (unreserved Satellite master
resource in the same layer). Pre-emption of extra traffic antenna
may also be used if spare resources are not available to television
carry the higher-priority protected traffic. As initiated
SME
by detection of a fault on the working path the selection
Small Medium
of a recovery path may be based on pre-planned config-
Enterprise
urations, network routing policies, or current network
status such as network topology and fault information.
SMP
Signalling is used to establish the new paths to bypass
Service Manage-
the fault. Thus, restoration involves a path selection
ment Point
process followed by rerouting of the affected traffic
from a working entity to the recovery entity. {source SMP
[RFC3386]} Significant
Market Power
Restoration Method of giving preference to protect higher-priority
priority traffic ahead of lower-priority traffic. Its use is to help SMS-C
determine the order of restoring traffic after a failure Short Message
has occurred. The purpose is to differentiate service Service Centre
restoration time as well as to control access to available
spare capacity for different classes of traffic. {source SNCP
[RFC3386]} Subnetwork SDH protection schema.
Connection
Restoration time Time interval from the occurrence of a network fault to Protection
the instant when the affected traffic is either completely
restored, or until spare resources are exhausted, and/or SOHO
no more extra traffic exists that can be pre-empted to Small Office
make room. {source [RFC3386]} Home Office
Revertive mode Procedure in which revertive action, i.e. switch-back SONET

from the protection entity to the working entity, is Synchronous
taken once the failed working entity has been repaired. Optical
In non-revertive mode, such action is not taken. To NETwork
minimize service interruption, switch-back in revertive
mode should be performed at a time when there is least SPRings
impact on the traffic concerned, or by using the make- Shared
before-break concept. {source [RFC3386]} protection rings
A recovery mode in which traffic is automatically
switched back from the recovery path to the original SRG
working path upon the restoration of the working path Shared risk Set of network elements that are collectively impacted
to a fault-free condition. This assumes a failed working group by a specific fault or fault type. For example, a shared
path does not automatically surrender resources to the risk link group (SRLG) is the union of all the links on
network. {source [Shar02]} those fibres that are routed in the same physical conduit
in a fibre-span network. Besides shared conduit, this
Roaming The possibility of changing operator or end user. concept includes other types of compromise such as
Roaming does not refer to technical handover between shared fibre cable, shared right of way, shared optical
access points or different networks such as GPRS and ring, shared office without power sharing, etc. The
WLAN. span of an SRG, such as the length of the sharing for
compromised outside plant, needs to be considered on
RSS a per fault basis. The concept of SRG can be extended
Remote sub- to represent a risk domain and its associated capabili-
scriber stage ties and summarisation for traffic engineering. {source
[RFC3386]}

