Fisheye State Routing in Mobile Ad Hoc Networks

Guangyu Pei Mario Gerla Tsu-Wei Chen

Computer Science Department Computer Science Department Bell Laboratories
University of California, Los Angeles University of California, Los Angeles Lucent Technologies
405 Hilgard Avenue 405 Hilgard Avenue 600 Mountain Avenue
Los Angeles, CA 90095 Los Angeles, CA 90095 Murray Hill, NJ 07974
email: pei@cs.ucla.edu email: gerla@cs.ucla.edu email: tsuwei@research.bell-labs.com

Abstract In this paper, we introduce a new routing scheme for ad

hoc wireless networks. It is a link state based routing pro-
In this paper, we present a novel routing protocol for tocol which is adapted to the wireless ad hoc environment.
wireless ad hoc networks – Fisheye State Routing (FSR). The rest of the paper is organized as follows. In section 2,
FSR introduces the notion of multi-level fisheye scope to we survey the existing wireless routing schemes. We de-
reduce routing update overhead in large networks. Nodes scribe the Fisheye State Routing (FSR) in section 3. Sec-
exchange link state entries with their neighbors with a fre- tion 4 presents the performance results and we conclude our
quency which depends on distance to destination. From link paper in section 5.
state entries, nodes construct the topology map of the entire
network and compute optimal routes. Simulation experi- 2 Brief Review of Routing Protocols
ments show that FSR is simple, efficient and scalable rout-
ing solution in a mobile, ad hoc environment.
Existing wireless routing schemes can be classified into
three categories according to their design philosophy: (a)
distance vector based; (b) link state based; (c) on demand.
1 Introduction Historically, the first type of routing scheme used in early
packet networks such as the ARPANET was the distance
As the wireless and embedded computing technologies vector type. The main advantages of the distance vector
continue to advance, increasing numbers of small size and approach are simplicity and computation efficiency. How-
high performance computing and communication devices ever, this approach suffers from slow convergence and ten-
will be capable of tetherless communications and ad hoc dency of creating routing loops. While several approaches
wireless networking. An ad hoc wireless network is a self- were proposed which solve the looping problem [17, 15],
organizing and self-configuring network with the capability none of them overcome the problem of slow convergence.
of rapid deployment in response to application needs. An The solutions to both convergence and looping come in the
important characteristic which sets ad hoc networks apart form of the Link State (LS) approach. LS is the preferred
from cellular networks is the fact that they do not rely on scheme for wired nets (e.g., Internet [14] or ATM [1]). In
a fixed infrastructure. Ad hoc networks are very attractive Link State, global network topology information is main-
for tactical communication in military and law enforcement. tained in all routers by the periodic flooding of link state
They are also expected to play an important role in civilian updates by each node. Any link change triggers an imme-
forums such as convention centers, conferences, and elec- diate update. As a result, the time required for a router
tronic classrooms. Mobility, potentially very large number to converge to the new topology is much less than in the
of mobile nodes, and limited resources (e.g., bandwidth and distance vector approach. Due to global topology knowl-
power) make routing in ad hoc networks extremely chal- edge, preventing routing loop is also easier. Unfortunately,
lenging. The routing protocols for ad hoc wireless networks as Link State relies on flooding to disseminate the update
have to adapt quickly to the frequent and unpredictable information, excessive control overhead may be generated,
changes of topology and must be parsimonious of commu- especially when mobility is high and frequent updates are
nications and processing resources. triggered. In addition, the small update packets make for

