![Consumer Vs Business Network Traffic (2011-2016); source [28].](https://www.researchgate.net/profile/Anders-Andrae/publication/255923829/figure/fig1/AS:392775032950785@1470656273226/Consumer-Vs-Business-Network-Traffic-2011-2016-source-28_Q320.jpg)
Content uploaded by Anders S.G. Andrae
Author content
All content in this area was uploaded by Anders S.G. Andrae
Content may be subject to copyright.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
6(
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(
41$(01'+2"#$(50)(
72.81$+(((
7$.&-,#(92"":$/(
Since 2009, several emerging technologies have initiated broad and disruptive impact
across the ICT sector: cloud computing promises efficiency of scale both in terms of capital
and operational costs; high-speed wireless networks promise near-ubiquitous network access
and thin-client solutions (smart-phones and tablets) provide appropriate, low power user-
interfaces to take advantage of this emerging next-generation ICT infrastructure. But despite
claims that this new consumer ICT infrastructure can reduce the overall energy costs of
society’s new digital lifestyle, there are few studies that encompass the total energy costs of
consumer ICT devices and the supporting communications networks and associated data
centers that have become so essential.
This work brings together the work of many prior researchers, while also introducing a
number of new methodological approaches to estimate growth in the portion of global
electricity consumption that can be ascribed to digital consumer devices. Baseline estimates
for the main categories of consumption - direct, manufacturing related, network-related and
data-center related - are determined for 2012. A number of methodological approaches are
outlined to extrapolate trends over the period 2013-2017 and projections based on best-case,
expected and worst-case scenarios are provided.
Key trends are identified. The most significant of these, which applies in all three scenarios
explored in this work, is that the proportion of direct electricity consumption by devices will
drop from c. 50% to 35% or less. Thus there is a strong trend to push electricity consumption
onto the network and data center infrastructure where energy costs are less transparent to
consumers. Some challenges are identified for networking and data-center sectors. Of these
the global roll-out of LTE will be a crucial determinant of future electricity demand.
There is a basis for significant optimism as new technologies in the TV panel and display
industries, combined with the replacement of desktop computers with laptops and thin clients
should lead to an overall drop in the direct energy usage of ICT devices. Our best-case
analysis shows a decline in consumption from 7.4% in 2012 to 6.9% of total global electricity
consumption in 2017; however the worst-case shows a rise to 12.0% driven primarily by
expansion of the network and data-center infrastructure.
Peter Corcoran
peter.corcoran@nuigalway.ie
Anders Andrae
anders.andrae@huawei.com
Senior Lecturer,
College of Engineering & Informatics,
National University of Ireland, Galway
Senior Expert
Emission Reduction/LCA/Sustainability
HUAWEI TECHNOLOGIES CO.,LTD.
Sweden 2012 Laboratories Division
HUAWEI TECHNOLOGIES SWEDEN AB
Skalholtsgatan 9, Kista, Sweden
;(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
):@,#(14(01'.#'.+(
!"!#$%&$'(#!&)*'%&'!+!,(#%,%(-',.&*/"0(%.&'1.#',.&*/"!#'%,(' 2!
3/(4.#*' 2!
3#(%,+!'*/""3#-' 2!
25'%&(#.)/,(%.&' 6!
"#$%#&$'!"(!)*&+!'"%,! -!
./&)+!"(!$/$%01! 2!
75'83,9$#./&)':',.&(!;(' <!
752'3'$+. 83+'0!#*0!,(%=!'.&'!+!,(#%,%(-',.&*/"0(%.&' <!
757'(4!'$#.>(4'.1',.&*/"!#'%,(' <!
34345!6$7&6)!8$#&6$!)%$/8+! 9!
75?'(4!'#.+!'.1'(4!'&!(>.#9' @!
756',+./)',."0/(%&$':')3(3',!&(!#*' @!
?5',3(!$.#%A3(%.&'.1'!&!#$-', .&*/" 0 (%.&B'"! ( 4.).+.$%,3+'
,.&*%)!#3(%.&*' C!
?52')%#!,('!&!#$-',.&*/"0(%.&'8-')!=%,!*' C!
:4545!8&+6.++&"/!;/ 8 !%$<;)$8!<&)$%;).%$! =!
:4543!)#!;/8!#&$' &/0 !8$#&6$+! =!
:454:!6<&$/)!&6)!8$#&6$+! >!
:454-!")*$%!6"/+.?$%!8$#&6$+! >!
?57'"3&/13,(/#%&$'!&!#$-':'+%1!',-,+!'3**!**"!&(' 2D!
:4345!8&+6.++&"/!;/ 8 !%$<;)$8!<&)$%;).%$! 5@!
:4343!)#!;/8!#&$' &/0 !8$#&6$+! 55!
:434:!6<&$/)!&6)!8$#&6$+! 55!
:434-!")*$%!6"/+.?$%!8$#&6$+! 55!
:4342!/$)'"%,+! 53!
:4349!8;);!6$/)$%+! 53!
?5?'!+!,(#%,%(-',.&*/"0(%.&'8-'&!(>.#9*' 27!
:4:45!8&+6.++&"/!;/ 8 !%$<;)$8!<&)$%;).%$! 5:!
:4:43!%;)&"/;<$!;/8!?$)*"8"<"0&$+! 5-!
?56'!+!,(#%,%(-',.&*/"0(%.&'8-')3(3',!&(!#*' 2<!
:4-45!8&+6.++&"/!;/ 8 !%$<;)$8!<&)$%;).%$! 59!
:4-45!%;)&"/;<$!;/8!?$)*"8"<"01! 59!
65'#!,!&(')3(3'*./#,!*'3&)'(#!&)*'E7DDC'F'7D2 7 G ' 2@!
652')%#!,('!&!#$-',.&*/"0(%.&'8-')!=%,!*' 2@!
-4545!A%$#&".+!%$+$;%6*$%+! 5B!
-4543!;/;<1+&+!"(!)*$!&/+);<<$8!8$#&6$!C;+$! 5=!
-454:!/$'!$((&6&$/6&$+!;/8!C$+)!A%;6)&6$! 5>!
-454-!0%"')*!&/!6"/+.?A)&"/!D3@@=73@53E! 3@!
657'"3&/13,(/#%&$'!&!#$-'3&)'+%1!H,-,+!',.&*%)!#3(%.&*' 72!
-4345!A%$#&".+!%$+$;%6*$%+! 35!
-4343!;/;<1+&+!"(!)*$!&/+);<<$8!8$#&6$!C;+$! 35!
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
A(
65?'!+!,(#%,%(-',.&*/"0(%.&'8-'&!(>.#9*' 77!
-4:45!A%$#&".+!%$+$;%6*$%+! 33!
-4:43!?$)*"8"<"0&6;<!6"/+&8$%;)&"/+! 3:!
-4:4:!0%"')*!&/!6"/+.?A)&"/!D3@@=73@53E! 3-!
656'!&!#$-',.&*/"0(%.&'8-')3(3',!&(!#*' 76!
-4-45!A%$#&".+!%$+$;%6*$%+! 32!
-4-43!&?A;6)!"(!$?$%0&/0!)$6*/"<"0&$+!F!+1/$%0&$+! 32!
-4-4:!0%"')*!&/!6"/+.?A)&"/!D3@@=73@53E! 32!
65I'83*!+%&!'!*(%"3(!*'1.#'7D27' 7<!
I5'0#.J!,(%.&*'1.#',!H%,('!& ! # $ -', . & */ "0(%.&'E7D2?'F'7D2@G' 7@!
I52')%11!#!&('$#.>(4'*,!&3#%.*' 7@!
24545!<"'!0%"')*!;++.? A)&"/+! 3=!
24543!$GA$6)$8!0%"')*!;++.? A)&"/+! 3=!
2454:!*&0*!0%"')*!;++.?A)&"/+! 3=!
I57'!+!,(#%,%(-',.&*/"0(%.&'8-',!H%,(')!=%,!*' 7K!
24345!$<$6)%&6&)1!6"/+.?A)&"/!C;+$8!"/!&/+);<<$8!8$#&6$!/.?C$%+! 3>!
24343!6"?A;%;)&#$!8;);+$)+! :3!
2434:!)*$!+&0/&(&6;/6 $!"(!8 $#&6$!< &( $) &? $ ! :-!
I5?'0#.J!,(%.&*'1.#'"3&/13,(/#%&$'!+!,(#%,%(-' ?<!
I56'&!(>.#9'!+!,(#%,%(-',.&*/"0(%.&' ?C!
24-45!C;+$8!"/!%;'!8;);!)%;((&6!0%"' )*! :>!
24-43!'&%$<$++!;66$++!/$)'"%,+! -@!
24-4:!+$A;%;)&/0!)*$!'&%$<$++!;66$++!6"?A"/$/)! -5!
I5I')3(3',!&(!#'!+!,(#%,%(-',.&*/"0(%.&' 6?!
<5'0#.J!,(%.&*'1.#'(.(3+',!H%,('!+!, ( # %,%( - ',. & * /"0(%.&' 66!
<52',."8%&!)'0#.J!,(%.&*'1.#'+.>L'!;0!,(!)'3&)'4%$4'$#.>(4'*,!&3#%.*' 66!
<57',."03#3(%=!'3&3+-*%*' 6<!
<5?')%*,/**%. & ' 6C!
94:45!A"+&)&#$!&?A;6)+!"(!$?$%0&/0!)$6*/"<"0&$+!F!+1/$%0&$+! -=!
94:43!H$?$%0&/0H!6*;<<$/0$+!;/8!)*$&%!A")$/)&;<!&?A;6)+! ->!
<56',.&,+/)%&$'#!"3#9*' ID!
#!1!#!&,!*' I2!
B(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
6>(5'.$1*2-.&1'(
Sales of ICT devices - laptops, home computers, their associated peripherals (networking,
storage and display) dominate market share and recent growth in the consumer electronics
industry [1]. ICT markets have been driven by the consumer sector for some time now. With
recent explosive growth in markets for new ICT devices - smartphones and tablets in
particular - and the transition of the television from a basic receiver into a digital media center
and entertainment hub these 'digital' consumer devices now dominate global sales of
consumer electronics. Throughout the remainder of this paper we refer to this broad category
of devices as as CE-ICT 'appliances'.
Since 2008, several emerging technologies have initiated broad and disruptive impact
across the ICT sector: (i) cloud computing promises efficiency of scale both in terms of
capital and operational costs [2–5]; (ii) high-speed wireless networks promise near-ubiquitous
network access [6–9] and (iii) thin-client solutions such as smart-phones and tablet devices
provide appropriate, low power user-interfaces to take advantage of this emerging next-
generation ICT infrastructure [10]. There are strong improvements in energy efficiency
promised both in terms of computational efficiencies [2], [11], [12] and data storage [13–15].
The main question we seek to answer in this paper is whether this new consumer ICT
infrastructure can actually reduce the overall energy costs of society’s new digital lifestyle, or
is it simply catalyzing a substantial rebound effect that could skyrocket ICT-related energy
costs over the next decade?
Throughout this article we employ the term energy usage as interchangeable with
electricity consumption - practically all of the energy that is utilized by today's manufacturing
industry, communications networks, data centers and the CE-ICT devices themselves is
derived from the electricity network. It is also useful to be able to estimate the percentage
total of global electricity consumption that is dedicated to support CE-ICT devices and the
associated infrastructure.
CD#$D&#E(14(.8&+(F1$G(
To reach this answer we begin with a review of the underlying global electricity usage that
can be ascribed to electronic consumer and ICT devices. Based on the work of previous
researchers [16], [17] we present a methodology where electricity consumption from ICT
devices is divided into 4 principle categories - (i) client devices, including PCs, laptops, TV
and home entertainment systems; (ii) network infrastructure; (iii) data center computation and
storage; and lastly, (iv) device manufacturing/replacement energy costs. We then proceed to
evaluate the current and projected energy consumption in each of these categories using,
where practical, more than one methodological approach.
Section 2 provides more background and context to this growth in energy usage by CE-ICT
devices. We discuss the rapid growth in the numbers of these devices, particularly thin clients
such as smartphones and tablets; the increasing reliance by such devices on network services
and the emergence of cloud-computing as a consumer resource are discussed.
In section 3 the approaches of past and current researchers are considered. The different
components of electricity usage driven by CE-ICT devices are considered and a unified
approach to evaluating the overall energy impact of these devices is presented. We then
consider the work of a range of past researchers from different fields and develop some
alternative methodologies to provide different approaches to estimate the different
components of CE-ICT electricity usage.
In section 4 data from the period 2008-2012 is matched with market predictions for new
client devices and network infrastructures and used to determine a baseline estimate for each
of the four main contributing components of electricity consumption - (i) direct consumption;
(ii) manufacturing consumption; (iii) network consumption; (iv) data center consumption.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
H(
Then in section 5 these baseline values are extrapolated to estimate consumption trends over
the period 2012-2017. Three different growth scenarios are employed - best-case, expected
and worst-case analysis with detailed assumptions provided for each. Section 6 reviews the
main trends that can be determined from this analysis.
