TABLE OF CONTENTS

EXECUTIVE SUMMARY ................................................................................................................... 2

1. INTRODUCTION .......................................................................................................................... 4
1.1 PAST AND CURRENT INITIATIVES .................................................................................. 4
1.2 OBJECTIVES AND APPROACH ........................................................................................ 5

2. ENERGY USE IN DATA CENTRES .................................................................................................. 6

2.1 ENERGY SOURCES .......................................................................................................... 6
2.2 FACILITY SYSTEMS ......................................................................................................... 6
2.2.1 Cooling .............................................................................................................. 8
2.2.2 Electrical Distribution ....................................................................................... 9
2.3 INFORMATION TECHNOLOGY (IT) SYSTEMS ............................................................... 10
2.3.1 Hardware ........................................................................................................ 10
2.3.2 Software ......................................................................................................... 11
2.4 DESIGN AND DEPLOYMENT ......................................................................................... 12
2.4.1 Multi‐Tiering ................................................................................................... 13
2.4.2 Modular Provisioning ..................................................................................... 13

3. TECHNOLOGY DEVELOPMENT ...................................................................................................15

3.1 DIRECTIONS IN RESEARCH AND INNOVATION ............................................................ 15
3.1.1 Facility Systems .............................................................................................. 15
3.1.2 IT Systems ....................................................................................................... 16
3.1.3 Integration ...................................................................................................... 18
3.2 COMPARATIVE IMPACT ASSESSMENT ........................................................................ 19
3.3 RESULTS AND ANALYSIS .............................................................................................. 19

4. CONCLUSION.............................................................................................................................21

5. REFERENCES ..............................................................................................................................22

6. APPENDIX .................................................................................................................................23

1
EXECUTIVE SUMMARY

The Green Data Centre Technology Roadmap sets out a framework to improve data centre
sustainability. The roadmap aims to reduce energy consumption and improve the energy
efficiency of the constituent systems of a data centre – facilities and IT. It assesses and
makes recommendations on potential directions for Research, Development &
Demonstration (RD&D) to improve the energy efficiency of data centres in Singapore’s
context.

Most data centres in Singapore, many in the middle of their lifespan, were designed and
constructed without sustainability and energy conservation in mind. Looking ahead, the
Singapore data centre industry is expected to experience strong and sustained growth.
Consequently, the Roadmap covers the green initiatives that span existing data centres and
new data centres.

Energy consumption in data centres can be broadly categorised into two categories – energy
consumption by IT systems and energy consumption by facility systems (i.e. supporting
infrastructure). Three main areas examined in this roadmap are facility, information
technology (IT) systems, and an integrated approach to design and deployment of data
centres.

Most of the global effort to improve the energy efficiency of data centres has centred on
facility systems. Of facility systems, cooling has received the most attention as it is generally
the single largest energy overhead. Singapore’s climate, with its year‐round high
temperatures and humidity, makes cooling particularly energy‐intensive compared to other
data centre hubs globally. The following technologies to improve the energy efficiency of
facility systems are assessed:

 Direct liquid cooling

 Close‐coupled refrigerant cooling

 Air and cooling management

 Passive cooling

 Free cooling (hardening of IT equipment)

 Power supply efficiency

Notwithstanding the importance of improving the energy efficiency of powering and cooling
data centres, the current focal point for innovation is improving the energy performance of
physical IT devices and software. Deficiencies in these areas provide opportunities for
innovation that would greatly improve the sustainability of data centres. The following
technologies to improve the energy efficiency of IT systems are assessed:

2
  Software power management

 Energy‐aware workload allocation

 Dynamic provisioning

 Energy‐aware networking

 Wireless data centres

 Memory type optimisation

An integrated approach to design and deployment is absent in the data centre sector. The
energy performance of a data centre depends on the complex interplay between its IT
systems and facility (i.e. non‐IT) systems. However, the two are often deployed with little
regard for one another. This silo‐ed approach to design and deployment often results in
wasteful over‐provisioning. Technology vendors have offered Data Centre Infrastructure
Management (DCIM) tools as a way for end users to address the silo‐ed nature of data
centre deployments. However, very often the DCIM solutions themselves remain limited to
the domains of their respective vendors, restricting their use to only piecemeal
optimisations. The Roadmap explores future directions in advanced DCIM to enable the
integration and automation of the disparate systems of the data centre.