This work was supported in part by NSF under contract ANI-9814675,

inefficient use of the wireless MAC layer. When wireless,
in part by DARPA under contract DAAB07-97-C-D321 and in part by In- ad hoc network size and mobility increase (beyond certain
tel. thresholds), current proactive routing schemes (i.e., the dis-
tance vector and link state) become infeasible since they odically exchanges control packets to inform all the other
will consume a large part of network capacity and node pro- nodes of its location. Each control packet is assigned a life
cessing power to transmit update control messages just to time based on the geographical distance from the sender.
keep up with the topology changes. DREAM sends short lived packet more frequently than long
The most recent addition to the family are the on de- lived packets due to the so called distance effect, i.e., the
mand routing schemes. These have been specifically intro- farther two nodes separate, the slower they seem to be mov-
duced in order to overcome some limitations of the proac- ing with respect to each other. The data packet is broadcast
tive protocols in mobile environments. Examples include to the nodes in the direction of the destination using only
AODV [18], TORA [16], DSR [6], ABR [20]. The basic location information stored at the sender.
idea behind these reactive protocols is that a node discov-
ers a route in an “on demand” fashion, namely, it com- 3 Fisheye State Routing (FSR)
putes a route only when needed. In on demand schemes,
query/response packets are used to discover (possible more 3.1 Topology Representation in FSR
than) one route to a given destination. These control packets
are usually smaller than the control packets used for rout-


 
The ad hoc wireless network is modeled as an undirected


ing table updates in proactive schemes, thus causing less graph , where is a set of nodes and is
overhead. However, since a route has to be entirely discov- a set of undirected links connecting nodes in . Each
ered prior to the actual data packet transmission, the initial node has a unique identifier and represents a mobile host
search latency may degrade the performance of interactive

with a wireless communication device with transmission
applications (e.g., distributed database queries). Moreover, range , and an infinity storage space. Nodes may move
it is impossible to know in advance the quality of the path

around and change their speed and direction independently.  
(e.g., bandwidth, delay etc) prior to call setup. Such a pri- An undirected link connecting two nodes and is  
ori knowledge is very desirable in multimedia applications,
 
formed when the distance between and become less than 
since it enables more effective call acceptance control. If

or equal to . Link is removed from when node
the route breaks down because of mobility, a packet may
need multiple route discoveries on the way to destination.
and move apart, and out of their transmission ranges.
In the FSR routing implementation, for each node , one

    
Route caching becomes ineffective in high mobility. Since
 
list and three tables are maintained. They are: a neighbor
flooding is used for query dissemination and route mainte-
! "
list , a topology table , a next hop table and a
nance, on demand routing tends to become inefficient when distance table . 
is defined as a set of nodes that are ad-
 
" $# %'& (  )# & *+(
traffic load and mobility are high and network size grows jacent to node . Each destination has an entry in table

 $# %' & , - /*+,

large [10]. which contains two parts: and .

  $#.&
A recent proposal which combines on demand rout- denotes the link state information reported by
ing and conventional routing is Zone Routing Protocol node . denotes the time stamp indicating the
(ZRP) [7, 8]. For routing operation inside a local zone,
  0  1  ,
time node has generated this link state information. Sim-
an arbitrary proactive routing scheme (e.g., distance vec- ilar, for every destination , 
denotes the next
( 
 !
tor) can be applied. For interzone routing, on demand rout- hop to forward packets destined to on the shortest path,
ing is used. The advantage of zone routing is its scalabil- while denotes the distance of the shortest path from
ity, as “global” routing table overhead is limited by zone
size. Yet, the benefits of global routing are preserved within
to .
Additionally, a weight function, weight: , is
32 64 7 5
each zone. The performance of ZRP is dependent on a key used to compute the distance of a link. Since min-hop short-
parameter: the zone radius. The choice of radius is deter- est path is the only objective in this paper, this weight func-

8
mined by network characteristics (e.g, node density, relative tion simply returns 1 if two nodes have direct connection,
node velocity etc.) [8], which dynamically change in ad hoc otherwise, it returns . This weight function may also be
networks. Moreover, the interzone route discovery packets replaced with other functions for routing with different met-
may loop back into zones already queried. This must be rics. For instance, a bandwidth function can be used to re-
avoided to prevent overhead which can be potentially worse alize a QoS routing.
than for flooding based queries [8].
With the availability of GPS [11] technology, any of the 3.2 Description of FSR protocol
previous routing protocols can be assisted by GPS location
information. For example, LAR [13] is an on demand pro- FSR is an implicit hierarchical routing protocol. It
tocol similar to DSR but it restricts control packet flooding uses the “fisheye” technique proposed by Kleinrock and
by using location information. DREAM [3] is a location Stevens [12], where the technique was used to reduce the
based proactive scheme. Each node in the network peri- size of information required to represent graphical data. The
eye of a fish captures with high detail the pixels near the fo- 2
cal point. The detail decreases as the distance from the focal
8
point increases. In routing, the fisheye approach translates 5 1
3 9