A group of links or nodes that share a common risk according to the operators Inter-operator Tariff (IOT),
component, whose failure can potentially cause the plus any bilaterally agreed arrangements or discounting
failure of all the links or nodes in the group. When schemes. The transfer of TAP records between the vis-
referring to link resources, Shared Risk Link Group ited and the home mobile networks may be performed
(SRLG) applies an example being fibre links in the directly, or more commonly, via clearing house.
same conduit (all links are affected when the conduit is Invoicing between the operators then normally happens
broken). once per month.
SS TCBH
Special Time-Consistent
subscriber Busy Hour
SSno7 TCP
Signalling A signalling protocol defined by the International Transmission
System No. 7 Telecommunication Union (ITU), basically as a way to Control Protocol
control circuit switched connections in ISDN although
used in several systems (including GSM, IN). TDM
Time Division
SSP Multiplexing
Service
Switching Point TE
Traffic
Stability An operational state in which a network does not oscil- Engineering
late in a disruptive manner from one mode to another.
{source [RFC3272]} TERA
Techno- Research project in the ACTS programme in FP4. The
STM Economic main objective of this project is to support consolida-
Synchronous Results tion, condensing and rationalising of deployment
transport module from ACTS guidelines for introduction of advanced communication
services and networks.
Strategic The process of deciding on the objectives of the http://www.telenor.no/fou/prosjekter/tera/;
planning organisation, on changes in these objectives, on the http://www.cordis.lu/infowin/acts/rus/projects/ac364.htm
resources used to attain these objectives, and on the
policies that are to govern the acquisition, use and dis- TEX
position of these resources. {source [Anth65]} Transit
Exchange
Supply side A congestion management scheme that provisions
congestion additional network resources to address existing and/or TIA 568 The U.S. standard for building cabling.
management anticipated congestion problems. {source [RFC3272]}
TITAN
Survivability The capability of a network to maintain service conti- Tool for Intro- EU research project.
nuity in the presence of faults within the network. Sur- duction strategies
vivability mechanisms such as protection and restora- and Techno-
tion are implemented either on a per-ink basis, on a economic
per-path basis, or throughout an entire network to alle- evaluation of
viate service disruption at affordable costs. The degree Access Network
of survivability is determined by the networks capabil-
ity to survive single failures, multiple failures, and TONIC
equipment failures. {source [RFC3386]} TechnO- Research project in the IST programme in FP5.
ecoNomICs of It concentrates on techno-economic evaluation of new
Switch-back The process of returning the traffic from one or more IP optimised communication networks and services, in order to
recovery paths back to the working paths. {source networks and identify the economically viable solutions that can
[Shar02]} services make the Information Society to really take place.
http://www-nrc.nokia.com/tonic/
Switch-over The process of switching the traffic from the path that
the traffic is flowing on onto one or more recovery TP
paths. This may involve moving traffic from a working Transmission
path onto recovery paths or may involve moving traffic point
from a recovery path onto more optimal working paths.
{source [Shar02]} Traffic A description of the temporal behaviour or a
characteristic description of the attributes of a given traffic flow or
TA traffic aggregate. {source [RFC3272]}
Traffic A collection of packets with a code-point that maps to
Aggregate the same PHB, usually in a DS domain or some subset Traffic A collection of objects, mechanisms and protocols that
of a DS domain. A traffic aggregate marked for PHB x engineering are used conjunctively to accomplish traffic engineer-
is referred to as the x traffic aggregate or x aggregate. system ing objectives. {source [RFC3272]}
This generalizes the concept of Behaviour Aggregate
from a link to a network. {source [RFC3564]} Traffic flow A stream of packets between two end-points that can
be characterized in a certain way. A micro-flow has a
TAP more specific definition: A micro-flow is a stream of
Transferred A standard maintained by the GSM organization, by packets with the same source and destination addresses,
Account which GSM operators exchange roaming billing source and destination ports, and protocol ID. {source
Procedure information. This is how roaming partners are able to [RFC3272]}
bill each other for the use of network and services
through a standard process. The TAP files are gener- Traffic intensity A measure of traffic loading with respect to a resource
ated and sent, at the latest 36 hours from call end time. capacity over a specified period of time. In classical
This means that operators can send one or many TAP telephony systems traffic intensity is measured in units
files per day. TAP files contain rated call information of Erlang. {source [RFC3272]}