To facilitate comparison with the work of previous researchers we include both industry
and residential ICT and some non-networked consumer electronics in our methodology. This
allows some understanding of how consumption patterns are changing as we transition to a
home environment where most appliances – even traditional white goods such as refrigerators
and washing machines - will be network connected. Our work is placed in context with a
range of contributions from earlier researchers and serves to emphasize the rapidly growing
significance of the CE-ICT sector as a global energy consumer.
Note that a key goal of this work is to outline a framework for evaluating future electricity
growth patterns. Thus, while we present a broad range of possible future outcomes, these are
supported with quite detailed sets of assumptions. It should thus be possible to review our
work and assumptions retrospectively and identify where our approach has aligned usefully
with future growth patterns and where it has not.
I'&.+(14(!'#$%/(
An important aspect of this article is to find a common metric that is suitable for
quantifying the energy usage of CE-ICT. Across different fields we find different units are
employed.
In the field of communications it is common to use the metric of joules/bit, or more
typically for modern communications networks the working unit is micro-joules/bit (µJ/bit).
In earlier work several researchers [16], [17] adopted the instantaneous units of gigawatts
(GW), while others [18], [19] have focused on the power consumption in kilowatts per
subscriber or household.
As our interest is in the total energy consumption, and as this is invariably provided in the
form of electrical energy we have adopted the measure of terra-watt hours per year (TWh/yr)
as used by Lambert et. al [20] in their study on communications networks. All data presented
in this work is converted to these common base units. Note that our interest is in the total
usage/consumption and thus per-device, or per-user metrics are scaled to reflect the installed
base of devices, or the estimated/measured user-base.
J(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
;>(K:-G%$ 12 '* (? (01'.#L.(
!"#$%$&'()*'$+ , -. / ,0 123 , $( 4 $5', 01 -20216 $7(4.89 /12(4$
Apart from a small annual decrease of 1.1% in 2009 [21], global electricity consumption
continues to rise on an annual basis [22]. In the residential sector governments have
responded with significant success by implementing policies designed to achieve long-term
market transformation in the supply and adoption of energy efficient appliances. As a
consequence the overall growth in electricity consumption in the residential sector has been
kept at acceptable growth levels over the past decade. However in 2010 there was a disturbing
6.5% increase in electricity consumption over 2009 indicating that new trends may be
displacing the significant success of past governmental policies to constrain growth in
electricity consumption [23].
In OECD countries those appliances that previously used the majority of electricity, such
as refrigerators, clothes washers and water heaters, are close to saturation levels and the net
electricity consumption per appliance has seen significant improvement across the OECD
[24]. However there appears to be a fundamental shift in residential electricity consumption
trends, particularly in the rising ownership of personal information and communications
technologies (ICT) and consumer electronics (CE) devices. In 2009 this group of appliances
accounted for approximately 15% of residential electricity consumption [24].
!"!$:;,$&-(<1;$(=$7(4.89,-$>7:$
While the market for many traditional CE appliances is stagnant in OECD economies, the
growth of consumer ICT is strong, showing a 50% growth rate over the period 2007-2012 [1].
When combined with mobile & personal entertainment CE devices (digital imaging,
communications & gaming appliances) these segments represent more than 70% of sales in
the consumer electronics sector in the US.
Growth in the CE sector has been further catalyzed by new network connected devices
such as smart-phones, tablet computers and more recently, smart-TVs. Interestingly these new
devices appear to have created new modes of usage that encourage multi-device ownership.
Throughout the remainder of this paper we will refer to these devices as ‘consumer
electronic-ICT’ (CE-ICT).
In parallel there has been an even more rapid growth in the provision of online services
designed to cater for the demand for content streaming, data storage and management driven
by these ‘new’ devices. In turn this has led to wide-scale deployment of large data centers and
the associated networking infrastructure by industry. This phenomenon is generically
presented as the practical realization of ‘cloud computing’ [25] and provides a strong synergy
for new CE-ICT devices [16].
;>;>6(0!M50)(N#D&-#()$#'*+(
A key element of the ‘new’ generation of ICT devices is that of network connectivity.
Typically these devices are either consumers or producers of digital audio/video and their
development owes much to improved coding and decoding technologies. Recent growth has
been further catalyzed through new network services – making access to content of
acceptable quality much easier and faster than was heretofore possible.
Smartphone sales were 800 million units in 2012 [26] and while growth must inevitably
slow the current trends suggest that close to 100% adoption will be the norm in developed
economies and with downwards pressures on unit costs and subsidies from service providers
these devices will likely be affordable to in most developing economies.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
O(
Multi-device ownership is also likely to be the norm. Tablet computers have established
their own place in the living room [27]. Within a family it is common for children to own or
have access to multiple connected devices prior to being considered old enough to own a
smart-phone – a handheld gaming terminal, a personal music/video player and dedicated
gaming consoles. In addition most families in the OECD will have multiple desktop or laptop
computers.
!"?$:;,$@(',$(=$1;,$A,1<(-B $
The role of the network as an ‘enabler’ for new CE-ICT devices has already been noted [1].
But the rapid growth in device numbers is in turn responsible for driving data traffic on
networks [28], [29]. And as can be seen from Fig 1 the bulk of network traffic is generated by
consumer activities.
But the network is also a significant source of energy consumption. In fact one recent study
[20] suggests that communications networks are responsible for 3.5% of global electricity
consumption and that 2.6% of that is directly attributable to consumer activity. It is also
arguable that improvements in network services have catalyzed growth in the CE-ICT sector,
in particular encouraging recent rapid growth in thin client devices such as smart-phones and
tablets. In turn this growth will drive further demand for faster and more ubiquitous network
service. Thus networking has come to represent a very significant component of electricity
consumption that is directly linked to CE-ICT devices.
Figure 1: Consumer Vs Business Network Traffic (2011-2016); source [28].
!"C$7'(8D$7(9/8124E$F$G*1*$7,41,-.$
Data centers have already achieved some notoriety in both academic and popular press as a
growing source of energy consumption [30]. More recently there has been some evidence of
significant improvements in efficiencies within data-centers [31]. Nevertheless it is clear that
the volume of data being handled by such centers is growing at unprecedented rates [29] and
is likely to be increasingly influenced by new growth patterns in consumer data traffic as
shown in Figure 1 above. Note in particular that consumer data growth is significantly
outpacing that of business/industry users.
This particular aspect of electricity consumption represented about 1.5% of global
electricity consumption in 2010 [31] [32]. However it is difficult to obtain direct information
on the trends that are influencing this sector since 2010. Historic growth rates vary from 12%
[17], [30], [33] to 8% [31]. We will comment further on this later and propose some
approaches to obtain sensible estimates for future growth.
P(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
A>( 0:.#%1$&Q: .&1'( 14( !'#$ %/ ( 01'+2 "3.&1'R(=#.81*1,1%&-:,(
01'+&*#$:.&1'+(
For this paper we have chosen a framework that is loosely based on that presented by the
authors of [17]. They have provided three device categories for direct energy consumption –
computers, TVs and other devices. In addition they define a distinct category for network
equipment and also for the energy consumption at the data center.
While we broadly follow these 5 categories but introduce some changes to facilitate our
subsequent analysis. Device consumption is merged into a single main category, and a new
category is introduced to cover the embodied lifecycle energy of each device type. This
effectively adds the manufacturing energy of devices into our analysis and is important
because the embodied manufacturing energy is significant when compared with the lifetime
operating energy for most electronics products. This is especially the case for short lifecycle
products such as mobile phones and many ICT devices where the manufacturing, or
embodied energy can be greater than the lifetime operating energy [34]. The communications
(network) costs associated with device use are also accounted for and account is also taken of
the growing phenomenon of cloud computing as this is a key enabling technology for newer
thin client appliances [16]. A more detailed discussion on each of these categories is provided
below.
?"#$G2-,01$5 4, -E 6$7 (4 . 89/12(4$)6$G,320,. $
In the context of our analysis the direct electricity consumption by devices should apply to
CE-ICT devices. However some earlier researchers have included other device categories in
their results so we will also consider non-ICT device categories with a view to making
comparisons. It is also important to understand how changing patterns of device usage within
the home may be affecting electricity consumption.
A>6>6(N&+-2++&1'(:'*(S#,:.#*(T&.#$:.2$#(
A very useful report was commissioned by the Consumer Electronics Association and
executed by the Fraunhofer Center for Sustainable Energy Systems [35]. This report is
particularly valuable for its methodological analysis of the usage patterns of a wide range of
CE devices. This study also provides an estimated average electricity usage for each device
category. We remark that these are provided in the context of the US, but they provide a
useful basis from which we can extrapolate to a more global consideration of electricity
consumption. Some more recent data is also available from the SMARTer 2020 report [36].
The values provided in this report align well with those in [35] and we shall discuss these
further in section 4.1.2 when we present an electricity consumption model based on the
installed base numbers for various devices.
A>6>;()U(:'*(U&#E&'%(N#D&-#+(
Our first category matches closely with Pickavet’s TV category [17]. As a baseline we
estimate the global number of TV sets in service to lie somewhere in the range of 1.8 and 2.0
billion units. Despite growth projections of 5%-8% back in 2010 [1], the annual sales of TV
display panels has stabilized around the 250 million mark [37][38] for the last 2-3 years
which is consistent with a unit life-cycle of approximately 8 years.
One significant change that we have seen is the switch-over to connected TVs. Again this
has not been as significant as the 100 million units predicted for 2013 [16] but the trend is
likely to grow until the majority of new TV panels feature network connectivity as a standard
feature. There is also an active market in add-on HDMI appliances that can provide this
connectivity for existing TV panels.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
V(
Another significant change that was not anticipated in 2011 [16] has been the substantial
improvement in energy performance of individual TV panels. The authors of [17] used a
figure of 330 Watts for a Plasma TV and 190 Watts for LCD TV. Today’s state of art LCD
panels are typically less than 100 Watts due to a transitioning to LED based displays – a
transition that is in the process of being effected across the industry in a remarkably short 4
year period [37]. Emerging technologies such as field emission display offer potential for
continued improvements for the future [39].
A>6>A(0,&#'.(50)(N#D&-#+(
This second primary device category follows closely on that of [17] but we have decided to
extend this to include devices such as smart-phones and tablet computers. Admittedly the
overall contribution of such devices to direct electricity consumption is relatively small, but
their rapidly growing numbers, and the fact that many households have multiple devices
suggests that they should be included in our estimates.
We also remark that the latest models of tablets are taking advantage of improvements in
battery technology to allow greater charges to be carried and to support brighter and higher
density display technology. While we do not expect this to be a long-term trend some high-
end tablet devices have essentially doubled their charge capacities in the latest models [40].
Eventually we might expect to see such improved battery technology deployed across all
smart-phone and tablet models. Thus, in contrast with other CE-ICT devices the energy
consumption per device of such thin clients is likely to grow over the next few years to
accommodate improved screen and CPU technology and emerging high-speed wireless
connections to the network.
Markets for desktop and laptop computers can be regarded as essentially stagnant or
declining in developed OECD economies, but there is some overall growth from developing
economies [16], [41]. We note that this is less than previously estimated [16], most likely due
in part to recent stagnation in the global economy and also because of the rapid adoption of
thin clients by the consumer sector. Again we can use estimates from [35] as a basis for
global calculations. This study shows a strong transition from desktop to laptop between 2005
and 2009. Now there are more laptops sold worldwide than desktops.
At the time of this study the installed base of desktop computers in the US was about 100
million and for laptops the figure was about 132 million. Worldwide the number of computers
sold is stagnant around the 350-400 million mark [16], [41], [42] and despite growth in
developing economies we expect that worldwide these will be stable, or even declining. The
initial evidence from 2013 is for a significant decline during the year [43].
From [35] we have an annual estimate of energy usage per desktop of 220 kWh/yr and for
laptops of 63 kWh/yr based on 2009 computer models. It is also worth noting that US
households are often multi-computer and that secondary and tertiary computers have lower
usage than the primary household computer. Bearing in mind that most households outside of
the OECD will be single-computer some correction factor is indicated. We also expect that
life-cycle will be significantly longer in the developing world where many older computers
continue in use as long as they remain serviceable.
A>6>B(C.8#$(01'+2"#$(N#D&-#+((
The authors of [17] introduced a third category that covers all other ICT equipment not
accounted for by the first two categories. They considered this category to include gaming
consoles, modems and wifi routers, audio/video (A/V) receivers, printers & multi-function
devices, digital and video cameras, MP3 players and similar digital media and stand-alone
video player devices.
In [17] the total power consumption of this category was estimated at 40 GW, or 350
TWh/yr which seems excessive. From our analysis, in order for this category to achieve such
a large usage of electricity most of these devices would need to be operated continuously.