to maintaining accurate distance and path quality informa- 9

6 4
7 Hop=1
tion about the immediate neighborhood of a node, with pro- 10
13 12
gressively less detail as the distance increases. 11
18 19
21 Hop=2

FSR is functionally similar to LS Routing in that it main- 36 14

15
16 17 23 22 Hop>2
20
tains a topology map at each node. The key difference is the 25 27 29 35
24
way in which routing information is disseminated. In LS, 26
28 34
30
link state packets are generated and flooded into the net- 32
31
work whenever a node detects a topology change. In FSR,
link state packets are not flooded. Instead, nodes maintain
a link state table based on the up-to-date information re-
ceived from neighboring nodes, and periodically exchange Figure 1. Scope of fisheye
it with their local neighbors only (no flooding). Through
this exchange process, the table entries with larger sequence
numbers replace the ones with smaller sequence numbers. a considerable fraction of link state entries are suppressed
The FSR periodic table exchange resembles the vector ex- in a typical update, thus reducing the message size. This
change in Distributed Bellman-Ford (DBF) (or more pre- strategy produces timely updates from near stations, but
cisely, DSDV [17]) where the distances are updated accord- creates large latencies from stations afar. However the
ing to the time stamp or sequence number assigned by the imprecise knowledge of the best path to a distant desti-
node originating the update. However, in FSR link states nation is compensated by the fact that the route becomes
rather than distance vectors are propagated. Moreover, like progressively more accurate as the packet gets closer to
in LS, a full topology map is kept at each node and shortest destination. As the network size grows large, a “graded”
paths are computed using this map. frequency update plan must be used across multiple scopes
In a wireless environment, a radio link between mo- to keep the overhead low.
bile nodes may experience frequent disconnects and recon-
nects. The LS protocol releases a link state update for each
such change, which floods the network and causes excessive
overhead. FSR avoids this problem by using periodic, in-
TT HOP
stead of event driven, exchange of the topology map, greatly 0
0:{1} 1 TT HOP
reducing the control message overhead. 1:{0,2,3} 0 0:{1} 2
2:{5,1,4} 1 1:{0,2,3} 1
When network size grows large, the update message 3:{1,4} 1 1 2:{5,1,4} 2
4:{5,2,3} 2 3:{1,4} 0
could consume considerable amount of bandwidth, which 5:{2,4} 2 4:{5,2,3} 1
3 5:{2,4} 2
depends on the update period. In order to reduce the size
of update messages without seriously affecting routing ac-
curacy, FSR uses the Fisheye technique. Fig. 1 illustrates 2 TT HOP
0:{1} 2
the application of fisheye in a mobile, wireless network. 4 1:{0,2,3} 2
The circles with different shades of grey define the fisheye 2:{5,1,4} 1
3:{1,4} 1
scopes with respect to the center node (node 11). The scope 5 4:{5,2,3} 0
5:{2,4} 1
is defined as the set of nodes that can be reached within a
given number of hops. In our case, three scopes are shown
for 1, 2 and 2 hops respectively. Nodes are color coded Figure 2. Message reduction using fisheye
as black, grey and white accordingly. The number of levels
and the radius of each scope will depend on the size of the
network. The FSR concept originates from Global State Routing
The reduction of routing update overhead is obtained (GSR) [5]. GSR can be viewed as a special case of FSR, in