Owing to the mission critical nature of data centre operations, end user risk aversion is the
largest barrier to implementing more effective energy strategies in data centres. It is
recommended that the most effective way to address risk aversion and promote the
adoption of new energy efficient technologies is collaboration between the public bodies
responsible for energy efficiency and the industry. To this end, proof‐of‐concept
demonstrations are essential if the adoption of new technologies is to be fast‐tracked in
Singapore.

3
1.INTRODUCTION

Singapore leads Southeast Asia as a data centre hub, accounting for more than 60% of the
region’s data centre market. Buoyed by political and economic stability, well‐developed
telecommunications infrastructure, the presence of a large number of multinational
companies, and government initiatives such as the planned Data Centre Park, the data
centre industry is expected to continue to experience strong growth.

The data centre industry in Singapore and many developed countries are subject to intense
scrutiny due to the high proportion of energy costs in the context of overall cost of
ownership.

To identify new technologies relevant to Singapore, the Infocomm Development Authority

(IDA) of Singapore appointed i3 Solutions Group to prepare the Green Data Centre
Technology Roadmap (the ‘Roadmap’) with input from the research community, industry,
and government agencies. The Roadmap provides insights on methods and systems to
reduce energy consumption in data centres without compromising system performance and
security requirements. It also identifies key areas of research where academic institutions
can take a lead role in developing emerging technologies.

1.1 PAST AND CURRENT INITIATIVES

In 2009, the Infocomm Development Authority of Singapore (IDA) initiated the Green ICT
programme with the aim of improving the energy efficiency and competitiveness of the
Singapore data centre industry.

Investment Allowance Scheme: The Investment Allowance was introduced to help data
centre operators improve the energy efficiency of their existing data centres. Successful
applicants are granted an investment allowance of 30% to 50% off the fixed capital
expenditure incurred on retrofitting their data centres.

Green Data Centre Standard: Published in 2011, the Singapore Standard 564 (SS564) was
developed by the Green DC Standards Working Group under the industry‐led Information
Technology Standards Committee. The SS564 aims to benchmark the energy efficiency of
data centres, while providing a set of best practices for the industry.

BCA‐IDA Green Mark for Data Centres: Jointly developed by the IDA and the Building and
Construction Authority (BCA), the Green Mark is a rating system that encourages the
adoption of energy efficient design, operation and management of data centres.

There is a clear necessity for data centre owners and users to account for their energy usage
and adopt methods to reduce energy consumption. Although some efforts have already

4
been made through the IDA‐BCA Green Mark Standard and SS564, these have primarily
focused on the facility (i.e. non‐IT) systems of the data centre.

New technologies applicable to facility systems and IT systems will enable further
improvements in energy performance beyond the best‐in‐class today. It is proposed that
Singapore becomes an early adopter of beneficial new energy efficient technologies and
apply medium to long term targets for energy efficiency.

1.2 OBJECTIVES AND APPROACH

The Green Data Centre Technology Roadmap considers how the carbon footprint of data
centres in Singapore can be reduced. This is done in the context of ensuring the continued
growth of the industry and enhancing Singapore’s position as a data centre hub for the Asia
Pacific region.

The desired outcomes of the Roadmap are:

 Development of a framework for the research community, the industry and the
Singapore Government to assess their technology options in reducing the carbon
footprint of the data centre industry.

 Identification of directions in Research, Development & Demonstrations (RD&D) that

will enable Singapore to achieve a leading position in Green Information and
Communications Technology (ICT) technology development.

The assessment of the suitability of individual technologies for Singapore data centres is
primarily based upon the extent of current product development and the relative impact
each technology has on energy consumption and sustainability. It should be noted that some
technologies complement each other to different extents, while others are mutually
exclusive. Additionally, some technologies are best applied to new data centres, retrofitted
to existing data centres or applied equally to both new and existing data centres.

The Roadmap presents recommendations on technological directions to improve data centre

energy performance. They are quantified with indicative incremental adoption costs,
deployment curves, abatement potential and marginal abatement cost.

Information presented has largely been gathered from the annotated reference sources. The
unique nature of this report requires considerable input based on expert opinion. Where
expert opinion is presented due to the lack of external references, explanations of the
reasoning and assumptions are included.

5
2.ENERGY USE IN DATA CENTRES

Data centres comprise three main energy groups as shown in Figure 2.1. Each group can
contribute to improving the sustainability and energy efficiency of the data centre.

Figure 2.1: Data Centre Energy Groups.

2.1 ENERGY SOURCES

While a comprehensive discussion of energy sources is beyond the scope of this report, a
short treatment is nonetheless included here for completeness.