6W(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
However the analysis of [35] shows that very few of these devices, many of which are TV or
computer peripherals, are operated continuously. Indeed many devices such as DVD players
and printers spend most of their time in low-power standby modes.
Rather than follow the authors of [17] it makes more sense to associate TV peripherals
directly with our TV category. Thus the present category comprises set-top boxes, game
consoles, DVD and Blue-ray players and A/V TV peripherals. This allows us to estimate
ratios of peripheral devices per TV-set and simplifies some of the work in extending from the
US-only study of [35] to a global analysis. Although not included in the present analysis,
cameras and music players are more correctly grouped as client devices - e.g. most mp3
players will connect to the network in order to obtain music from online stores and digital
cameras upload pictures and video onto a laptop or home network, frequently for sharing
online or printing via an online service.
Another category that is beginning to emerge is that of smart-appliances – examples
include network connected white goods, smart thermostats, home energy management,
lighting and security systems amongst others. While this category is not currently a
significant contributor to electricity consumption if we are to believe the proponents of the
Internet of Things (IoT) we can expect that it will emerge strongly over the next few years
[44]. For now, however, we do not attempt to include IoT appliances, cameras or music
players in our calculations. The current contributions of such devices to direct consumption
are relatively minor and their contribution to network traffic is typically via a computer.
However as more content generating and sharing devices become directly active on the
network their role is expected to expand significantly.
?"!$H*48=*018-24E$54,-E 6 $F $I 2=, $7 6 0 ', $%..,..9,41$
It is not just the energy consumption of electronics that is important. In fact for most
consumer electronic devices the electricity that is used in the manufacturing process is often
as large as the lifetime operational electricity of the device [45], [46]. In some cases it can be
multiples of the operational energy [34].
According to the International Energy Agency (IEA) there were more than 3.5 billion
mobile phones subscribers, 2 billion TVs and 1 billion personal computers in use around the
world in 2009 [24]. Moreover these various CE-ICT devices are distributed across developing
economies - in Africa 1 in 9 of the population owned a mobile phone. And, importantly,
many older low-efficiency devices are refurbished and resold within these economies even
where the operational and disposal costs do not make sense [47], [48].
Recent governmental policies have significantly reduced the energy consumption of many
traditional household appliances such are refrigerators and washing machines, but the rapid
growth in new CE-ICT devices was not predicted. Thus the residential component of
electricity consumption continues to grow worldwide, despite general improvements in white
goods appliance efficiency and policy measures [24].
The reality is that the energy requirements of semiconductor and nano-material
manufacturing processes can be 5-6 orders of magnitude greater than the traditional
manufacturing processes used to build, say, an automobile. To manufacture a kilogram of
state-of-art integrated circuits requires tens of thousands of megajoules [49], in contrast with
no more than 10 megajoules for conventional manufacturing [46]. The scope of
manufacturing is here Raw Material Acquisition, Production of Parts, Assembly of the
Devices and Distribution to Use. These life cycle phases are defined by the ETSI LCA
standard for ICT [50].
A>;>6(N&+-2++&1'(:'*(S#,:.#*(T&.#$:.2$#((
Because the manufacturing electricity of CE-ICT devices is typically much higher than for
other manufactured goods the total lifetime becomes very important in determining the
overall energy consumption of an appliance. Many CE-ICT devices only reach a balance
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
66(
point where the operational electricity is greater than the manufacturing electricity after
several years of operation. Yet many devices become rapidly obsolete - smartphones and
tablets are a good example - and may be replaced by their owners after as little as 1-2 years of
operation. Another cause for concern is that many of the latest CE-ICT devices are not
designed with repairability in mind [51]. This reduces their total lifecycle significantly.
Overall the potential lifecycle of an electronic device has a very significant influence both
on the total electricity consumption of that device, and also on the installed base. As an
example, consider TV panels where we assume an average lifecycle of 8 years and an
estimated installed base of 2 billion TV panels with annual manufacturing of 250 million new
devices. At the end of each year an approximately equal number of TV panels are introduced
into service as are withdrawn and the nett installed base remains around 2 billion units.
However if the lifecycle were to increase to 12 years then there would only be 166 million
TV panels leaving service each year yet 250 million entering service and the installed base of
TV panels will grow by 80 million units per annum. This is even more important for short-
lifecycle devices where the differences between a 2-year or 4-year lifecycle will have an even
greater influence on both growth of the installed base of devices and the total lifetime
electricity consumption of the device.
Most previous studies focus entirely on the use phase of the ICT Sector. However, several
authors [16], [17], [20] mention that the production of the ICT Equipment is and will become
even more a significant contribution to the overall electricity usage of the ICT Sector as LCA
studies have shown the beginning of life to be important for short-lived devices. In the present
study this presumption will be quantified for the first time transparently for the global
situation.
From published Life Cycle Assessment (LCA) studies the manufacturing electricity for
each piece of consumer ICT Equipment can be derived and thus the annual global
manufacturing electricity for each item of equipment. In this way, combined with annual sales
data, LCA studies can be used to estimate the upstream electricity for devices, networks and
data centers.
A>;>;()U(:'*(U&#E&'%(N#D&-#+(
Previous investigations [52], [34] arrive at a range between 400-559 kWh/TV.
A>;>A(0,&#'.(50)(N#D&-#+((
For desktops the sources are relatively rich compared to other ICT Equipment [34], [42],
[53]. These studies propose values from 60-215 kWh/desktop.
For monitors [54], [34] suggest a range from 187-334 kWh/piece.
For laptops [34] suggests a range of 75-167 kWh/piece.
For smartphones the best estimations come from Apple [55]; Further France Telecom has
an Eco-Rating method [56], [57] for smart phones based on metrics such as areas and masses.
From FTs method two metrics were used: 0.4 kWh/cm
2
screen and 6-35 kWh/IC chip
depending on flash memory storage. From these references we conclude that 30-60
For tablets again Apple [58] and France Telecoms Eco-Rating method [57], applied to
open metrics for typical tablets having 7 inch and 10 inch screens were used. For these studies
we conclude that 75-287 kWh/tablet is used in manufacturing. The variability is due to the
importance of the screen size.
A>;>B(C.8#$(01'+2"#$(N#D&-#+((
For this category only two LCAs were found in the literature, one for a set-top-box [59]
indicating manufacturing costs of 50 kWh/STB and one for a DVD-player [60]. The AV
6;(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
Receiver 50 kWh/unit and GC, 100 kWh/unit, were estimated with the help of a proxy from
[61] and [62].
A>;>H(X#.E1$G+((
Network infrastructure comprises many different types of equipment including a wide
range of radio base station technologies and more established wired and optical switches,
routers and . The method of annual shipped units and electricity used per unit was not applied
due to data gaps. Instead the annual use stage electricity consumption of Networks derived in
the present study was used as basis. Then combined with previous LCA studies [34], [52],
[61], [63], estimating that the manufacturing share of the total lifetime electricity of fixed and
wireless networks lies in the range of 10-20%, the lifecycle ratio method proposed by
Greenhouse Gas Protocol ICT Supplement [64] was applied.
Note that a sensitivity check is performed for radio base stations as these make up a large
share of wireless network equipment manufacturing.
Total sales of radio base stations for mobile communications were 1.3 million in 2012. Of
these 95% were macrocells. The expected sales for 2017 are around 1.6 million (70 % macro
and 30% smaller cell sizes) [65]. It takes around 3 MWhr per radio base station [66] so 5
TWh per year. Radio base stations then constitute a reasonable share of the Network
manufacturing, approximately 10% of 51 TWh in 2012 - see Table 3(b). On the other hand
there are alternative interpretations which estimate that 85% of the world’s population will be
covered by 3G mobile internet in 2017 and that 4G coverage will reach 50% in the same
timeframe [67].
This latter scenario implies a larger global market for radio base stations. In general market
reports/estimations on shipped unit fluctuate considerably and so it is challenging to predict
and even back-cast a very exact number.
A>;>J(N:.:(-#'.#$+(
The data centers consist of many different types of ICT Equipment and peripherals such as
servers and storage systems.
Instead the annual use stage electricity consumption of Data centers derived in the present
study was used as basis. Then combined with previous LCA studies (Honee 2012 [68],
Andrae 2013 [61]), estimating that the manufacturing share of the total lifetime electricity of
fixed and wireless networks lies in the range of 10-20%, the lifecycle ratio method proposed
by Greenhouse Gas Protocol ICT Supplement [64] was applied. A sensitivity check is
however done for annual shipments of servers as these contribute a significant share of data
center equipment manufacturing electricity [61].
Around 10 million servers were sold in 2012 [69]. It takes around 0,4 MWhr per server
[70] so 4 TWh for 2012. Servers thus constitute a reasonable share of the Data Center
manufacturing, about 10%, or 4 of 40 TWh in 2012 (See Table 3(b)).
?"?$5',01-20216$7(4.89/ 12( 4 $) 6 $A , 1<(-B.$
The networking component of electricity consumption is very significant in out
calculations. At this point in time we are seeing a major transformation in ICT with a shift
from desktop and laptop computing to a thin client model. The very rapid adoption of smart-
phones, tablets and related thin client devices coupled with the equally rapid growth in cloud-
based consumer services signals a disruptive shift in ICT usage.
In turn this moves much of the computational effort from the local client and displaces it to
a back-end data center. There is evidence that, with proper management, this can improve the
efficiency of many computational tasks In turn this moves much of the computational effort
from the local client and displaces it to a back-end data center. There is evidence that, with
proper management, this can improve the efficiency of many computational tasks [4], [13],
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
6A(
[15], [71], [72] and thus reduce the electrical energy required per CPU cycle. However the
data that is to be processed and the resulting outcome must be transmitted from client to data-
center and back again over a network infrastructure which presents challenges [15], [71],
[73].
A further complication with network connections is that they are 'always on' and thus [74]
consuming energy even if not active.
even if not [4], [12], [14], [82], [83] and thus reduce the electrical energy required per CPU
cycle. However the data that is to be processed and the resulting outcome must be transmitted
from client to data-center and back again over a network infrastructure which presents
challenges [14], [82], [84]. A further complication with network connections is that they are
'always on' and thus [85] consuming energy even if not used. The situation is even more
complicated for mobile networks, where networking connections can be established as needed
and then disabled, but there is a high cost to make and break connections, particularly for
LTE [75], and while client devices can power down when not accessing the network the
background network must be continually active.
In fact the optimal network connection would be under 100% constant loading in order to
achieve maximum efficiency per unit of data transferred. But this is not how practical
networks, either wired or wireless, work in the real world. Thus much of the peripheral
infrastructure of a network will be lightly loaded and in consequence operate with very low
levels of energy efficiency.
A>A>6(N&+-2++&1'(:'*(S#,:.#*(T&.#$:.2$#(
Networks can be broadly divided into a core backbone and a local access network [18],
[76]. The core networks is relatively efficient, but depending on the type of local access
network employed the energy costs can be at least an order of magnitude higher [77]. This is
especially the case when high-speed wireless networks are used to provide the local access.
As an example, the authors of [20] determined a model that predicts that electricity
consumption by the total combined network components of ICT would exceed a baseline
global figure of 350 TWh/yr in 2012. But if the applied electricity consumption figures per
user from [78], drawn from access networks that primarily employs LTE technology, are
incorporated into the same model then that figure would rise to 850 TWh/yr - a multiplier
factor of 2.4 for electricity consumption.
Another useful study is by Baliga et al [76] analyzing the relative energy consumption in
different wired and wireless networks. Their conclusions reinforce our comment that at data
rates above 10 Mb/s wireless networks become power hungry consuming 10 times the
electricity of the equivalent wired network. This leads us to another key point that is
emphasized in a whitepaper [79] from the Center for Energy Efficient Communications
(CEET) and relates to the significant growth in wireless access networks. The majority of
today's new thin-client devices use wireless access, either via WiFi or 3G. In fact growth in
mobile data has been somewhat restricted due to WiFi offloading where users restrict data
connectivity on the 3G interface due to significantly higher costs [28]. But the real unknown
in terms of wireless networking is the pending global rollout of 4G/LTE data services. This
will be discussed later, but the CEET whitepaper gives a useful perspective, illustrating that
the growing energy requirements for wireless access networks is significantly higher and
growing more rapidly than the energy requirements for data networks within data centers.
While this CEET whitepaper is not the first research to study wireless access networks
[76], [80], [81] it appears to be the first to explicitly highlight the significant growth in energy
usage of such networks as wireless becomes the global norm for subscriber access. In an
earlier publication Corcoran [16] also incorporated a substantial impact on network energy
due to new rollouts of LTE networks. However many deployments were delayed and it now
6B(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
appears that 2013/14 will see most global network operators begin large scale rollouts of
G4/LTE networks [82], [83].