8
by using different exchange periods for different entries which there is only one fisheye scope level and the radius
in routing table. More precisely, entries corresponding is . As a result, the entire topology table is exchanged
to nodes within the smaller scope are propagated to the among neighbors. Clearly, this consumes a considerable
neighbors with the highest frequency. Referring to Fig. 2, amount of bandwidth when network size becomes large.
entries in bold are exchanged most frequently. The rest of Through updating link state information with different fre-
the entries are sent out at a lower frequency. As a result, quencies depending on the scope distance, FSR scales well
 
 

proc Node 
 proc FindSP
NodeInit Dijkstra’s
G '
 shortest-path algorithm
while TRUE do
% 
   5 
if PktQueue
  !! packet received 2  4  C
8
foreach pkt  PktQueue do
 foreach 2    " 
"E
 $ "=
 do

 pkt  source
  if
%   2    
  2 
 
PktProcess
 pkt then  AHI  weight HJ
od NEXT
%  2  '&   AHI )(+* 
fi 
 else NEXT
CheckNeighbors
  
 !  fi
   
 od
G
FindSP while
$ do

 H ( G  G
RoutingUpdate foreach HI$  >K do
od Find > HI such  that
%  9 HI 
. weight >
ML#NPO > weight > 


proc NodeInit od  1 H 
G G
foreach "# $ % AHI  %    9 HI
do
    @AHI  > weight   >
"  ' & 
NEXT NEXT >
% 
"   )(+* 
od
NEXT  "  0(+*  .

; 
-,/. " proc PktProcess  pkt 
od
  !  1324 
 2   source
   pkt   source     
link exists    "  Q " source 
  
 !  
%  
  !  5  foreach "E 1$
 
 
 do
 C RSH 6 QU @
 
NEXT if "1
 T,/. "  T,/. "
6 7!5 
 
8  then begin
6       RSH 6 
-,/.
   -,/.  "  RJH 6  -,/.  "
.   "   "


proc RoutingUpdate end
6 7 6 9* 
fi
  
  6  
   -,/
.  
od
  2   .
 

foreach proc CheckNeighbors F
do  
  
 1:2  foreach
A"E XZY 6 
 do 

   &

     if V W
S([ "  
od 
; "I
message 2  Senderid fi
foreach =< do * od
for ScopeLevel >?
to  do .
@ A
5 
if
C
clock %  2  mod UpdateInterval @ B
FisheyeScope  B
then message  TT 
D  2  
message  TT   
fi
od
od   F 
broadcast " message to all "E
.

Figure 3. The FSR Protocol

to large network size and keeps overhead low without com- received routing messages, which contain link state in-
promising route computation accuracy when the destination formation broadcasted by its neighbors. PktProcess(i,pkt)
is near. By retaining a routing entry for each destination, makes sure that only the most up to date link state in-

# & *+(
FSR avoids the extra work of “finding” the destination (as formation is used to compute the bestd route by compar-
in on-demand routing) and thus maintains low single packet ing the embedded sequence number, eTf 
, with the
transmission latency. As mobility increases, routes to re-  ones stored in node ’s local storage, for each destination
mote destinations become less accurate. However, when a . If any entry in the incoming message has a newer se-  "1 # %'& (
packet approaches its destination, it finds increasingly ac-
# %'&
quence numberd regarding destination , ,
will be
"  # & (

+
*
#.& /*+,-
curate routing instructions as it enters sectors with a higher replaced by eTf , and will be replaced

d
refresh rate. by e-f .
After the routing messages are examined, node rebuilds
3.3 Pseudo code for FSR 
the routing table based on the newly computed topology ta-
ble. Node then exchanges the latest link state informa-

  
Fig. 3 provides FSR pseudo code. Initially, each node tion with its neighbors. Procedure RoutingUpdate(i) scans
   
8g

" 
starts with an empty neighbor list , and an empty topol- through the topology table according to the distance
  ) 
g hg

  $ # %'&
ogy table . After node initializes its local^J_avariables
`cb ;d
between and . If 8g is within the range of fisheye
with proper values as described in procedure ]\ , scope level i , will be included in the update
it learns about its neighbors by examining the sender ID message. Note that the update interval for entries which
  