Energy sources may reside outside the data centre (in the case of power from a utility
company) or within the data centre (in the case of on‐site generation options). In either
case, data centres select their energy sources to achieve the best balance of cost‐
effectiveness and reliability.

Internationally, many large internet companies have made using renewable or green energy
sources for their data centres a priority. However, the scale, climate and tidal range of
Singapore restricts the use of geothermal, tidal, hydro and wind, either as viable sources of
energy or for use as economisers for data centres. Approximately 96% of energy supplied in
Singapore is fossil fuel based, derived from natural gas and other petroleum products [1].

2.2 FACILITY SYSTEMS

Facility systems refer to the infrastructure and processes needed to support the operation of
IT equipment in a data centre. They comprise two principle components – cooling and
electrical distribution. To date, most industry initiatives have been focused on improving the

6
efficiency of data centre facility systems. This has been driven by the high proportion of
facility system energy costs relative to the entire data centre. This is illustrated in Figure 2.2.

IT systems Facility 37%

51% 49%

12%

IT Systems Cooling Electrical Distribution

Figure 2.2: Energy profile of a typical data centre’s Facility Systems in Singapore.

Power Usage Effectiveness (PUE) [2] is the primary metric used to measure the energy
efficiency of a data centre’s facility systems. PUE is calculated as follows:

TOTAL DATA CENTRE POWER

PUE
IT EQUIPMENT POWER

The ideal PUE is 1.0; in this case, all power supplied to the data centre reaches the IT
equipment. In practice, power is required to provide cooling to the data centre. Power losses
also occur in electrical conversion and distribution systems, and ancillary systems within the
data centre. Numerous claims of low PUE have been made by companies around the world,
with some stating PUE ratings as low as 1.05.

A 2012 site measurement study surveyed 23 data centres in Singapore and recorded an
average PUE of 2.07 [3]. These findings are comparable to the average PUE ratings of 2.2 in
the United States and 2.02 in Europe. However, a separate industry survey of 100
respondents in 2013 reported an average PUE of 2.61 [4]. A possible explanation for this
discrepancy is self‐selection by larger data centre operators in the 2012 Singapore study. On
the other hand, the 2013 survey was able to capture a larger cross section of the data centre
landscape.

Apart from PUE, other efficiency metrics include Water1 and Carbon2 Usage Effectiveness [5,
6]. Currently, only 24% of organisations globally collect data on carbon emissions and 34%

1
WUE™ is the annual site water usage divided by IT equipment energy. This includes water used for
humidification and water evaporated on‐site for energy production or cooling of the data centre.

7
collect data on water usage [7].The Green Data Centre Primer further elaborates data centre
performance metrics [8].

2.2.1 Cooling

On average, 37% of the total energy consumed by data centres in Singapore is used to cool
IT equipment (Figure 2.2). There are many ways to reduce the energy involved in cooling.
However, Singapore’s climate poses a major disadvantage. The high temperatures and
humidity of Singapore’s tropical climate makes cooling data centres significantly more
energy‐intensive compared to other locations in the world.

The majority of data centres today operate the cooling systems of their data centres within
the guidelines set by ASHRAE (formerly the American Society of Heating, Refrigerating and
Air Conditioning Engineers). The guidelines set recommended operating envelopes with
regard to temperature and humidity for various equipment classes (Figure 2.3) [9].

Figure 2.3: ASHRAE‐recommended operating envelopes for data processing environments (Source: ASHRAE).

Current guidelines from ASHRAE for the upper limits measured at the air intake of IT
equipment are detailed in Table 2.1. In practice, most data centres operate at temperatures
well below the recommended upper limit set in the guidelines, resulting in high cooling
costs.

Table 2.1: ASHRAE operating ranges for Class A1 equipment.

2
CUE™ is the total CO2 emissions caused by the total data centre energy divided by IT equipment energy. CUE
is a source based metric and includes carbon generated in the energy distribution path.

8
 Recommended Allowable
o
Dry Bulb Temperature 27 C 32oC
Dew Point 15oC Not specified
Relative Humidity 60% 80%

The guidelines also provide an assessment of the increase in IT equipment failure rates
associated with various operating temperatures. This is detailed in Table 2.2. The most
common failure modes in IT equipment are power supply failure and storage device failure.

Table 1.2: ASHRAE IT equipment failure rates at various operating temperatures.