Traffic matrix A representation of the traffic demand between a set of USO
originating and destination abstract nodes. An abstract Universal
node can consist of one or more network elements. Service
{source [RFC3272]} Obligation
Traffic The process of observing traffic characteristics at a Vertical An abstraction, or reduction in information, which
monitoring given point in a network and collecting the traffic hierarchy would be of benefit when communicating information
information for analysis and further action. {source across network technology layers, as in propagation
[RFC3272]} information between optical and router networks.
In the vertical hierarchy the total network functions are
Traffic trunk An aggregation of traffic flows of the same class which partitioned into a series of functional or technological
are placed inside a Label Switched Path. {source layers with clear logical, and maybe even physical sep-
[RFC3564]} aration between adjacent layers. Survivability mecha-
A traffic trunk is an abstraction of traffic flow travers- nisms either currently exist or are being developed at
ing the same path between two access points which multiple layers in networks. The optical layer is now
allow some characteristics and attributes of the traffic becoming capable of providing dynamic ring and mesh
to be parameterised. {based on [ID-ipofw]} restoration functionality, in addition to traditional 1+1
An aggregation of traffic flows belonging to the same or 1:1 protection. The SDH layer provides survivability
class which are forwarded through a common path. A capability with automatic protection switching (APS),
traffic trunk may be characterized by an ingress and as well as self-healing ring and mesh restoration archi-
egress node, and a set of attributes which determine its tectures. Similar functionality has been defined in the
behavioural characteristics and requirements from the ATM layer with work ongoing to also provide such
network. {source [RFC3272]} functionality using MPLS. At the IP layer rerouting is
used to restore service continuity following link and
Transit traffic Traffic whose origin and destination are both outside of node outages. Rerouting at the IP layer, however,
the network under consideration. {source [RFC3272]} occurs after a period of rerouting convergence, which
may require from a few seconds to several minutes to
Trust domain A trust domain is a network under a single technical
complete. {source [RFC3386]}
administration in which adequate security measures are
established to prevent unauthorized intrusion from out- Visited provider The provider that provides services to a user with
side the domain. Hence, it may be assumed that most whom the service providing provider does not have
nodes in the domain are deemed to be secured to trusted any legal ties but who is a subscriber of a provider
in some fashion. Generally, the rule of single admin- with whom a roaming agreement exists.
istrative control over a trust domain may be relaxed in
practice if a set of administrative entities agree to trust VoD
one another to form an enlarged heterogeneous trust Video on
domain. However, within a trust domain, any subverted demand
node can send control messages which can compromise
the entire network. {based on [ID-ipofw]} VoIP
Voice-over-IP Voice transmission using Internet Protocol to create
TSC digital packets.
Transit Switch-
ing Centre VPN
Virtual Private A network that is constructed by using public wires to
UDP Network connect nodes. For example there are a number of sys-
User Datagram tems that enable you to create networks using the Inter-
Protocol net as the medium for transporting data. These systems
use encryption and other security mechanisms to
UMTS ensure that only authorized users can access the
Universal Mobile The candidate for a 3rd generation system standardised network and that the data cannot be intercepted.
Telecommuni- and promoted by 3GPP. The standardisation work
cation System began in 1991 by ETSI but was transferred in 1998 to Wavelength A lightpath is said to satisfy the wavelength continuity
3GPP as a cooperation between Japanese, Chinese, continuity property if it is transported over the same wavelength
Korean and American organisations. It is based on the property end-to-end. Wavelength continuity is required in opti-
use of WCDMA technology and is currently deployed cal networks with no wavelength conversion feature.
in several European countries. The first European ser- {based on [ID-ipofw]}
vices opened in 2003 and more are expected in 2004
and 2005. In Japan, NTT DoCoMo opened its pre- Wavelength A lightpath that satisfies the wavelength continuity
UMTS service FOMA (Freedom Of Mobile multi- path property is called a wavelength path. {based on
media Access) in 2000. The system operates in the 2.1 [ID-ipofw]}
GHz band and is capable of carrying multimedia
traffic. It offers a maximum user bit rate of 2 Mbit/s, WDM
however more realistic rates are 144 kbit/s and 384 Wavelength A means of data transmission that uses optical multi-
kbit/s for data services. http://www.3gpp.org/ division plexing to enable two or more wavelengths, each with
multiplexing its own data source, to share a common fibre optic
UPSR medium.
Uni-diretional Wavelength Division Multiplexing is a technology that
path switched allows optical signals operating at different wave-
ring lengths to be multiplexed onto a single optical fibre
and transported in parallel through the fibre. In general,
UPT each optical wavelength may carry digital client pay-
Universal loads at a different data rate and in a different format.
Personal Tele- For example, a mixture of SDH 2.5 Gbit/s and 10
communications Gbit/s can be carried over a single fibre. An optical
system with WDM capability can achieve parallel
URI transmission of multiple wavelengths gracefully while
Universal maintaining high system performance and reliability.
Resource
Identity

In the near future, commercial dense WDM systems References
are expected to concurrently carry more than 160 [Anth65] Anthony, R N. Planning and Control System; A Framework for
wavelengths at rates 10 Gbit/s or above, resulting in Analysis. Cambridge, MA, Harvard University Press, 1965.
more than 1.6 Tb/s. {based on [ID-ipofw]}
[BCD01a] Broadband Content Delivery forum: Requirements for end-to-
Wi-Fi alliance A non-profit international association formed in 1999 end delivery of broadband content. 2001.
to certify interoperability of wireless Local Area Net-
work products based on IEEE 802.11 specification. [ID_ipofw] Rajagopalan, B et al. IP over Optical Networks: A Framework.
Currently the Wi-Fi Alliance has 207 member compa- IETF draft. Draft-ietf-ipo-framework-05.txt. Oct. 2003. Work in progress.
nies from around the world, and over 1000 products
have received Wi-Fi certification since certification [ID-Temea03] Lai, W S, Tibbs, R W, Van den Berghe, S. Requirements
began in March 2000. http://www.wifialliance.org/ from Internet Traffic Engineering Measurement. Draft-ietf-tewg-measure-
06.txt. July 2003. Work in progress.
WISP
Wireless Internet A commercial provider of WLAN services in public [MEFsite] Metro Ethernet Forum. www.metroethernetforum.com
Service Provider places.
[RFC3272] D. Awduche et al. Overview and Principles of Internet Traffic
WLAN Engineering. IETF RFC 3272. May 2002.
Wireless Local Usually used to define wireless access to local area net-
Area Network works based on IP. The term is often used specifically [RFC3386] Lay, W, McDysan, D. Network hierarchy and multiplayer sur-
to denote the 802.11 standards, or Wi-Fi. A WLAN vivability. IETF RFC 3886. Nov. 2002.
access point (AP) usually has a range of 20300 m. A
[RFC3564] Le Faucheur, F, Lai, W. Requirements for Support of Differenti-
WLAN may consist of several APs and may or may not
ated Services-aware MPLS Traffic Engineering. IETF RFC 3564. July 2003.
be connected to the Internet.
[Shar02] Sharma, V, Hellstrand, F. Framework for MPLS-based Recovery.
WLL
IETF draft. Draft-ietf-mpls-recovery-frmwrk-07.txt. Mar. 2002.
Wireless Local A means of provisioning a local loop facility without
Loop wires. Usually employing low power radio systems
running in the microwave range. Widely deployed in
Asia and other developing countries where they offer
the advantages of rapid deployment, and rapid configu-
ration and reconfiguration, as well as avoidance of the
cost of burying wires and cables. WLL is also attractive
to new operators bypassing the incumbent PNOs in a
deregulated, and competitive environment.
Working entity Entity that is used to carry traffic in normal operation