A key point to make here is that it is the data rate at the access point that will be the
primary determinant of network electricity consumption. Thus if users are expecting 2-3 Mb/s
data rates a single 10 Mb/s could support a maximum of 4-5 concurrent users. As LTE/4G
networks will mainly be deployed in crowded urban environments we expect that operators
will seek to optimize the number of users per base-station thus running base-stations at higher
data rates, in theory up to 100 Mb/s. This, in turn, could drive the electricity consumption of
these access networks up by two, even three orders of magnitude higher than the equivalent
wired backbone. In fairness many researchers are aware of this issue and a variety of
proposals to reduce the power requirements for mobile networks have been proposed and will
be discussed in due course.
Typically to operate at a data rate of 10 Mb/s or higher over LTE/G4 mobile networks
requires at least on order of magnitude higher electricity consumption compared with wired
access network technologies. To date the roll-out of such high-speed wireless networks has
been delayed by the global recession, but various OECD economies are in the process of
rolling out these networks over the next few years and China has indicated that it intends to
complete the world’s largest high-speed LTE/G4 network by the end of 2013 [83]. This will
feature 200,000 base-stations across 100 cities and will cater for 500 million people. It will be
very interesting to see the impact of this infrastructural development on local electricity
supply.
Note that while some improvements in the efficiency of such high-speed networks might
be expected, practical physical limitations arise from Maxwell and the 1/R
2
loss of power
with distance from source, R. This has led Telco operators to investigate more granular
networking architectures using femto-cells, effectively opting for more access points with
lower data rates [84–90]. In practice we do not know what level of service users will expect
and this will strongly influence how LTE/G4 deployments evolve. If users want faster data
rates and more ubiquitous access then we could see soaring electricity consumption by mobile
operators.
If we next consider the roll-out of data services in the developing economies it becomes
clear that LTE/G4 offers a very attractive infrastructure that can be rapidly deployed in large
urban environments without modern telecommunications backbone or wired switching
infrastructure. These wireless networks do not require substantial civil works such as the
laying of additional optic fiber backbone, or installing MAN networks or the equipping of
local digital exchanges. More importantly they can be deployed rapidly – note our earlier
comments about China’s ambitions in this regard. Given the rapid growth in such urban
centers, particularly in Africa and Asia, LTE/G4 offers a very attractive approach to bring
broadband infrastructure to developing economies.
A>A>;(S:.&1':,#(:'*(=#.81*1,1%&#+(
The energy consumption of networks is very well studied in the literature [91], [78], [80],
[92]. There is also a lot of focus on certain categories of network. In particular optical
networks [93–98] are important as these form the backbone or 'core' network that handles all
bulk data traffic. Another field where substantial research is taking place is that of energy
usage and efficiency in wireless or mobile networks [77], [91], [99–101].
For our purposes we will take a fairly top-level view of communications networks.
Nevertheless there are different components to any network infrastructure and it is important
to make some key differentiations. One of the reasons for doing this is that we will later use
some of the available growth projections for network traffic to help understand the likely
future growth in network electricity consumption. In this regard we do need to make some
distinction between different categories of network technology as will be discussed in the
next section.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
6H(
H()2',$*4D$J2K,D$G*1*$
In a very broad sweep we could assert that practically all networking is converging towards
the Internet. Or perhaps a better way to express this is that the Internet, as it continues to
evolve, is defining a global scalable networking infrastructure. Yes, there are many 'secure' or
'partitioned' networks, but how many of these do not support at least some Internet protocols?
And rely on commoditized 'Internet' routing and interface hardware?
Now the biggest change to the Internet model in the last decade has been the introduction
of mobile data. To date this is still a small proportion of the overall network traffic, but since
the introduction of smartphones and, more recently, tablet devices and the rapid market
adoption of both technologies the growth of mobile data has accelerated at a very fast pace.
Indeed it has been dampened significantly by high tariffs applied by most service providers,
and the growing availability of Wi-Fi hotspots, but has still grown at 60%+ year on year for
the last 3-4 years. Now that high-speed mobiles networks have started to be more widely
deployed by ISPs mobile data looks set for a significant stimulus in 2013.
This, in turn, leads to our consideration of telecoms networks in two primary components.
We see the 'core' network and its wired extensions, including the wired access network, as
one primary entity. This network infrastructure will continue to expand as it has in the past
and its energy efficiency will be essentially static. As we shall discuss shortly, this means that
we can estimate the growth in energy consumption over the next few years from the growth in
network data traffic.
The second component is that of the wireless access network - essentially today's mobile
phone networks. As explained above this is the most interesting component of the network as
we are on the cusp of a widespread deployment of LTE networks. These wireless networks
are capable of much higher data rates than today's mobile networks, but in parallel they have
potential to increase energy consumption substantially.
Thus our methodology seeks to separate the more conventional optical and wired network
components, which may be modeled based on the growth of core networks traffic, from the
wireless, or mobile component which can be modeled based on the predicted growth in
mobile traffic. The details of this approach are given in section 4.2.3.
A,1<(-B$:-*==20$&-(<1;$+-(L,012(4.$
Growth of core network traffic.
Figure 2(a): This illustrates the growth in core network traffic, including traffic within
data centers; about 70% of total traffic is within or between data centers.
6J(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
$
Figure 2(b): Monthly mobile traffic; this excludes traffic from mobile devices that is
offloaded to Wifi connections; in 2012 most mobile traffic is carried by 2G/3G networks but
by 2017 as much as 45% of mobile data will be carried by new LTE/G4 mobile networks.
?"C$5',01-20216$7(4.89/ 12( 4 $) 6 $G * 1* $7 , 41 , -.$
Data centers have existed as recognized entities for at least the past two decades. The
concept of a centralized computing resource can be traced back to the 1960s when the
original concept of 'cloud computing' was presented by Douglas Parkhill [25]. In recent years
we have seen data centers growing from enterprise computing facilities to provide the
backbone for Internet growth and, more recently, to emerge as an essential back-end
infrastructure [102] for a new generation of thin-client consumer electronics devices [16].
Naturally the size and scale of these data centers continues to grow and today they are seen
as a key element in the next stage of growth for the ICT industry [72], [103].
A>B>6(N&+-2++&1'(:'*(S#,:.#*(T&.#$:.2$#(
Koomey [30], [31], [33] is a key contributor to studies on electricity usage by data centers.
Other authors have focused on the sheer size of today's data center [102], or on the energy
efficiency of the underlying architecture [104], [105] or specific operational aspects [106],
[107], [108] or have considered energy related issues of cloud computing [109], [3], [5], [73]
but it is Koomey who has quantified the total energy costs.
In [31] Koomey has reviewed and arrived at a revised and lower estimate of power
consumption by worldwide datacenters for 2010 lying in a range between 203 and 272
TWh/year. This compares with a best guess/worst case of 301/397 TWh/year from his earlier
work. Much of this improvement derives from the migration of data from in-house data
centers where most servers operate at low loading, to cloud computing centers where server
loads are balanced using virtual machines (VM) techniques [2]. A second contributory factor
is improvements in server power consumption through use of lower power chipsets and
architectures [108]. We will use Koomey's estimates as a basis for our calculations in section
4 below.
A>B>6(S:.&1':,#(:'*(=#.81*1,1%/(
Data centers and networks are highly interlinked and many authors are happy to consider a
combined metric for both infrastructures. As commented above we prefer to try and separate
the data center component of electricity consumption from the network component. In
practice this is not as difficult as it may seem. The internal network infrastructure of data
centers will be wired, and mainly optical thus having relative low power consumption. This is
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
6O(
confirmed by Table 2 of Koomey's [31] most recent analysis to be approximately 6% of data
center electricity consumption. Note that this also explains the low figures for Data Centers
used in [79] - these authors are comparing the communications network infrastructures rather
than the total energy use of data centers.
Thus it is relatively straightforward to separate the network dependent portion of energy
within data centers. This, in turn, facilitates the use of network focused studies such as that of
Lambert et al. [20] in combination with the work of Koomey to provide separate evaluations
for the electricity usage by data centers and by the global network infrastructure.
B>(S#-#'.(N:.:(912$-#+(:'* () $ #' * + (Y; W WP (Z(; W 6 ;[(
In one sense much of this data is historic and so it should be possible to obtain fairly
accurate estimates. However, it is not practical to know the exact energy consumption of even
a single device once it is installed, as we do not have direct access to measure its on-site
usage. Thus any determination of the overall usage of electricity by the CE-ICT ecosystem
can only be an estimate. Nevertheless we believe a reasonable convergence can be shown
across the work of a number of different researchers.
In this section we review a range of such work across a number of different areas that are
relevant to CE-ICT and in parallel develop our own methodologies based on alternative data
sources. The main goal is to establish some reasonable convergence in estimates for today's
rate of electricity usage by the CE-ICT device ecosystem.
C"#$G2-,01$5 4, -E 6$7 (4 . 89/12(4$)6$G,320,. $
A first challenge is to consider the direct usage of electric power by CE-ICT devices. As
was explained in section 3.1 it makes sense to separate devices into two main categories - (i)
TV & TV-centric devices, and (ii) ICT devices.
The latter category are also referred to throughout this report as 'client' devices because
their primary usage tends to be accessing networked content and services. Another way to
describe this categorization is as 'broadcast' and 'network' devices. From a practical
perspective almost every modern CE-ICT device falls into one of these categories.
B>6>6(<$#D&12+(S#+#:$-8#$+(
We begin by comparing estimates from the literature for the client device category. The
two principle research papers that identify and attempt to quantify this particular group of
devices are [17] and . These are compared with the total business as usual estimate from [24]
for all CE-ICT devices.
Table 1(a): Electricity consumption for client ICT devices 2008-2012
Citation
Devices
2008
2009
2010
2011
2012
Pickavet, et al. [17]
PC, Ltop, Display
262.8
282.5
303.7
326.5
351.0
Corcoran [16]
PC, Ltop, Display,
smart-phone & tablet
262.8
286.5
312.2
340.3
371.0
IEA [24]
ICT & CE
670
720
776
830
885
Table 1(b): Electricity consumption for TV devices and peripherals 2008-2012.
Citation
Devices
2008
2009
2010
2011
2012
Pickavet, et al. [17]
TV, Plasma, LCD
385
420
458
499
544
Corcoran [16]
TV, Plasma, LCD,
Smart-TV
350
350
350
438
548
IEA [24]
ICT & CE
670
720
776
830
885
6P(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
There are various inconsistencies between the approaches used by these, and other
researchers, but the main point of tables 1(a) & 1(b) is that there appears to be a good
consensus as to the combined electricity usage of devices in 2011/2012. Note that the IEA
figures are a combined total for all CE-ICT devices.
In 2011, all three research reports agree that combined CE-ICT electricity consumption lies
between 778 and 830 TWh/year; in 2012 this quantity lies between 885 and 919 TWh/year.
B>6>;(7':,/+&+(14(.8#(5'+.:,,#*(N#D&-#(K:+#(((
As indicated in the introduction to this section it is possible to find detailed information on
the installed base of devices in a range of different categories from various government &
industry sources. It is also possible to apply a detailed usage analysis model such as those
developed by [35] to each category of devices. In table 2(a) a summary of our results is
presented, extrapolating the usage models of this report to apply more broadly to the installed
base of global CE-ICT devices.
There are several adjustments/refinements that are worth noting. Firstly the annual kWhr
rating for a typical US TV panel was estimated at 180 kWhr in 2010. Here a 10% higher
figure of 200 kWhr has been adopted to reflect two factors: (i) in emerging economies we
expect a higher proportion of older high-power TV sets in the installed base, and (ii) there
will be a larger proportion of single-TV households in developing economies and these
primary TV sets have higher annual usage than 2nd and 3rd TVs in multi-TV households.
Table 2(a): Electricity consumption (TWh/yr) for CE-ICT devices and peripherals 2011-2013.
Installed Devices (x10
6
)
2011
2012
2013
kWh/yr (2010)
Desktops
579
588
598
220
Monitors
608
617
628
97
Laptops
729
832
946
80
Smartphones
700
1000
1350
5
Tablets
50
150
250
15
TV
1900
2000
2100
200
TV STB
722
760
798
100
TV GC
380
400
420
135
A/V Receiver
570
600
630
65
DVD/Blueray
665
700
735
28
TWh/yr
Desktops
127
129
132
Monitors
59
60
61
Laptops
58
67
76
Smartphone
4
5
7
Tablets
1
2
4
TV
380
400
420
TV STB
72
76
80
TV GC
51
54
57
A/V Receiver
37
39
41
DVD/Blueray
19
20
21
Total (TWh/yr)
808
852
897
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
6V(
For laptops the [35] study used an estimate of 63 kWhr/yr per laptop, based on a usage
model where 43% of portable computers are second computers for their user and where
energy star power savings strategies are mandatory. In developing economies we expect a
higher proportion of older or second-hand laptops without advanced energy saving features
and almost all portable computers will be primary devices for their users thus incurring a
higher level of usage. For these reasons a higher base usage figure of 80 kWhr was adopted to
reflect such differences. (Note that the SMARTer 2020 report [36], which is more recent,
does not have an explicit figure for laptops which we assume are included as part of the
energy consumption by desktop computers.)