Inlet Dry Bulb (oC) Increase in Failure Rate

20 0%
22 13%
25 24%
27.5 34%
30 42%
32.5 48%

The increase in failure rates at temperatures even well below the upper temperature limit is
significant to mission critical data centres, where higher failure rates are undesirable. Failure
rates published by ASHRAE are likely to deter organisations from adopting higher operating
temperatures. Further RD&D is recommended to test the failure rates stated by ASHRAE.
Vigorous quantification of failure rates will encourage data centres to operate at higher
temperatures, thus reducing cooling needs.

2.2.2 Electrical Distribution

The most significant component affecting the energy efficiency of the electrical system is the
UPS (uninterruptible power supply). Various UPS types are used in Singapore; the most
common type being the on‐line double conversion UPS. Assuming a typical 2N UPS
configuration and an average UPS utilisation of 30%, typical of a mature data centre, the
electrical system efficiency is likely to be approximately 89%; excluding transformer and
distribution losses.

An alternative to the on‐line mode of operation is the off‐line mode where the UPS’ normal
condition is off and power flows through the static bypass. Operating off‐line can improve
efficiency up to 98%. Distribution losses, which account for around 37% of the average
losses in the electrical system, can be reduced, though not to the same extent as the UPS
system.

9
Another method to reduce losses in the electrical distribution system is the use of DC power
over the current AC power distribution systems. The inherent benefit that DC data centres
have over AC data centres is a reduced component count. This inevitably results in a higher
reliability configuration since there are fewer conversion steps compared to an AC system.
Typical AC data centres in Europe and Asia (excluding Japan) have five conversion steps
associated with normal power delivery to the server while DC systems require only three
steps. As a result DC systems have less equipment usage, less heat generation and increased
reliability because there are fewer complex components and potential points of failure.

2.3 INFORMATION TECHNOLOGY (IT) SYSTEMS

2.3.1 Hardware

Information technology (IT) infrastructure plays a major role in the energy efficiency of a
data centre. Of the various physical IT components, servers account for the majority of
physical IT energy use (Figure 2.4). This is followed by storage devices and network devices.

5%
10%
Facility IT Systems
49% 51%
36%

Facility Systems Network Storage Servers

Figure 2.4: Energy Profile of a Typical Data Centre’s IT Systems in Singapore.

Efforts have been made to reduce the energy consumption of servers and increase their
energy efficiency. Techniques such as energy‐aware workload management tools, voltage
and frequency scaling and power capping have been introduced by hardware
manufacturers. Currently the most significant issue affecting server energy performance is
the power used at zero processor utilisation. Intuitively one would expect this to be zero or
close to zero; instead the average is 50% of the maximum power consumption [10].
Techniques such as device power‐down and invoking processor sleep states can address this
problem. The drawback of this is a latency penalty as additional time is required to wake
from a sleep state in response to demand spikes.

Storage capacity is the fastest growing segment of physical IT equipment. The emergence of
cloud computing, in particular the public cloud is driving an extraordinarily high growth rate
in storage. Globally storage capacity is expected to grow to approximately 40YB3 by 2020
[11]. Some measures have been taken by manufacturers to address energy usage, such as

3 1YB= 1024 bytes

10
using different drive technologies and new techniques for deploying data storage. There
are, however, many aspects of storage research that can be addressed in the context of
sustainability and energy efficiency.

Contemporary data centres use both fibre‐optic and copper cables for networking. As
network speeds increased in conjunction with the commoditisation of fibre products, a
larger proportion of fibre is used in the data centre. Similarly, advances in fibre and copper
technology have enabled higher throughputs across the data centre network. In parallel,
network switches have become more energy efficient through a combination of improved
power supply efficiency and component design. Energy saving techniques involving the use
of intelligent routing methods, node power‐down and sleep states could potentially realise
significant network energy savings.

2.3.2 Software

Software that is optimally configured for energy performance directly reduces the energy
consumed by IT equipment and has a cascading effect on the energy consumption of
downstream devices. The Software‐to‐Facility Energy Cascade Effect makes software one of
the most important areas to address in terms of reducing energy consumption. This is
illustrated in Figure 2.5.

Figure 2.5: The Software‐to‐Facility Energy Cascade Effect.

With the exception of mobile platform designers, for whom extending the battery life is a
major factor in platform software and application design, most applications and operating
system developers do not consider how energy efficiency and usage is impacted by
software. In the case of fixed platforms, the impact of energy‐aware language structures,
coding and compilers have largely been overlooked by software developers.

11
The industry trend towards cloud based services (IaaS, PaaS and SaaS) particularly in public
and hybrid cloud environments will inevitably become more energy‐focused. Cloud service
providers operate a usage‐based charging model for software and infrastructure. Since
energy costs usually account for a large proportion of a data centre’s operating expenditure,
cloud providers will have to prioritise energy efficiency to maximise their return on
investment.