^Jl d _m`b d _anmoJl
of each received packets. That is, assuming that all nodes belong to fisheye scope level i is jke ihp . FSR


can be heard by are ’s neighbors, node adds all routing uses different exchange intervals for different entries in the

packet senders to its neighbor list, .
Node then invokes PktProcess(i,pkt) to process the
table. To be precise, entries corresponding to nodes that are
nearby (within a predefined scope) are propagated to the
neighbors more frequently than entries of nodes that are far
away. 
!  
shortest path routes. Dijkstra’s algorithm requires typically
steps to compute the shortest paths from one source
FindSP(i) creates a shortest path tree rooted at . In prin-
ciple, any existing shortest path algorithm can be used to
!     [19].
!  
to all destinations, although it is possible to reduce it to
memory space is required to store
!  
]i\

create the tree. In this paper, however, the procedure listed the network topology represented by a connection matrix.

 0  
in Fig. 3 is based on the Dijkstra’s algorithm [19] with mod- As for DBF, it has complexity of for computing and


ifications so that the next hop table ( ) and the dis- memory, as it only keeps the distance information for each
tance tables( ) are computed in parallel with the tree re- destination, and computes shortest paths in a distributed
construction. 




 fashion.

 
At node , FindSP(i) initiates with , then it it- For the line b overhead, in FSR each node broadcasts in-
     
^

     6
erates until . In each iteration, it searches for a formation for links on average for nodes in the
  
 -


`

!    # . In
node such that node minimizes the value of f scope of level with update interval . So the to-

_ d

  ( 8 
b ^
, for all and f , where ,f
  
 
f

 0        "
$!   #(#,# % . So

`

the case of GSR,   is equal to for

and weight(k,j) . Once node is found,_ 7isd aug- tal line overhead LO for FSR will be
0 (

mented with ,

is assigned to f f

and
 
 
is assigned to

f . That is, as the
%    #
&    #
(' # . LS, on the other
b ^

 "!   "!
` b ^ ` ^ `
shortest path from to has to go through f , the successor
for to is the same successor for to f .
hand, has similar accumulated `
data size for each link up-
3.4 Complexity date, but its update interval may become extremely small
when mobility increases.
In addition, as LS transmits one short packet for each
!  
In this section, we analyze the complexity of the FSR
scheme and compare it with that of other three routing link update, its control packet complexity CO can be as high
schemes: GSR, DBF and LS. Note that the GSR is a the as when the mobility is high. On the other hand, both
FSR and DBF transmit a fixed number of update tables us-

)!* 
special case of FSR, where we have only one scope and the
radius is the the diameter of the network. ing longer packets to optimize the MAC throughput. Thus,
The complexity is studied under five aspects: CO .
 Computation Complexity (CC): the number of com-
Lastly, the convergence time for FSR is also superior
than that for DBF. In fact, if the short update interval is used,
putation steps for a node to perform a routing compu- FSR can converge as fast as LS.


tation after an update message is received;

Memory Complexity (MC): the memory space re- 4 Performance Evaluation


quired to store the routing information;
4.1 Simulation model and methodology
Line Overhead (LO): the aggregate volume of control


bytes exchanged by a node per unit time. We implement our routing scheme in a multihop, mobile
wireless network simulator using the parallel discrete-event
Control Packet(CO): the average number of routing
simulation language PARSEC [2]. In most of experiments


packets exchanged by a node per unit time.
unless specified, the network consists of 100 nodes roam-
Convergence Time (CT): the time required to detect ing randomly in a 1000x1000 meter square. The roaming
a link change. area for network sizes 50, 200, 400 and 1000 is 700x700,
1500x1500, 2000x2000 and 3000x3000 meter square re-

 
Table 1 shows the results of our comparison. In the table, spectively. The radio transmission range is 120 meters and
denotes the number of nodes in network ( ), denotes channel capacity is 2 Mbits/sec. A free space propagation
^
the maximum
`
hop distance, the diameter, in the network, channel model is assumed. We use IEEE 802.11 MAC pro-