Based on these estimates we see that annual consumption for all devices is estimated at 808
TWh/yr in 2011, which fits nicely between the estimates of 778 to 830, TWh/yr determined
from the work of prior researchers. For 2012 these suggest a figure of 852 TWh/yr with is
somewhat lower than prior estimates but this can be explained in part by changes in the both
PC and TV markets. PC sales have stagnated in recent years and have been strongly displaced
by laptop sales and more recently tablet computers. Thus there is a shift to lower power client
devices in the ICT category although there is still a significant installed base of older PCs and
laptops in 2012.
On the TV side we note that the market has stagnated at c. 250M units per annum [37],
[38] in contrast to a 7.5% or higher growth rate predicted by past researchers. Again there is
evidence of changing use patterns that lead to more viewing of content by consumers on low-
power ICT client devices. Thus the figures presented in table 2(a), although slightly lower
than previous estimates, appear to offer a sensible baseline for direct electricity consumption
by CE-ICT device.
A(1,$(4$24=(-9*12(4$.(8-0,.$=(-$1;,$9*24$D,320,$0*1,E(-2,.$
For clarity we provide an outline of our methodology for the main devices.
Desktop and Laptop Computer: Market estimates for these are taken from IDC market
analysis [41] with an assumed lifecycle of 4 years. For later projections the lifecycle is varied
from 3-5 years depending on the growth scenario.
TV Panels and Peripherals: Market estimates are taken from [37] with a lifecycle
assumption of 8 years. Most TV peripherals are assumed to have a lifecycle that tracks the
TV itself.
B>6>A(X#E(!44&-&#'-&#+(:'*(K#+.(<$:-.&-#(
In the analysis presented above we applied the power consumption in annual KWhr per
device from the [35] study. These values were based on a study of CE and ICT devices in the
US in the period 2009-2010, but some categories of device will have improved on their gross
power consumption even in the last few years.
Notably the TV industry has been very successful in migrating from older TV panel
technologies such as plasma and LCD to more power efficient LED panels. Thus in 2011
25% of TV panels were LED based, rising to 55% in 2012 and >75% projected for 2013 [38],
[110].
Thus while past researchers predicted that larger display sizes would increase electricity
power consumption, in fact new TV panels will actually tend to reduce overall power
consumption over the next few years. Based on our estimation of 150M 'high inefficiency' TV
panels replaced annually from the 250M sales of new TVs we expect the power consumption
of those 150M replaced devices to be halved. This assumes the replacement of a high-power
panel (200-300 Watts) by a state-of-art LED panel (<100 Watts).
Taking as an example our estimates above for 2011-2013 and our replacement rate of
150M per annum. Thus over this three year period we have added 300M additional 150 Watt
TV sets, but we have also displaced 225M @ 200+ Watt TV sets with 100 Watt TV sets -
;W(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
thus about 50% of the increase in electricity consumption is offset by replacement of high-
power TV panels or older CRT based TVs. As the industry approaches 100% LED
technology, expected in 2014/15 then we could expect to see a net reversal in total electricity
consumption by TVs for a number of years.
However the actual replacement rate of TV panels is quite significant and if it is
sufficiently high then even with positive market growth we could see substantial reduction in
the direct electricity consumption of TV panels as older models are scrapped. On the other
hand if consumers hold onto their TV panels beyond their 10-year lifecycle - as 2nd or 3rd
household TVs - we can expect that electricity consumption could continue to grow even in a
declining market for new devices. In the current state of the global economy it is challenging
to understand exactly how consumer behavior will evolve over the next few years.
B>6>B(\$1E.8(&'(01'+2"3.&1'(Y;WWPM;W6;[(
It is helpful to compare our three main sources of recent data to see how each predicts the
increased consumption of CE-ICT devices. The IEA estimates [24] begin from 2010 and are
closely in line with Pickavets estimates [17]. Corcoran is lower which is most likely because
that work restricted itself to the major categories of ICT Clients (PC, laptop and other
consumer ICT devices) and to TV displays and associated peripherals [16] (i.e. Pickavets
includes an additional 'other' category that is only partly covered by Corcoran's estimates).
Nevertheless we see a close convergence in 2011 and 2012 from all three of these
researchers and this supports our model described in section 4.1.2 above.
Table 2(b): Electricity consumption (TWh/yr) for all primary CE-ICT devices and peripherals
(2008-2012) from previous researchers, compared with the individual device model of this work.
2008
2009
2010
2011
2012
IEA [24]
776
824
873
Pickavet, et al. [17]
648
703
762
826
895
Corcoran [16]
613
625
645
824
870
1
This Work
808
852
The main goal of this section is to obtain a consensus estimate for the electricity
consumption of devices in 2012. Given the improvements in energy efficiencies for TV sets
and, to a lesser extent, for PC and laptop clients it seems reasonable to use the 852 TWh/year
estimate from our model as a basis for predicting the future trends in electricity by CE-ICT
devices in section 5.
1 The original estimate provided in [16] had assumed strong growth in TV sales driven by a broad adoption of
3D technologies and the introduction of Smart-TV; the figure provided here has been modified to reflect the actual
market condition of 2012.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
;6(
C"!$H*48=*018-24E$54,-E6$*4D$I2=,M760',$7(4.2D,-*12(4.$
(
B>;>6(<$#D&12+(S#+#:$- 8# $+(
To the authors knowledge there are no overall global ICT electricity footprint estimations
that transparently include the manufacturing electricity. In this sense this research will shed
light on the importance of the relative importance of the manufacturing of ICT as compared
to the use stage. Individual more or less transparent LCA studies do exist for individual parts
of the present scope. Here facts from these LCA studies are put together.
(
B>;>;(7':,/+&+(14(.8#(5'+.:,,#*(N#D&-#(K:+#(((
Several sources present annual numbers of shipped devices that can be used for 2008-2012
[26], [37], [38], [69], [111] [34], [41], [112]. The present set top box and game console shipments
can be regarded as low estimates [113], [114]. Monitor sales follow the desktops shipments [114]. In
Table 3(a) the basis for 354 and 281 TWh for Use stage Networks and Data Centers are presented
below in Sections 4.3 and 4.4, respectively.
Table 3(a): Numbers of Shipped Devices and per-unit Manufacturing Electricity
2008
2009
2010
2011
2012
KWhr/unit
(2010)
Desktops
141
143
146
148
151
188
Monitors
157
159
162
165
169
268
Laptops
143
170
201
215
246
134
Smartphones
120
160
350
460
700
40
Tablets
1
1
50
100
150
75
TV
241
244
252
255
258
400
TV STB
92
93
96
97
98
45
TV GC
48
49
50
51
52
150
A/V Receiver
72
73
76
77
77
100
DVD/Blueray
84
85
88
89
90
200
Networks eq
(TWh)
354
Manuf: 15%
Use 85%
Data center eq
(TWh)
281
Manuf: 15%
Use 85%
To get the TWh for Desktops in 2008 141 million units was multiplied with 188 kWh/unit.
Then an annual improvement in electricity efficiency is assumed so the TWh for Desktops in
2009 in 143 million×188kWh× (1-0,05)=26 TWh, in 2010 146×188×(1-0,05)
2
, and so on.
For Networks and Data Centers the manufacturing electricity was estimated with a formula
from the Greenhouse Gas Protocol which uses a so called LCA stage ratio factor from which
Network equipment manufacturing electricity (E
man
) can be estimated. The expected share of
the use stage electricity (E
use
) for a Network is around 85%, i.e. C
use
= 85%.
E
man
= {E
use
/ (C
use
/100)} x (1 – C
use
/100)
E
man
= {354 kWhr / 85%/100)} x (1 – 85%/100) = 62 TWh. We assume a 5% annual
improvement of the manufacturing electricity since 2008, 62×(1-0,05)
4
=51 TWh
;;(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
Table 3(b): Manufacturing Electricity (TWh/yr) by Device Category
2008
2009
2010
2011
2012
Desktops
27
26
25
24
23
Monitors
42
40
39
38
37
Laptops
19
22
24
25
27
Smartphones
5
6
13
16
23
Tablets
0,04
0,07
4
8
11
TV
96
93
91
87
84
TV STB
4
4
4
4
4
TV GC
7
7
7
7
6
A/V Receiver
7
7
7
7
6
DVD/Blueray
17
16
16
15
15
Networks equipment
51
Data center equipment
40
Total
277
309
322
327
327
C"?$5',01-20216$7(4.89/ 12( 4 $) 6 $A , 1<(-B.$
B>A>6(<$#D&12+(S#+#:$- 8# $+(
It is not a surprise to find that the electricity usage of networks has been a topic considered
by many earlier researchers. Both Pickavets [17] and Corcoran [16] identified networks as a
key component of the overall electricity usage by CE-ICT devices. A number of other
researchers have studied, in varying degrees of details, the growth in electricity requirements
of telecommunications networks and the networking phenomenon known as the 'Internet'.
Now it is not our goal to perform a very granular study of networks, but rather to try and
obtain a sensible 'top-level' understanding of how they are evolving and obtain a consensus
estimate of the current level of electricity consumption in 2012. Thus we focus on a number
of the most relevant research studies supporting this objective.
Lambert [20] provides a detailed study of the use phase electricity consumption in
communications networks, specifically telecom operator networks and including customer
premises equipment and an estimate for office and building networks. They also provide a
very helpful comparison of their results with a number of other recent research studies. The
overall estimate for 2012 is included below in section 4.3.3 and will be used as a baseline for
later projections of growth in electricity consumption due to network growth.
One interesting observation made by these researchers is that when they take per-user
electricity consumption figures from another study by Kilper [78] and applied these in their
model they found that the total electricity figures were more than doubled to 812 TWh/year.
This study by Kilper was based on an early LTE network and this importance of the access
network technology in determining the overall electricity consumption of the network
infrastructure is underlined by this thought experiment. We already discussed the operational
power consumption of LTE and other high-speed networking technologies in some detail in
section 3.3.1 and there will be further discussion on the potential impact of LTE when we
discuss our future projections.
One key point that is emphasized in a whitepaper [79] from the Center for Energy Efficient
Communications (CEET) relates to the significant growth in wireless access networks. The
majority of today's new thin-client devices use wireless access, either via WiFi or 3G. In fact
growth in mobile data has been somewhat restricted due to WiFi offloading where users
restrict data connectivity on the 3G interface due to significantly higher costs [28]. But the
real unknown in terms of wireless networking is the pending global rollout of 4G/LTE data
services. This will be discussed later, but the CEET whitepaper gives a useful perspective,
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
;A(
illustrating that the growing electricity requirements for wireless access networks is
significantly higher and growing more rapidly than the electricity requirements for data
networks within data centers.
While this CEET whitepaper is not the first research to study wireless access networks
[76], [80], [81] it appears to be the first to explicitly highlight the significant growth in
electricity usage of such networks as wireless becomes the global norm for subscriber access.
In an earlier publication Corcoran [16] also incorporated a substantial impact on network
electricity due to new rollouts of LTE networks. However many deployments were delayed
and it now appears that 2013/14 will see most global network operators begin large scale
rollouts of G4/LTE networks [82], [83].
B>A>;(=#.81*1,1%&-:,(01'+&*#$:.&1'+(
+*-1212(424E$1;,$A,1<(-B$
Based on our considerations of the literature, and the shifting perspectives of various
researchers in the last 2-3 years it is clear that networks can be broadly divided into 'core' and
'access' components. The 'core' component is the main wired backbone, based on optic-fiber
technology with some conventional copper wire in the outer branches. The 'access'
component comprises of a number of different technologies and connects consumers to the
actual 'core' network. In practice we can subdivide the access network into two main
components - wired and wireless. The wired access network uses digital subscriber lines and
in some regions even older analog phone lines to provide connectivity to consumers. The
wireless access network is mainly mobile technology such as GPRS and G3, or where fixed
wireless access makes economic sense WiMax or HPSA may be used. And in some
geographical areas more advanced G4/LTE mobile networks have begun deployment.
From an energy perspective the access network is the greedy part of any service provider's
network. For a typical ISP it will consume 2/3 of the network electricity although as some
components (e.g. ADSL modems) are on the customer's premises these do not impact on
electricity costs for the service provider. It is also worth remarking that wireless technologies
are significantly more costly in terms of electricity consumption. There is a significant body
of recent research, in particular on LTE (long-term evolution) networks that will provide the
infrastructure for next-generation mobile networking and we will discuss this in more detail
in section 5.4 when we consider forward projections.
A,1<(-B.$*4D$G*1*0,41,-.$
In much of the literature the networking and datacenter components are intertwined and
from an electricity consumption perspective it is not always easy to disentangle them. Now
while it is tempting to combine these into a single category we feel it is important to
distinguish between the electricity growth of data-centers and that of the networking
infrastructure.