mode. Depending upon the context, an entity can be
a channel or a transmission link in the physical layer,
a Label Switched Path (LPS) in MPLS or a logical
bundle of one or more LSPs. {source {RFC3386]}
Working path The protected path that carries traffic before the occur-
rence of a fault. The working path exists between a
PSL and PML. The path is a hop-by-hop routed path,
a trunk, a link, an LSP or part of a multipoint-to-point
LSP. Commonly primary path and active path are also
seen as synonyms to working path. {source [Shar02]}
WTSA
World Tele- The World Telecommunication Standardization
communication Assembly (WTSA) is held every four years by the
Standardization International Telecommunication Union (ITU), a
Assembly specialised agency of the United Nations, to define
general policy for ITUs Telecommunication Standard-
ization Sector (ITU-T Sector). WTSA adopts the work
programme, sets up the necessary Study Groups (SGs),
designates chairmen and vice-chairmen of the Study
Groups, reviews the working methods and approves
Recommendations. The WTSA is open for participa-
tion from both Member States and Sector Members
(operators and industry) of the T-Sector.
http://www.itu.int/
xDSL Various configurations of digital subscriber line:

X = ADSL asymmetric, VDSL very high speed,
SHDSL single pair high speed, SDSL symmetric,
HDSL high speed.

T03 3-4

Uploaded by

Copyright:

Available Formats

T03 3-4

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

T03 3-4

Uploaded by

Copyright:

Available Formats

Contents

Telektronikk Network Planning

Editorial office: 99 Portfolio Evaluation Under Uncertainty

Layout and illustrations:

Prepress and printing:

The essentials of markets must be understood as

NOW time Before Lunch

The virtual operator choice is also accompanied

Figure 3 Options available

Year 2004 Year 2010

different sales representatives holding meetings multipoint configurations, self-management and

So, there seems to be several options open in the

Having a full-fledged set of tools is essential to A network operator, as an instance of an enter-

Merging and Politicial

Objectives to reach (future revenue, number of

Influence on performance indicators, in partic- Marketing

Statement Relations Purposes (examples)

Income Net income = revenues expenses to analyze potential profit

EBITDA The elements in the business plan contain finan-

Figure 4 illustrates relations between key mea-

Customers Shareholders Banks, lenders

Sales of services Increase of equity Increase of debt

Cash flow (inflow outflow)

Increase the number of new customers, sales et

Limit the churn rate of customers by keeping -2

By network segment Access, aggregation, region, core, end-to-end

By geographical area City X, Region Y, Country Z, etc.

By network technology ISDN, GSM, UMTS, IP, ATM, SDH, etc.

By service group Voice, leased capacity, broadcasting channels, etc.

Strategic, business modelling, design,

Description of network architectures, node

Financial and business planning

Coherent procedures and tools

Broadcast TV e.g. MPEG2 2 6 Mbit/s1) ~1 s 10-5 Yes

PC Focused Services Typical Typical Delay Packet On

High Speed Internet Access (browsing,

Capacity management and traffic management

The first phase refers to definition of control

Teletraffic theory can be defined as (ref. [ITU-

Traffic demand characterisation Grade of Service objectives

Traffic Traffic QoS requirements

End-to-end GoS objectives

Traffic controls and dimensioning

The yearly variations; some days or seasons

Comprehensible the significance of the User impatience combined with an excessively

probability, say pi, that a new attempt will be (1 B) (t + s (1 B p) Figure 23 Schematic

9 Forecasting Descriptive approach, where the basic

Time Time Time

Kalman filter models, Regression models, Logit Sum of business

In several networks, measurements of aggregates where

Performance quality control

Performance Design participation

Performance estimation modelling

Define and build test cases

Figure 28 Performance activities related to system development phases (based on [Sing02])

where Co is a constant. P(n1, n2, ..., nK) =

Linear programming Simplex method

Non-linear programming Gradient-based procedures

Flow optimisation Critical path, maximum flow, etc.