+
  % %
and denote the degree of node connectivity and the rout- tocol with Distributed Coordination Function (DCF) [9] as

b
ing update interval, respectively. is the average number the MAC layer in our experiments. The random waypoint

 
of nodes in scope , where ... ( is the number model [4, 6] was used in the simulation runs. In this model,
of scope levels used in FSR). is the damping factor of a node selects a destination randomly within the roaming
update frequency for level . area and moves towards that destination at a predefined
FSR and LS have same memory complexity and com- speed. Once the node arrives at the destination, it pauses
putation complexity as both maintain the topology for the at the current position for 5 seconds. The node then selects
whole network and use Dijkstra’s algorithm to compute another destination randomly and moves towards it, pausing
 !  
Protocol CC MC LO CO CT
!   !     #
b ^
!   ! ' 
!  "! '
` `

!   !   # !   ! ' 

FSR

!   !   !   '  # !   ! 

^ ` `
GSR
!  !   !   # !   !  ' 
^ `


LS
` `
DBF

Table 1. Complexity Comparison

0.9
there for 5 seconds, and so on. Note that the pause time is DBF
0.8 LS
not considered in computation of node speed. Each simula- GSR
FSR R=1
tion executed for 600 seconds of simulation time. Multiple 0.7 FSR R=2
FSR R=3

Weighted Inaccuracy
runs with different seed numbers were conducted for each 0.6

scenario and measurements were averaged over those runs. 0.5

0.4

4.2 Performance Measurements 0.3

0.2
The performance measures monitored in our study are: 0.1
(a) weighted routing inaccuracy; (b) control overhead 0
(O/H); and (c) packet delivery ratio. The variables are: mo- 0 10 20 30 40 50 60 70
mobility (km/hr)
bility, number of nodes, update interval, and; fisheye scope
radius. For FSR, we use 2-level fisheye scoping in our ex- Figure 4. Inaccuracy vs Mobility
periments.

4.2.1 Weighted Routing Inaccuracy

The route update interval also has impact on routing in-
Routing inaccuracy is checked by comparing the next hop accuracy. As shown in Fig. 5 (scope radius is 2), FSR ac-
table of each node with the tables generated by an off-line curacy degrades as the update interval increases. Intuitively
algorithm. This off-line algorithm has knowledge of the in- one would expect that the higher the speed, the smaller the
stantaneous network topology and computes optimal routes. interval to achieve a given accuracy. This is true for low
For a destination which is far away, an incorrect value in the speeds. Above a certain speed threshold (say, 20 km/hr in
next hop table is less critical than for a destination close by. Fig. 5) the accuracy is relatively insensitive to speed.

Considering this, we define the weighted routing inaccuracy Fig. 6 reports the inaccuracy of FSR with different


for node , , as: scopes v.s. network size. The mobility is 50 km/hr. For

"
  
  "   large network size, increasing the radius will improve the

 \3e f accuracy but not very dramatically. This is because in a


 
 
  
 mobility environment, a change on a link far away from the

   isis thethe diameter

where of the network.
   and bQ_ gKd
f
source does not necessarily cause a change in the routing
table at the source. Moreover, the fact that the route error
 is weighted by distance obviously reduces the sensitivity to
\3e f next hop and hop distance to destination

inaccuracy is computed by averaging   ,

calculated by the off-line algorithm respectively. Finally, network size. Thus, receiving updates about far away nodes
  routing
f

the overall at low frequency will not significantly affect the routing ac-
for all . curacy. On the other hand, using larger radii will cause more
Fig. 4 shows the inaccuracy of different routing schemes control O/H as shown in next section.
at different average speeds. LS performs best at all speed
ranges, since it reacts the fastest to topology changes. The 4.2.2 Control Overhead
inaccuracy for FSR decreases with the increases of scope



radius. However, FSR still performs better than DBF even Fig. 7 plots link control O/H as a function of network size.
with radius due to the fact nodes are exchanging link Since GSR exchanges full size link state vectors, the con-
state information among each other and thus keep track of trol O/H grows linearly with the network size. In contrast,
the entire network topology. the control O/H is greatly reduced in FSR. Note that the
 0.9
T = 1 sec GSR
T = 2 sec 0.5 FSR R=1
0.8 FSR R=2
T = 3 sec
T = 4 sec FSR R=3
0.7