2.4 DESIGN AND DEPLOYMENT

Technologies should be complemented with design and deployment best practices to

ensure that they are deployed to their fullest effect. In most cases, the focus has been on
“right‐sizing” data centre deployments – meeting IT demands without over‐provisioning
resources.

An essential and often overlooked factor in the design of data centres is part‐load
performance. Usually, energy efficiency is considered at full design load, but this is rarely
relevant since it does not convey the progressive increase in data centre occupancy and
hence energy use of the data centre.

Four factors that limit data centre capacity are physical space, cooling, power and network
connectivity. Poor design and operational practices often fail to consider that these factors
can grow at significantly different rates. This results in one of these design factors creating a
bottleneck that prevents the data centre from realising its full design capacity and return on
investment. The problem is further compounded by the use of fault tolerant UPS systems.
At 2N topology4, the UPS rarely operates above 50% utilisation. In reality, it would spend
most of its tenure operating at utilisation levels between 10% ‐ 30%. It is widely
acknowledged by industry experts that IT and facility departments are not aligned to
common objectives when it comes to the data centre. IT departments cannot predict their
infrastructure requirements beyond 3‐5 years, yet data centre construction capital
expenditure is usually amortised over 15 to 20 years. This problem is referred to as the
Technology‐Facility Paradox.

In a typical scenario, during the initial stage of data centre planning, IT departments
approximating future requirements tend to over‐provision. The problem is exacerbated by
the MEP (mechanical, electrical and plumbing) engineering teams who, due to their lack of
understanding of IT infrastructure, further over‐provision with the use of conservative
designs. The outcome is unnecessarily high capital and operating expenditures for data
centre deployments.

4
2N topology is a UPS architecture providing multiple‐redundancy. In this case, each power supply is
connected to its own UPS, providing protection from both power supply failure and failure of the UPS.
12
2.4.1 Multi‐Tiering

Current data centre designs are generally wasteful in terms of both embedded energy and
operational energy use. A typical data centre comprises a single tier of MEP design that is
applied uniformly across the computer floor. Nonetheless, from the end‐user’s perspective,
not all service‐lines or applications are of the same criticality. Single‐tier data centres offer
unnecessarily high redundancy for applications that do not require it.

Multi‐tiering is the design of data centres to provide multiple levels of reliability that
commensurate with service‐line or application criticality, the underlying application and
technology infrastructure. A simple two‐tier arrangement is shown in Figure 2.6. Multi‐tiered
data centres avoid redundancy, which reduces capital costs and fixed losses, and when used
in conjunction with a progressive modular design, avoids the over‐provisioning of MEP
systems. The resulting data centre is also more adaptive to IT changes, and is able to provide
variable and scalable availability.

Figure 2.2: A Simple Two‐Tier Data Centre Deployment (Source: HP)

2.4.2 Modular Provisioning

Over‐provisioning occurs when the capacity of MEP systems deployed far exceed the needs
of the data centre’s IT load. This results in energy wastage due to fixed losses. Fixed losses
are inherent to all electrically powered equipment. They occur when equipment is powered
on and remain constant irrespective of whether the equipment is utilised.

Modular provisioning is the design of data centres to allow for IT and MEP resources to be
deployed in progressive stages. In turn, MEP systems are able to better match the actual IT
load. This approach to designing and deploying data centres has significant benefits in terms
of deferred capital expenditure and operating expenditure. Energy costs can be reduced by

13
the elimination of fixed losses. Figure 2.7 illustrates the improved efficiency profile of
modular deployments compared to conventional monolithic deployments.

Figure 2.3: Efficiency of Modular compared to Non‐Modular Deployments [12].

14
3.TECHNOLOGY DEVELOPMENT

3.1 DIRECTIONS IN RESEARCH AND INNOVATION

In this section, we review potential research directions in facility, information technology

(IT) systems, and integrated approach to design and deployment of data centres. A total of
13 technology areas are presented and analysed. The areas were selected in consultation
with the research community, industry, as well as with input from government agencies.
Together, they represent the major technological developments in data centre energy
efficiency pursued today.

3.1.1 Facility Systems

The high cost of cooling is Singapore’s main competitive disadvantage when it comes to
hosting a data centre. Even in locations with favourable climates, cooling is relatively
expensive. Apart from operating temperatures closer to the upper limit of the ASHRAE
recommendations, energy efficiencies of data centres can be improved by reducing cooling
needs. Cost of cooling can be reduced by improved cooling technologies and reducing the
need for active cooling. Technologies to enable these two approaches are elaborated below.