The deployment of a new generation of data-centers is targeted specifically at consumer
applications and services and indications are that this rebirth of the 'cloud computing' concept
of the 1960s will catalyze a new growth phase for the ICT industry. But it is important to
realize that this will be very different from the 'passive' data network we call the Internet
where content was stored and managed on a relatively small number of 'servers' and delivered
to a much larger number of clients.
New data services aim to encourage consumers to generate much of their own data and
content that will then form the 'raw material' of this new growth phase. Much of this data will
be stored and processed in data centers and shared over a variety of 'social networking'
channels. Naturally growth in the volume of raw data will also drive network capacity, but we
feel it is important to understand this relationship, how increasing data volumes affect both
growth of the underlying infrastructure and the fundamentals of energy consumption. For this
;B(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
reason we look to research that separates the data-center component of energy from the
networking component.
B>A>A(\$1E.8(&'(01'+2"3.&1'(Y;WWPM;W6;[(
Table 4(a) shows a comparison of the total electricity consumption of global networks from
a number of previous researchers. Note that the network growth figures for Corcoran [16]
have been adjusted to take account of the delayed deployment of LTE networks. After this
adjustment we can see that all researchers have broadly convergent estimates leading to a
consensus figure of the order of 350 TWh/yr for 2012.
Table 4(a): Electricity consumption (TWh/yr) for all data communications networks (2008-2012)
from previous researchers; note that Kilper, et al. [78] is included to illustrate the potential impact of
LTE wireless access technology on overall power consumption.
Citation
Devices
2008
2009
2010
2011
2012
Pickavet, et al. [17]
Datacom & Teleco
Networks; excludes Data
Center networks
219
245.3
274.7
307.7
344.6
Corcoran [16]
Datacom & Teleco
Networks; excludes Data
Center networks
219
251.9
276
318
2
3652
Lambert, et al. [20]
All networks - based on
subscriber numbers
240
265
320
354
Kilper, et al. [78]
Mobile Network
812
Note that the figure provided for Kilper, et al. [78] is derived from table 5 of Lambert, et al.
[20] and illustrates the potential effects of LTE access networks if these were introduced
across the entire communications infrastructure.
One way to view this figure of 812 TWh/yr is to consider it as an upper bound on
electricity consumption for the network infrastructure deployed in 2012. It also tells us that a
communications network that uses LTE, or equivalent wireless technology exclusively for its
access infrastructure can consumer up to 2.2 times as much energy compared with a network
where the primary access is via wired connections.
C"C$54,- E6 $7 ( 4 .8 9/12(4$)6$G*1 *$7 , 4 1, -. $
Obtaining reliable information on Data Centers is difficult. Some corporations are very
open about their energy policies and usage, but others are not [115].
It is clear that with the recent growth in cloud computing and the outsourcing of a wide
range of IT services that there are many providers of data-centers. Some are large
corporations, but there are also a significant number of smaller and more specialized cloud
service providers. Similarly, there is a wide range of data-center infrastructure ranging from
massively scalable and geographically distributed infrastructures, employed by major
corporations like Google and Amazon, to smaller more specialized local installations that are
customized for a particular service, or application.
We can expect to see ongoing rationalization in the sector over the next decade, and
eventually standards and industry best practice should lead to consolidation and improved
efficiencies across the sector. But today it is one of the fastest growing technology sectors
and as a consequence it is difficult to arrive at a clear picture of where cloud computing and
the underlying data-center infrastructure is heading. Nevertheless we should be able to find a
2 The original estimate provided in this research paper [14] assumed that significant LTE deployments would
begin in 2011; in fact major roll-outs of LTE networks have been rescheduled by most global ISPs to a 2013/14
timeframe; the estimates providedTable here are modified to take account of this change.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
;H(
sensible baseline value for 2012 energy consumption and this is the main focus of this
section.
B>B>6(<$#D&12+(S#+#:$- 8# $+(
Much of the definitive work in this area has been undertaken by Koomey [30], [108], [33].
His work was used by other researchers [16], [17] to arrive at estimates for electricity
consumption in data centers over the period 2008-2012. These estimates are provided in
Table 4b below.
More recently Koomey has undertaken a detailed audit of his earlier work and employing a
number of different methodologies arrived at a refined, and lower estimate for electricity
consumption of data centers with a specific focus on the accuracy of his 2010 estimates [31].
This employs a number of alternative methodologies to develop a range of estimates and
ultimately the output is a significantly reduced estimate of the global energy consumption
figures for 2010.
B>B>;(5"3:-.(14(!"#$% &' % () # - 8 ' 1 ,1 % &# +(?(9/'#$%&#+(
There are a number of technological reasons that can help explain the significant
differences between the earlier work of Koomey and his most recent results [31].
Firstly we note the trend of data centers to allocate resources using virtual machines, rather
than physical servers. This enables operational hardware to operate at close to 100%
utilization and unneeded physical servers to be turned off, keeping a buffer of operational
servers to handle fluctuations in load demand. By operating physical servers at high load and
switching off unneeded computing capacity the overall efficiency of the data center is
optimized. Several researcher have written extensively on this approach [3], [11], [14], [116]
[90], [91] [119], [120] and it has been widely adopted as 'best practice' across the industry. It
is not clear what the overall impact of this change in operational procedures has been across
the industry but it has certainly contributed to the revised estimates provided in [31].
A second trend on the hardware side has been a transition to low-power, multi-core CPU
platforms [121], [122] and even GPU clusters [123], [124] for computationally intensive
works such as video processing and transcoding. As new CPU designs are currently driven by
the rapid growth of the smartphone and tablet markets we expect further enhancements in
multi-core and GPU on-chip CPU designs. Many of these architectural enhancements will
likely find their way into next-generation data centers and such trends have undoubtedly
contributed to the improvements presented in [31].
B>B>A(\$1E.8(&'(01'+2"3.&1'(Y;WWPM;W6;[(
It should be clear from the above discussions that the work of Pickavet et al [17] and that
of Corcoran [16] were based on the original estimates of Koomey [30], [33] and as a
consequence they significantly overestimate the actual power consumption of data centers.
This is illustrated in Table 4b where Koomey's original work [30], [33] and his later revisions
[31] are both included.
As Koomey has only provided an estimate for 2010 it was also decided to include an
estimate of consumption for 2011 and 2012 as our goal is to provide a reference baseline
value for 2012 in each of the four main categories of electricity consumption. To this end we
have adopted an annual growth rate of 12% for 2011 and 2012. This is in line with the growth
rates identified by Pickavet, et al. [17]. Now Koomey [31] has indicated a lower rate of
growth from 2005-2010 of about 8%. However he has speculated that the slowdown in the
growth of data centers was at least in part due to external economic conditions. Moreover we
note that a significant number of new consumer cloud services have been introduced since
2009 (Dropbox, Sugarsync, Amazon cloud drive, Skydrive, Google drive, etc). In addition
online storage, sharing and processing of personal content such as pictures (Flickr, Snapfish,
Photobucket) and video (youtube) went through a strong growth phase during this period.
;J(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
Given these developments it seems reasonable to revert to a 12% growth rate for data
centers for 2011 and 2012.
Table 4b: Electricity consumption (TWh/yr) for data centers from previous researchers.
Citation
Devices
2005
2008
2009
2010
2011
2012
Pickavet, et al. [17]
Severs &
Data Center
254.0 (1)
284.5
318.7
356.9
399.7
Corcoran [16]
Severs &
Data Center
254.0 (1)
285.6
343.0
411.5
514.4
Koomey [30], [33]
Severs &
Data Center
152.5
268.7
Koomey [31]
Severs &
Data Center
237.4
265.9 (2)
297.8 (2)
Notes: (1) Pickavet and Corcoran have lower estimates for 2008 than Koomey because network
energy (c. 6%) within the data center is not included; (2) These are extrapolated from Koomey's
revised 2010 estimate at an annual growth rate of 12%; these values include c. 6% network energy.
C"N$O*.,'24,$5.129*1,.$=(-$!P#!$$
This brings us to our final estimates for 2012 summarized in Table 5 below.
It is worthwhile to recap our rationale for establishing these values. They are based on a
combination of previous research work and estimates and our own best estimates deduced
from a combination of known market, economic and, where available, hard data for each
metric.
For direct consumption by devices we have compared the estimates of previous researchers
and matched this with our own model of device categories and estimated per device
consumption and usage patterns. The selected value is lower than some of the estimates of
previous researchers, but seems sensible given global economic conditions over the last few
years.
The contribution of networks offers perhaps the best consensus. This does require some
adjustment of the projections of Corcoran [16] to correct for delays in the roll-out of LTE/G4
networks. After this we find the estimates of network energy usage are within 20 TWh/yr of
each other and the recent work of Lambert et al [20] offers a median estimate. We remark that
the core network component of datacenters is included in this estimate and thus estimates for
energy consumption in datacenters shoudl be reduced accordingly - by approximately 6%.
Table 4(c): Best Estimate Values for 2012 in each category.
2012 Baseline
Device Consumption
852 TWhr/yr
Manufacturing (LCA)
330 TWhr/yr
Networks
352 TWhr/yr
Data Centers
281 TWhr/yr
Total
1,815 TWhr/yr
Given the relatively close agreement between these estimates for network we do not
provide an alternative model at this point, but in section 5 the energy consumption of the core
network is separated from that of the wireless access network. This separation is based on the
observed trend that mobile data will employ wireless access networks and given the projected
growth in such data [28] there will be a equivalent growth in the energy consumption by these
wireless access networks [79]. This will be discussed in more detail in section 5.4 where our
projections will be made separately for core network and wireless access networks.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
;O(
For Data Centers we note that much of the previous research was influenced by Koomey
[30] who has since provided more accurate, revised estimates for the year 2010 [31]. Given
the detailed nature of the latter work we adopt this estimate for 2010 but assume a year-on-
year growth rate of 12% from 2010 to 2012 rather than the 8% annual growth rate identified
from 2005 to 2010 in the same work by Koomey. The rationale for this higher growth rate is
the ramp-up in 'cloud computing' since 2010, in particular the introduction of a wide range of
new consumer services. Further details were given in section 5.4 above. Our 2012 estimate of
almost 300 TWh/yr for Data Centers is then reduced by 6% to account for the energy usage of
core data networks inside data centers.
H>(<$1]#-.&1'+(41$(0!M 50 ) (!'# $%/(01'+2"3.&1'(Y;W6A(Z(; W 6 O[ (
Having made some estimates using different approaches for electricity consumption in
2012 we now seek to extrapolate to determine likely growth scenarios over the period 2013-
2017. This can be achieved by extending the same approaches used to determine our 2012
figures for each principle category of energy consumption.
For direct consumption by devices we develop projections that cover likely market growth
patterns over the period 2013-2017. We can compare these with projections from the 2009
IEA study [24] and these can be matched with overall growth rates in energy consumption by
the installed base of CE-ICT devices. Here it is reasonable for a number of reasons to assume
that overall consumption by devices will remain in a state of overall balance as increased
efficiency of new devices balances any overall growth in the installed base.
For network consumption we take the approach to separate energy consumption by the core
network from that of the peripheral wireless access network. As was discussed earlier, the
core network is already very efficient and thus we can use the global metric of TWh/EB
which is available for many network operators to provide an estimate of core network
electricity consumption. Naturally we expect this to be continually improving and thus our
various scenarios assume a fixed rate of annual improvement. Against this is the steady
growth of data traffic which is predicted to rise from 2600 Exabyte in 2012 to 8500 Exabyte
in 2017 [29].
Electricity consumption by data centers will also rise inevitably, driven in particular by
increased consumer adoption of thin client devices. We can use our estimates of global
growth in communications networks to determine overall growth rates in the scale of
underlying infrastructure. In turn these growth rates suggest how the data-center infrastructure
will grow to accommodate the increase in network data. It seems reasonable to assume a
linear relationship between increased data traffic and the processing and storage infrastructure
associated with this.
N"#$G2==,-,41$&-(<1;$Q0,4*-2(.$
Based on the above approaches to extrapolating energy consumption we can now provide a
set of criteria that define our three main growth scenarios. We first consider a low-growth
case that assumes best conditions in all three categories of electricity consumption. Then an
expected-growth assumption where sensible growth estimates are balanced against improved
energy efficiencies. The final case looks at continued growth in all categories with only small
improvements in efficiency acting to restrain energy consumption.
Interestingly we can match these three scenarios with the three cases outlined in Annex I of
the IEA study [24]. Detailed tables of each scenario will be provided later. In the remainder of
this section we provide the main input assumptions and a summary of the outcomes in terms
of electricity consumption.
;P(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
H>6>6(T1E(\$1E.8(7++2"3.&1'+(
For devices we assume that the installed base will remain essentially fixed from 2013-2017
with market size also remaining stable. Thus every new CE-ICT device displaces an older
model. This leads to an overall year-on-year improvement in energy efficiency of 5% for
most device categories. In this scenario we see energy usage by the installed base of CE-ICT
devices falling from 852 TWh/yr to 682 TWh/yr.