Control O/H (Mbps/node)

0.4
Weighted Inaccuracy

0.6

0.5 0.3

0.4
0.2
0.3

0.2
0.1
0.1

0 0
0 10 20 30 40 50 60 70 0 100 200 300 400 500 600 700 800 900 1000
mobility (km/hr) Number of nodes

Figure 5. Inaccuracy vs Update Interval Figure 7. Control O/H vs network size

0.6
GSR
FSR R=1 rapidly degrades for size larger than 100 nodes, as shown in
0.55 FSR R=2
FSR R=3
Fig. 9. When network size grows large, routing O/H will
cause considerable performance degradation of the GSR.
Weighted Inaccuracy

0.5
The advantage of FSR is clearly shown as FSR maintains
0.45 high delivery ratio across different network sizes.
0.4 1

0.35
0.8

0.3
Delivery Ratio

0 100 200 300 400 500 600 700 800 900 1000 0.6
Number of nodes

Figure 6. Inaccuracy vs network size 0.4

0.2 LS
GSR
FSR R=3
average number of neighboring nodes is independent from
0
network size since node density is kept constant. The reason 0 10 20 30 40 50 60 70
mobility (km/hr)
why FSR reduces O/H is that only a fraction of the entries
are updated each time. In a two-level fisheye hierarchy, the Figure 8. Delivery ratio vs mobility (for 100
smaller radius, the smaller fraction of entries updated in the nodes)
“fast” interval, and the lower the control O/H. The tradeoff
between routing accuracy and control O/H must be taken
into account when choosing the scope radii of the fisheye
solution.
5 Conclusions
4.2.3 Packet Delivery Ratio
The packet delivery ratio is the ratio of data packets deliv- In this paper, we present a new routing scheme, Fisheye
ered to the destinations versus data packets originated by State Routing, which provides an efficient, scalable solution
the sources. This number presents the routing effectiveness for wireless, mobile ad hoc networks. The routing accuracy
of a protocol. Fig. 8 shows the packet delivery ratio as func- of FSR is comparable with an ideal LS scheme and the rout-
tion of node mobility. As node mobility increases, the per- ing overhead is kept low. As a results, FSR is more desir-
formance of the Link State is dramatically degraded due to able for large mobile networks where mobility is high and
flooding O/H. While the routing tables are maintained accu- the bandwidth is low. By choosing proper number of scope
rate, there is not much bandwidth left for data traffic. GSR levels and radius size, FSR proves to be a flexible solution
faces better than LS; in fact, it performs as well as FSR for to the challenge of maintaining accurate routes in ad hoc
small to moderate size networks. However, performance networks.
 1
LS [8] Z.J. Haas and M. R. Pearlman “Determining the Opti-
0.9 GSR
FSR R=3 mal Configuration for the Zone Routing Protocol,” In
0.8
IEEE Journal on Selected Areas in Communications,
0.7
Aug. 1999, pp. 1395-1414.
Delivery Ratio