Direct Liquid Cooling or immersion cooling is the use of dielectric liquids as alternatives to
air as cooling mediums. Some electrical transformers and supercomputers have used
dielectric fluid submersion for many years. Until recently, this method involved higher costs
compared to air cooling. Direct liquid cooling is similar to direct‐to‐chip water cooling,
except that unlike direct‐to‐chip water cooling, the dielectric fluid passes over all surfaces
inside the IT equipment. The fluid is then transferred to an external heat exchanger where it
is cooled for recirculation back to the IT equipment. Trials involving immersion cooling have
demonstrated significant energy savings. However, perceptions of increased complexity and
restrictions on the type of hardware that may be used have hindered widespread
deployment.

Close‐Coupled Refrigerant Cooling is the use of liquid refrigerants with micro‐channel heat
exchangers to bring cooling as close as possible to the actual heat sources. Beyond being
more efficient than air as a cooling medium, the close proximity of the cooling and heat
source allows for greater precision in the application of cooling. Thus data centres do not
need to be overcooled to address hotspots. Trials have reported energy efficiency
improvements of 80% over traditional water cooled chillers.

The following are measures to reduce the need for energy‐intensive cooling.

15
Air and Cooling Management refers to the broad suite of technologies and accompanying
practices that allow for the optimisation of data centre cooling resources. It remains the
single largest near and medium term opportunity to improve data centre energy efficiency
due to the continued popularity of air as a cooling medium. Today, best practices with
regard to air and cooling management are well‐known, if inconsistently followed. Examples
of recent research efforts include the use of computational fluid dynamics (CFD) simulations
to aid in the design of racks and servers, to improve airflow and heat management.

Passive Cooling aims to eliminate or reduce the need for energy‐intensive active cooling
systems by improving passive heat transfer from chips, servers or racks. An example of this
is the use of heat pipes to reduce hotspots on chips or servers. Passive cooling can reduce
data centre cooling demand by up to 40%, significantly reducing data centre operating
costs. This potentially allows higher computational densities to be achieved without the
complexity and cost of liquid cooling architectures or immersion tanks.

Free Cooling (Hardening) refers to improving the protection of IT equipment against

adverse environmental conditions, namely particulate matter, temperature and humidity.
The need to cool the data centre has traditionally been a large energy overhead. Globally,
the data centre industry has responded by exploiting free‐cooling techniques in locations
where practicable and/or moving operations to those locations. Singapore’s hot and humid
tropical climate restricts the use of free‐cooling. Improving the environmental operating
envelop of IT equipment would allow data centres to be operated with lower levels of
cooling than is possible today.

With regard to power distribution, efforts are focused on reducing conversion losses or
avoiding conversions altogether.

Power Supply Efficiency refers to improving the efficiency of power transfer between the
energy source and IT equipment. This is largely a function of losses in electrical conversion.
Incentivised by industry programmes like 80 Plus, power supply efficiencies have improved
significantly since 2006, improving from 67% to over 90% at ideal loads.

3.1.2 IT Systems

Due to the Software‐to‐Facility Energy Cascade Effect (Figure 2.5), developments in software
energy efficiency can effect a magnified improvement throughout the data centre
technology stack. However, software has until recently, received little to no attention from
the data centre industry. Conversely, software energy efficiency has been a priority for the
mobile platform industry for many years.

To some extent, new research can take advantage of the gains made by mobile platforms.
However, there are profound differences in data types used by mobile and fixed platforms;

16
more work is needed to understand the techniques that can be translated and where new
techniques need to be developed.

Significant progress has been made in improving energy efficiency using dynamic voltage
and frequency scaling. However, there are numerous areas that require further research
and development to progress IT energy performance.

Suggested areas for research include:

Software Power Management, which refers to development and translation of techniques

to reduce computational demand by software when not performing useful work. Device idle
power consumption has improved considerably. However, it remains a major barrier to
energy proportional computing. The best levels of idle energy are approximately 25% of the
maximum power draw for the most recent generation of servers; the average is
approximately 50%. The use of sleep states during CPU idle‐time is fundamental to
increasing battery life in mobile platforms. Although the use of sleep states is widespread in
mobile devices these techniques are rarely used in fixed platform IT. This omission presents
a significant opportunity to reduce data centre energy consumption.