For network technology it is clear that the growth in data traffic will not allow any overall
reduction in energy consumption. We choose a fairly ambitious target of 15% per annum
improvement in the efficiency of the core network from a starting point of 0.135 TWh/EB.
Even with this level of across-the-board improvement energy consumption in the core
network is expected to rise from 351 TWh/yr to 509 TWh/yr. This scenario also assumes
relatively slow growth in wireless traffic via the access network with only 5% of overall
electricity consumption being due to wireless access networks and an total for core + wireless
of 535 TWh/yr.
Finally our determined rate of growth in infrastructure for this low-growth case is about
7.5%. Applying this to our 2010 base for data center electricity usage we determine 258
TWh/yr in 2013 rising to 370 TWh/yr in 2017.
H>6>;(!L3#-.#*(\$1E.8(7++2"3.&1'+(
For devices we assume a stagnating market for desktop computers, TVs and device
screens. There is a modest growth of 8% in laptops and the markets for smartphones and
tablets will plateau around 1.2 billion and 400 million units per annum by 2017 (These
numbers are actually conservative compared with current market estimates [112]). Energy
efficiency improvements for most devices are limited to 2% per annum - this implies more
older devices are retained in use, being resold or reused within the same household. This
scenario keeps electricity consumption from devices stagnant at 854 TWh/yr over the period
from 2013-2017.
For networking technology we assume a starting point of 0.14 TWh/EB and an annual
improvement in energy efficiency of 10% across the core network. Wireless access network is
limited to 9% of total electricity usage and total consumption rises to 733 TWh/yr by 2017. In
turn this determines a growth rate for infrastructure for this mid-level scenario of 14% leading
to a data-center figure for electricity usage of 558 TWh/yr.
H>6>A(^&%8(\$1E.8(7++2"3.&1'+(
In the case of devices we assume that current trends in terms of market sizes and
replacement rates continue. The market for TV panels continues at 250 million units with
similar replacement rates (8 year lifecycle). This leads to an increase in the installed base of
TV sets from 2 billion to 2.5 billion over the period 2013-2017. Desktop PCs and monitors
continue a steady but small growth of 2% while the installed base of laptops continues a
strong growth rate c.14% and smartphones and tablet markets grow to 1.6 billion and 450
million respectively. Annual efficiency improvements vary from 1% to 5% depending on the
particular device category. The net result is a growth in device related electricity consumption
from 852 TWh/yr in 2012 to 1087 TWh/yr in 2017.
Our high-growth scenario for networks assumes only a 5% year-on-year improvement in
efficiency that leads to a growth in core network electricity consumption to 870 TWh/yr. The
wireless access network grows to contribute 15% of total electricity consumption leading to a
combined figure of 1027 TWh/yr. In turn this indicates a growth rate in excess of 20% for the
underlying infrastructure. Applying this to our starting point for data centers we arrive at a
figure of 800 TWh/yr by 2017.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
;V(
N"!$5',01-20216$7(4.89/12(4$)6$75M>7:$G,320,.$
Here we take a look in more detail at these individual numbers.
H>;>6(!,#-.$&-&./(01'+2"3.&1'(@:+#*(1'(5'+.:,,#*(N#D&-#(X2"@#$+(
Here we take the device-centric model developed in section 4.1.2 and apply it to estimate
growth in the installed base for the various devices categories provided. As before there are
two main groupings - TV and TV-peripherals form one of these, the other being that of ICT
client devices ranging from desktop PC's to smartphones.
Our estimates for energy use per devices are drawn from [35] applying a fixed year-on-
year improvement in device efficiency to each device category. Most ICT devices are
assumed to have a short life cycle of 1.5-3 years, but TV peripherals are typically much
longer with an 8-year lifecycle.
%$J2-.1$Q,1$(=$%..89/12(4.$M$O8.24,..$*.$R.8*'$=(-$7'2,41S$:T$H*-B,1.$
Our initial assumptions correspond to a high-growth scenario and are based on continued
growth in the installed base numbers for each device category. It is further assumed that many
older devices are resold or reused thus there is significant growth in the installed bases of
laptops and TV panels. There is only very slight growth in the installed base of desktop
computers and monitors as laptops, tablets and even smartphones are gradually superseding
these.
For the various categories of TV peripheral we assume that: (i) 38% of TV panels have
some form of set-top-box; (ii) 20% have a game console; (iii) 30% have an external A/V
surround amplifier/speaker system and (iv) 35% have a DVD or Blue-ray disk player. Many
TV sets will have more than one of these peripherals and the power ratings applied reflect the
somewhat intermittent usage of some peripherals. For example, DVD players have an average
electric power consumption of 28 KWhr/yr because they are used less frequently than a set-
top box (100 KWhr/yr) which is often operated on a 24/7 schedule or an A/V receiver (65
KWhr/yr) that is typically switched on while the TV panel is operational. A detailed
discussion of methodology can be found in [35].
Year-on-year device efficiency for older ICT devices is assumed to improve by 5%
whereas newer smartphone and tablet devices are actually considered to lose efficiency at a
similar annual rate due to the use of improved battery technology allowing increases in the
total electricity use of these devices. In other words these devices will be able to carry more
battery charge but that will drive a trend towards more power hungry devices. Thus we
anticipate a negative efficiency for such devices, at least for the next few years.
In due course we shall compare this high-growth scenario to the 'business as usual' scenario
of IEA [24] and it will be clear that there is a close correspondence between the two. It should
be remarked that there was already a sharp drop (c 10%) in the sales of desktop and laptop
computers in the last quarter of 2012 and the first quarter of 2013 [43]. It is not yet clear if
this marks an underlying trend, or is simply a market aberration caused by the introduction of
Windows 8
TM
, followed by a slow initial adoption by consumers. However it does seem likely
that our high-growth scenario delineates the likely worst-case electricity consumption over
the next 5 years.
Table 5(a): Worst case consumption (2013-2017): direct energy consumption by category of device.
Device Numbers
x10
6
(Clients)
2012
2013
2014
2015
2016
2017
Desktops
588
598
608
619
629
640
Monitors
617
628
639
649
660
672
Laptops
832
946
1076
1223
1391
1581
AW(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
Smartphones
1000
1200
1296
1400
1512
1633
Tablets
150
190
236
293
363
451
Device Numbers
x10
6
(TV)
TV
2000
2100
2205
2315
2431
2553
TV STB
760
798
838
880
924
970
TV GC
400
420
441
463
486
511
A/V Receiver
600
630
662
695
729
766
DVD/Blueray
700
735
772
810
851
893
Power Consumption
(TWhr/year)
Desktops
129
132
127
129
131
134
Monitors
60
61
61
62
63
64
Laptops
67
76
82
93
106
120
Smartphones
5
6
7
7
8
9
Tablets
2
3
4
5
6
7
TV
400
399
419
440
462
485
TV STB
76
78
82
86
91
95
TV GC
54
57
60
63
66
69
A/V Receiver
39
41
43
45
47
49
DVD/Blueray
20
21
22
23
24
25
Total (TWhr/year)
852
872
905
952
1,003
1,057
%'1,-4*123,$%..89/12(4.$U#$M$7'2,41S$:T$H*-B,1.$G,0'24,$1;,4$&-(<$Q/(-*D20*''6$
An alternative scenario is our 'expected growth' estimate. This broadly corresponds to the
least life-cycle cost scenario of IEA [24]. In this case we have introduced a sudden 10% drop,
year-on-year, in 2013 for the desktop and laptop markets. This is followed by a slow 2% year-
on-year growth in the installed base for desktops but a much higher 14% growth year-on-year
for laptops. Despite the rapid adoption of smartphones and tablets we still consider that laptop
computers are more versatile for most users and the installed base will continue to grow
driven by developing markets in Asia and South America.
We are also conservative with regard to the installed base of TV panels. We see this
peaking in 2014 and entering a slight decline, stabilizing at 2.1 billion units. Given the
increased number of tablet devices and smartphones and the growth of online video content
there is less need for a TV set in the bedroom and we believe that many 2nd and 3rd TV sets
will find themselves gathering dust. Consumers are also likely to engage with proactive
recycling programs to dispose of older energy-hungry panels. TV panels represent 50% of the
overall electricity consumption budget for CE-ICT devices so any significant savings start
with this device category.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
A6(
Table 5(b): Expected Consumption (2013-2017): direct energy consumption by category of device.
In this 'expected consumption' scenario we anticipate some year-on-year improvements in
device efficiencies but these will be gradual. With TV sales of 250 million units per annum it
would take 8 years to replace all existing TV panels assuming 1-to-1 replacements. Over the
5-year period covered by our estimates lets assume that a panel with 20% less power
consumption replaces 50% of TV panels. This would provide overall net savings of 10%, or
annual savings across all TV panels of 2% year-on-year.
Note that we did not chose a higher efficiency for this 'expected consumption' scenario
because many upgrades are to a larger TV panel or a smart-TV and this will offset some of
the gains from higher efficiency panels.
$
%'1,-4*123,$%..89/12(4$U!$M$Q1*E4*41$H*-B,1.$<21;$&-(<1;$24$5==202,406$$
Our last scenario uses similar growth assumptions for the ICT sector and caps the installed
base of TV panels at around 2 billion in 2017. As discussed in the 'expected consumption'
scenario devices like tablets and smartphones are seen as displacing secondary TV sets during
this period in the same way they have recently begun to displace desktop and laptop
computers. However most households will retain a main TV set.
Device Numbers
x10
6
(Clients)
2012
2013
2014
2015
2016
2017
Desktops
588
529
537
546
554
563
Monitors
617
556
565
573
582
591
Laptops
832
749
812
880
955
1035
Smartphones
1000
1143
1176
1209
1244
1279
Tablets
150
190
229
275
331
398
Device Numbers
x10
6
(TV)
TV
2000
2100
2205
2100
2000
2100
TV STB
760
798
838
798
760
798
TV GC
400
420
441
420
400
420
A/V Receiver
600
630
662
630
600
630
DVD/Blueray
700
735
772
735
700
735
Power Consumption
(TWhr/year)
Desktops
129
116
116
118
119
121
Monitors
60
54
54
55
55
56
Laptops
67
60
64
69
75
81
Smartphones
5
6
6
6
7
7
Tablets
2
3
4
4
5
6
Power Consumption
(TWhr/year)
TV
400
412
432
403
376
380
TV STB
76
78
82
77
72
72
TV GC
54
57
60
57
54
57
A/V Receiver
39
41
43
40
38
39
DVD/Blueray
20
21
22
21
20
21
Total (TWhr/year)
852
846
881
849
821
840
A;(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
Table 5(c): Best Case Consumption (2013-2017):, energy usage by category of device.
In this scenario, which is analogous to the 'best available technology' case of IEA [24] we
assume year-on-year improvements in efficiency of 5% - implying double the savings in
electricity consumption for TV panels, desktop and laptop computers. When combined with
stagnating markets this leads to a net decline in energy consumption, dropping to 665 TWh/yr
by 2017. It is not clear that such savings are realizable, but recent trends such as the
substantial improvements in power consumption realized by TV panel manufacturers over the
last 3 years and significant declines in PC sales during 2013 do suggest that this scenario may
not be such an outlier. If further technology breakthroughs are combined with smart and
proactive government policies there is no reason.
H>;>;(01"3:$:.&D#(*:.:+#.+((
It is interesting to match our model projections with those of other research studies, but
unfortunately there are not too many detailed studies available in the literature.
Another approach is to match our three model scenarios with equivalent fixed growth rates.
This latter approach provides us with an overall trend for device consumption that can be
easily matched with economic and market trends.
Device Numbers
x10
6
(Clients)
2012
2013
2014
2015
2016
2017
Desktops
588
529
521
579
588
529
Monitors
617
556
547
608
617
556
Laptops
832
749
656
729
832
749
Smartphones
1000
1143
1149
1156
1163
1169
Tablets
150
190
217
247
281
321
Device Numbers
x10
6
(TV)
TV
2000
2000
2050
2101
2050
2000
TV STB
760
760
779
798
779
760
TV GC
400
400
410
420
410
400
A/V Receiver
600
600
615
630
615
600
DVD/Blueray
700
700
718
735
718
700
Power Consumption
(TWhr/year)
Desktops
129
116
109
103
111
90
Monitors
60
54
50
53
51
42
Laptops
67
60
50
53
57
46
Smartphones
5
6
6
6
6
6
Tablets
2
3
3
4
5
6
Power Consumption
(TWhr/year)
TV
400
390
399
370
343
324
TV STB
76
70
72
63
55
49
TV GC
54
54
56
53
51
49
A/V Receiver
39
38
39
38
37
37
DVD/Blueray
20
19
20
18
17
16
Total (TWhr/year)
852
810
804
761
733
665
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
AA(
7(4.89/12(4$7*'08'*1,D$=-(9$>5%$[24]$
This detailed study by IEA [24] is the most detailed research study available on the
electricity usage of CE-ICT devices. Even though it dates from 2009 many of the finding
remain very relevant.