0.6

0.5 [9] IEEE Computer Society LAN MAN Standards Com-

0.4 mittee, Wireless LAN Medium Access Protocol (MAC)
0.3 and Physical Layer (PHY) Specification, IEEE Std
0.2
802.11-1997. The Institute of Electrical and Electron-
0.1
ics Engineers, New York, NY, 1997.
0
0 100 200 300 400 500 600 700 800 900 1000
[10] A. Iwata, C.-C. Chiang, G. Pei, M. Gerla, and T.-W.
Number of nodes Chen, “Scalable Routing Strategies for Ad-hoc Wire-
less Networks,” In IEEE Journal on Selected Areas in
Figure 9. Delivery ratio vs network size (speed
Communications, Aug. 1999, pp. 1369-1379.
= 50 km/hr)
[11] E.D. Kaplan (Editor), Understanding the GPS: Prin-
ciples and Applications, Artech House, Boston, MA,
Feb. 1996.
References [12] L. Kleinrock and K. Stevens, “Fisheye: A Lenslike
Computer Display Transformation,” Technical report,
[1] The ATM Forum, “Private Network-Network Inter- UCLA, Computer Science Department, 1971.
face Specification v1.0,” 1996. [13] Y.-B. Ko and N.H. Vaidya, “Location-Aided Routing
(LAR) in Mobile Ad Hoc Networks,” In Proceedings
[2] R. Bagrodia, R. Meyer, M. Takai, Y. Chen, X. Zeng, J. of ACM/IEEE MOBICOM’98, Dallas, TX, Oct. 1998,
Martin, and H.Y. Song, “PARSEC: A Parallel Simula- pp. 66-75.
tion Environment for Complex Systems,” IEEE Com-
puter, vol. 31, no. 10, Oct. 1998, pp.77-85. [14] J. Moy, “OSPF Version 2,” In IETF RFC 1583, 1994.
[15] S. Murthy and J.J. Garcia-Luna-Aceves, “An
[3] S. Basagni, I. Chlamtac, V.R. Syrotiuk, and B.A.
Efficient Routing Protocol for Wireless Networks,”
Woodward, “A Distance Routing Effect Algorithm for
ACM/Baltzer Mobile Networks and Applications, vol.
Mobility (DREAM),” In Proceedings of ACM/IEEE
1, no. 2, Oct. 1996, pp. 183-197.
MOBICOM’98, Dallas, TX, Oct. 1998, pp. 76-84.
[16] V.D. Park and M.S. Corson, “A Highly Adaptive Dis-
[4] J. Broch, D.A. Maltz, D.B. Johnson, Y.-C. Hu, and J. tributed Routing Algorithm for Mobile Wireless Net-
Jetcheva, “A Performance Comparison of Multi-Hop works,” In Proceedings of IEEE INFOCOM’97, Kobe,
Wireless Ad Hoc Network Routing Protocols,” In Pro- Japan, Apr. 1997, pp. 1405-1413.
ceedings of ACM/IEEE MOBICOM’98, Dallas, TX,
[17] C.E. Perkins and P. Bhagwat, “Highly Dynamic
Oct. 1998, pp. 85-97.
Destination-Sequenced Distance-Vector Routing
[5] T.-W. Chen and M. Gerla, “Global State Routing: A (DSDV) for Mobile Computers,” In Proceedings of
New Routing Scheme for Ad-hoc Wireless Networks,” ACM SIGCOMM’94, London, UK, Sep. 1994, pp.
In Proceedings of IEEE ICC’98, Atlanta, GA, Jun. 234-244.
1998, pp. 171-175. [18] C.E. Perkins and E.M. Royer, “Ad-Hoc On-Demand
Distance Vector Routing,” In Proceedings of IEEE
[6] D.B. Johnson and D.A. Maltz, “Dynamic Source WMCSA’99, New Orleans, LA, Feb. 1999, pp. 90-100.
Routing in Ad Hoc Wireless Networks,” In Mobile
Computing, edited by T. Imielinski and H. Korth, [19] R. Sedgewick,Weighted Graphs, chapter 31,
Chapter 5, Kluwer Publishing Company, 1996, pp. Addision-Wesley, 1983.
153-181. [20] C.-K. Toh, “Associativity-Based Routing For Ad Hoc
Mobile Networks,” Wireless Personal Communica-
[7] Z.J. Haas, “A New Routing Protocol for the tions Journal, Special Issue on Mobile Networking
Reconfigurable Wireless Networks,” In Proceedings and Computing Systems, Kluwer Academic Publish-
of IEEE ICUPC’97, San Diego, CA, Oct. 1997, pp. ers, vol. 4, no. 2, Mar. 1997, pp. 103-139.
562-566.