Energy‐Aware Workload Allocation is the automated migration of workloads in virtualised

environments to make better use of the most energy efficient servers or data centres across
a network. Where and when possible, workloads could potentially be consolidated onto the
most energy efficient servers or distributed in an energy efficient manner. Unused servers
can then either be put into sleep mode or (time permitting) be powered off. All this needs
to be achieved while maintaining quality‐of‐service and security standards.

Dynamic Provisioning is the automated scaling of a data centre’s compute, storage and
network resources, and by extension its energy expenditure, to match demand. This is a key
feature of energy proportionality in the data centre. The main challenge is dealing with the
inherent trade‐off between energy proportionality and latency in ramping up resource
availability in response to demand spikes.

Energy‐Aware Networking is the design and use of the network in a manner that minimises
its energy footprint. The need to ensure network resilience and meet demand spikes has led
to the practice of deploying a large number of devices to achieve this. The aim of energy‐
aware networking is the dynamic minimisation of network energy footprint while
maintaining performance and resilience.

Wireless Data Centres is the use of wireless networking in data centres to augment or
replace cabled communications. While interest in wireless data centres is mainly in cable
management (or the lack thereof) and the associated physical flexibility afforded, some
energy efficiency potential has been noted.

Memory Type Optimisation in an energy efficiency context determines how data is stored
and accessed using energy efficient technology whilst meeting execution quality‐of‐service

17
standards. Among the most significant developments are those in non‐volatile memory
devices, offering a significant impact on the energy efficiency of IT at silicon level due to its
inherent ability to retain its state under zero power. Depending on the workload, this could
significantly reduce energy consumption.

3.1.3 Integration

Sustainability in data centres must be approached in an integrated manner, with

opportunities pursued, at each level of the data centre technology stack. So far optimisation
efforts have restricted themselves to either the facility systems or IT systems. Further
opportunities are afforded by integrating facility systems, IT systems and overall alignment
with actual service‐line requirements.

Data Centre Infrastructure management (DCIM) market offerings are currently disparate
and not always comparable with each other. This is because the scope of DCIM is not
subject to any widely accepted definition, with many domains sitting under the broad DCIM
heading. The majority of DCIM toolsets do not offer a solution within every domain.
Domains typically include, but are not limited to:

 Utility power

 Mechanical and electrical plant

 Asset management

 Capacity management

 Space visualisation

 Simulation

Advanced DCIM should enable centralised monitoring, management, and intelligent

capacity planning of critical systems. Subsequent generations of data centres will
increasingly benefit from operational advantages that DCIM tools offer provided
implementations include the integration of IT, Facilities and Application Management

18
3.2 COMPARATIVE IMPACT ASSESSMENT

In this section, the environmental and economic potential of each technology described
above are assessed. The methodology used is described in the Appendix.

We consider the impact of each technology applied in isolation until 2030. This has the
following important implications for the calculations that follow: i) they are useful for
quantitatively assessing the relative impact of each technology, but ii) they are not useful
for determining the absolute quantitative impact of each technology.

The technologies are presented here solely in terms of their environmental and economic
potential. However, we must remain cognisant of the fact that technology development and
deployment is determined not just by technical potential, but also by business and market
considerations. However, the analysis presented here forms a useful starting point for the
research community and business leaders to assess their technology options for increased
sustainability of data centres in Singapore.

3.3 RESULTS AND ANALYSIS

The results of the calculations are illustrated in Figure 3.1. Most of the technologies are NPV‐
positive (i.e. they appear in the upper positive area of the plot).

The largest opportunities for energy efficiency in Singapore data centres are split evenly
between technologies for the facility (in blue) and for IT systems (in red). Integration (in
green) also presents a significant opportunity.

In reality, improvements to the facility are likely to be the first to yield results in terms of
actual energy savings. This is because facility improvements are often less disruptive to data
centre operations compared to similarly ambitious technologies for IT systems. This is
reflected by the faster deployment curves for facility technologies.

Nonetheless, the even spread of opportunities across the data centre technology stack
emphasises the need to pursue energy efficiency in a concerted manner at all layers of the
data centre technology stack.

19
Figure 3.1: Abatement Potential vs Net Present Value.

20
4. CONCLUSION

There is a necessity for data centre owners and users to account for their energy usage and
to introduce methods to reduce energy consumption. New technologies applicable to
facilities, IT infrastructure and software will enable further improvements in energy
performance beyond what is available from the best‐in‐class today.

It is recommended that Singapore focus on the following 3 areas to improve the long‐term
energy sustainability of data centre industry:

 Lowering the high cost of cooling in Singapore’s tropical climate via i) energy efficient
cooling methods, and ii) the hardening of IT equipment to withstand higher
temperature and humidity levels, thereby reducing the need for cooling.

 Improving the energy proportionality of data centre IT systems via i) software power
management, ii) the development of energy efficient hardware and architectures,
and iii) the automated optimisation of resource provisioning and workload allocation
in software‐defined data centre environments.

 Integration and optimisation across the traditionally silo‐ed IT systems and facility
systems of the data centre.

21
5. REFERENCES

[1] EMA, Singapore Energy Statistics 2013, 2013.

[2] The Green Grid, PUE: A Comprehensive Examination of the Metric, 2012.

[3] NEA‐HP, Data Centre Energy Efficiency Benchmarking, 2012.

[4] Digital Realty, Singapore Survey Results, 2013.

[5] Green Grid, Water Usage Effectiveness: A Green Grid Data Centre Sustainability Metric,
2011.

[6] Carbon Usage Effectiveness: A Green Grid Data Centre Sustainability Metric, 2010.

[7] Uptime Institute, 2012 Data Centre Industry Survey, 2012.

[8] NCCS‐NRF, Green Data Centre Primer, 2014.

[9] ASHRAE, Thermal Guidelines for Data Processing Environments, 2011.

[10] Google, The Case for Energy‐Proportional Computing, 2007.

[11] IDC, The Digital Universe in 2020, 2012.

[12] Fujitsu, Characterisation of Data Centre Energy Performance, 2012.

22
6. APPENDIX
APPENDIX A

Methodology

The merit of each technology is assessed using two metrics:

1. Abatement Potential, α2030, the cumulative mass of CO2 emissions avoided from the
present to 2030; and

2. Net Present Value, β2030, the economic benefit per unit mass of CO2 emissions
avoided from present to 2030.

For each of the identified technologies, assumptions are made about the:

 Efficiency Potential, µ , the efficiency gain as a ratio of the total energy consumption
of a data centre;

 Cost, C , the incremental cost as a percentage of the capital outlay of a data centre;

 Market Penetration, P20XX, measured as a percentage of the Singapore data centre

market by capital outlay in 20XX.

For each technology,

Where,

, is the projected electricity consumption by data centres in Singapore in 2030,

, is the projected price per unit energy in Singapore in 2030,

, is the projected capital outlay of data centres in Singapore in 2030, and

, is the projected Grid Emission Factor of Singapore in 2030.

Table A.1 details the assumptions for the assessed technologies.

Technologies are grouped according to the Energy Group on which they bring about energy
efficiencies, regardless of where the technology is actually deployed. The Facility Group of
technologies reduces the energy consumption of facility systems; the IT Group (encompassing both
physical IT and software) reduces the energy consumption of IT systems; and the Integration Group
reduces energy consumption across the entire data centre.

23
 Table A.1: Model input (in orange) of the assessed technologies.

µ C P2020 P2025 P2030

Facility Air and Cooling Management 0.1 0.50 80 100 100
Passive Cooling 0.1 1.00 0 30 50
Free Cooling (Hardening) 0.2 0.01 0 40 80
Close‐coupled Refrigerant Cooling 0.2 1.00 5 25 50
Direct Liquid Cooling 0.4 2.00 0 25 50
SMPS Efficiency 0.0 0.10 0 25 50
IT Software Power Management 0.4 1.00 0 10 50
Dynamic Workload Allocation 0.0 0.01 0 10 30
Dynamic Provisioning 0.4 0.50 0 0 40
Energy‐Aware Networking 0.0 0.01 0 10 30
All Wireless Data Centres 0.0 0.30 0 10 30
Memory Type Optimisation 0.0 0.50 10 50 75
Integration Advanced DCIM 0.0 0.50 20 50 75

24
Disclaimer, Limitation of Liability
This report is the result of a collaborative effort between the i3 Solutions Group Pte Ltd, the
Infocomm Development Authority and other Singapore Government Agencies. Users of this
report shall make their own independent business decisions at their own risk and, in
particular, without undue reliance on this report. Nothing in this report shall constitute
professional advice, and no representation or warranty, expressed or implied, is made in
respect of the completeness or accuracy of the contents of this report. i3 Solutions Group
accepts no liability whatsoever for any direct or indirect damages resulting from any use of
this report or its contents. A wide range of experts reviewed the draft reports; however, the
views expressed do not necessarily represent the views or policy of i3 Solutions Group or
other contributors to this report.

25
 Lead Agencies

Commissioning Agencies