We note that IEA identified three different growth scenarios, similar to our own, shown in
table 6(a). These are taken from Annex I, p372 of [24] with intermediate values obtained by
simple linear interpolation. These projections are quite compelling, aligning closely with our
own estimates from section 5.2.1 and
The equivalent to our high-growth estimate corresponds to the IEA business as usual
scenario. Similarly our expected growth equates to IEA's least life cycle (LLC) and our low
growth matches their best available technology (BAT). Numerically all the IEA estimates are
in similar ballparks. Their LLC is a bit low in its 2012 estimate, but matches our 852 TWh/yr
closely by 2017; the BAT scenario shows negative growth, achieving 678 TWh/yr by 2017
compared to our 665 TWh/yr; finally the high growth estimate from IEA lies close to our 852
TWh/yr in 2012 and grows at a similar rate achieving a higher 1,112 TWh/yr to our 1,057
TWh/yr, but from a higher starting point of 873 TWh/yr.
Table 6(a): TWh/yr for IEA scenarios - BaU (high), LLC and BAT (low) for period 2013-2017.
IEA Predicted
(TWhr/year)
2012
2013
2014
2015
2016
2017
Low Growth (Best
Available Technology
740
722
704
686
682
678
Expected Growth
(Least Life Cycle)
793
802
810
819
838
857
High Growth
(Business as Usual)
873
921
970
1,018
1,065
1,112
7(4.89/12(4$)*.,D$(4$J2K,D$&-(<1;$@*1,.$
While there are not many detailed research studies on electricity consumption of CE-ICT
devices is interesting, as discussed above, to match our model predictions with overall growth
trends. If we take our 2012 baseline of 852 TWh/yr we find the results of table 6(b) for
growth rates of -5%, 0% and +5%. These align very closely with our three different growth
scenarios.
Table 6(b): TWh/yr for fixed growth rates of -5%, 0% and +5% (2013-2017).
Growth Rate
2012
2013
2014
2015
2016
2017
Low ( -5%)
852
809
769
730
694
659
Expected ( 0%)
852
852
852
852
852
852
High (+5%)
852
895
939
986
1,036
1,087
Our conclusion here is that the three scenarios that we explored in section 5.2.1 can be
reasonably matched with these three simpler scenarios. There is no doubt that the per-unit
electricity usage by CE-ICT devices is improving. New TV panel display technology and the
transition from desktop computer clients to laptops, tablets and smartphones will all bring
down the average consumption per device - in some cases quite significantly. At the same
time we are seeing stagnation in traditional market segments, but rapid growth in some new
device segments. A final consideration is that many older devices are making their way to
developing economies, often as e-waste [48]. Yet much of this e-waste is repaired and re-
enters the local market extending its lifecycle. It is to be expected that such practices will
continue and become more widespread as long as a global economic downturn persists, even
where the economic benefits are marginal.
When we balance all these factors we see that while the stagnation of traditional markets
for PC and TV panels suggests that the installed base of such devices will not grow
significantly there is likely to be an extension of device lifecycle as consumers 'hold on' to
devices beyond their normal lifecycle or where devices that are scrapped and exported as e-
AB(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
Waste are brought back into active service in the gray economy. In parallel there is a gradual
improvement in per-device efficiency as newer devices gradually replace older ones. For the
newer devices, their contribution to direct electricity consumption is very small compared
with the PC and TV segments. A reasonable conclusion is that growth will stagnate, or even
decline on an annual basis. Only if the global economy can achieve a significant turnaround
in the next couple of years are we likely to see any return to growth in traditional CE-ICT
markets over the next 4-5 years.
H>;>A()8#(9&%'&4&-:'-#(14(N#D&-#(T&4#.&"#((
The total electricity consumption by CE-ICT devices is dependent on the installed base. In
turn the installed base depends on the number of new devices that are entered into service and
the corresponding number of devices removed from service. These, in turn, are correlated
with the device lifetime.
Of course device lifetime varies considerably even across a single type of device.
Nevertheless we can say that a TV typically has a longer lifetime than a desktop computer or
laptop. Here, however, we are simply interested in the average lifetime of each device
category. More specifically the dependency between shipped units and lifetime has not been
considered in sections 5.2.1 and 5.3. The number of new devices entering service can be
estimated and this establishes the installed base numbers based on a fixed device lifetime.
However variations in device lifetime can have very pronounced effects on both the installed
base of devices and the total energy consumption.
The results for Best, Expected and Worst Device electricity consumption are provided
below and cans be used as a sensitivity analysis with regard to previous sections 5.2.
():@,#(OY:[_(Short (Best) lifetimes
Device Numbers x10
6
(Clients)
2012
2013
2014
2015
2016
2017
Desktops (3y)
445
452
459
467
475
482
Monitors (3y)
495
499
504
510
525
540
Laptops (3y)
662
745
862
1000
1159
1343
Smartphones (2y)
1160
1600
2075
2565
3020
3556
Tablets (2y)
250
400
600
850
1070
1347
Device Numbers x10
6
(TV)
TV (8y)
1973
1993
2016
2042
2071
2100
TV STB (8y)
691
697
706
715
725
735
TV GC (8y)
296
299
302
306
311
315
A/V Receiver (8y)
592
598
605
613
621
630
DVD/Blueray (8y)
691
698
706
715
724
734
Power Consumption (TWhr/year)
Desktops
98
94
96
93
90
82
Monitors
48
46
46
45
44
41
Laptops
53
57
66
72
79
83
Smartphones
6
8
11
13
16
19
Tablets
4
6
9
13
18
23
Power Consumption (TWhr/year)
TV
395
379
383
369
355
340
TV STB
69
63
63
58
53
48
TV GC
40
40
40
40
39
38
A/V Receiver
38
37
37
36
35
33
DVD/Blueray
19
19
19
18
17
17
Total (TWhr/year)
770
747
771
756
746
725
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
AH(
):@,#(OY@[_(Medium (Expected) lifetimes
):@,#(OY-[_(Long (Worst) lifetimes
Device Numbers x10
6
(Clients)
2012
2013
2014
2015
2016
2017
Desktops (5y)
729
741
753
766
778
791
Monitors (5y)
811
820
830
843
859
875
Laptops (5y)
975
1116
1278
1460
1689
1953
Smartphones (2y)
1160
1600
2075
2565
3020
3556
Tablets (2y)
250
400
600
850
1070
1347
Device Numbers x10
6
(TV)
TV (10y)
2455
2475
2498
2524
2553
2582
TV STB (10y)
859
866
874
883
893
904
TV GC (10y)
368
371
375
379
383
387
A/V Receiver (10y)
737
743
749
757
766
775
DVD/Blueray (10y)
859
866
874
883
893
903
Power Consumption (TWhr/year)
Desktops
160
163
162
165
164
164
Monitors
79
79
79
80
80
80
Laptops
78
89
100
114
130
147
Smartphones
6
8
11
14
17
23
Tablets
4
6
10
14
20
28
Power Consumption (TWhr/year)
TV
491
485
490
485
481
467
TV STB
86
85
86
85
84
82
TV GC
50
50
51
51
52
52
A/V Receiver
48
48
48
48
48
48
DVD/Blueray
24
24
24
25
25
25
Total (TWhr/year)
1,025
1,039
1,061
1,082
1,101
1,115
Device Numbers x10
6
(Clients)
2012
2013
2014
2015
2016
2017
Desktops (7y)
1011
1023
1038
1055
1072
1090
Monitors (7y)
1281
1133
1146
1164
1185
1206
Laptops (7y)
975
1116
1278
1460
1689
1953
Smartphones (3y)
1510
2060
2775
3465
4195
5079
Tablets (3y)
300
500
750
1100
1420
1833
Device Numbers x10
6
(TV)
TV (10y)
2455
2475
2498
2524
2553
2582
TV STB (10y)
859
866
874
883
893
904
TV GC (10y)
368
371
375
379
383
387
A/V Receiver (10y)
737
743
749
757
766
775
DVD/Blueray (10y)
859
866
874
883
893
903
Power Consumption (TWhr/year)
Desktops
223
225
224
227
227
226
Monitors
124
110
109
111
110
110
Laptops
78
89
100
114
130
147
Smartphones
8
11
15
19
24
32
Tablets
5
8
12
19
26
38
Power Consumption (TWhr/year)
TV
491
485
490
485
481
467
TV STB
86
85
86
85
84
82
TV GC
50
50
51
51
52
52
A/V Receiver
48
48
48
48
48
48
DVD/Blueray
24
24
24
25
25
25
Total (TWhr/year)
1,135
1,135
1,158
1,184
1,207
1,227
AJ(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
One immediate observation is that variations in the device lifetime can have very
significant effects on the installed base and consequently the total electricity consumption.
Longer device lifetimes are better in terms of reduced manufacturing energy, but imply that
older, less energy efficient appliances remain in use longer. We did not find any extensive
studies in the literature but there is a clear need to better understand patterns of appliance re-
use and, in the global context, how emerging secondary markets (e.g. Africa) are evolving.
N"?$+-(L,012(4.$=(-$H*48=*018-24E$5',01-20216$$
For the future projection of manufacturing electricity the variable is the kWh/unit and the
share of the Use stage of total lifetime electricity for Networks and Data Centers. The
number of shipped units is also variable but has not been explored.
Data sources for device shipments are given in Section 4.2.2. The tables presented below
are deduced using the same methods described in Section 4.2 (table 3). We provide three
examples from the “expected case” scenario to clarify the TWh obtained in these tables,
namely 1) tablets in 2015 2) data centers in 2017 and 3) monitors in 2014. (Note that the
starting year for these estimates is 2008 manufacturing data, thus 5% efficiency improvement
is applied over 7 years for 2015 estimates, 9 years for 2017 estimates, etc)
1) Tablets: 500 million × 75kWh×(1-0,05)
7
= 26 TWh,
2) Data Centers: 541 kWhr / 0.85)} x (1 – 0.85) ×(1-0,05)
9
= 60 TWh,
3) Monitors: 170 million × 268kWh × (1-0,05)
6
= 33 TWh.
):@,#(PY:[_((Unit numbers entering service and estimated manufacturing electricity per unit.
Best Case
2013
2014
2015
2016
2017
KWhr/unit
(2010)
Desktops
153
156
158
161
164
60
Monitors
165
170
175
180
185
187
Laptops
284
332
384
443
511
75
Smartphones
900
1175
1390
1630
1911
30
Tablets
250
350
500
570
650
78
TV
261
264
267
270
273
400
TV STB
99
100
101
102
103
41
TV GC
52
53
53
54
55
150
A/V Receiver
78
79
80
81
82
100
DVD/Blueray
91
92
93
94
95
200
Networks eq (TWhrs)
383
416
453
495
544
Use stage 90%
Data center eq (TWhrs)
302
325
349
375
403
Use stage 90%
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
AO(
):@,#(PY@[_(Manufacturing Electricity Consumption (TWh/yr) - best case scenario.
2013
2014
2015
2016
2017
Desktops
23
23
23
22
22
Monitors
35
34
34
33
33
Laptops
28
30
32
34
36
Smartphones
27
32
35
38
42
Tablets
14
19
26
28
31
TV
81
81
81
81
81
TV STB
3
3
3
3
3
TV GC
6
6
6
6
6
A/V Receiver
6
6
6
6
6
DVD/Blueray
14
14
14
14
14
Networks eq
52
53
54
56
57
Data center eq
41
42
42
43
43
Total (TWhr/yr)
330
343
356
365
375
(
):@,#(VY:[_((Unit numbers entering service and estimated manufacturing electricity per unit.
Expected
2013
2014
2015
2016
2017
KWhr/unit
(2010)
Desktops
153
156
158
161
164
188
Monitors
165
170
175
180
185
268
Laptops
284
332
384
443
511
134
Smartphones
900
1175
1390
1630
1911
40
Tablets
250
350
500
570
650
75
TV
261
264
267
270
273
400
TV STB
99
100
101
102
103
45
TV GC
52
53
53
54
55
150
A/V Receiver
78
79
80
81
82
100
DVD/Blueray
91
92
93
94
95
200
Networks eq (TWhrs)
407
469
542
629
731
Use stage 85%
Data center eq (TWhrs)
320
365
416
475
541
Use stage 85%
):@,#(VY@[_(Manufacturing Electricity Consumption (TWh/yr) - expected case.
2013
2014
2015
2016
2017
Desktops
22
22
21
20
19
Monitors
34
33
33
32
31
Laptops
29
33
36
39
43
Smartphones
28
35
39
43
48
Tablets
15
19
26
28
31
TV
81
78
75
72
69
TV STB
3
3
3
3
3
TV GC
6
6
6
5
5
A/V Receiver
6
6
6
5
5
DVD/Blueray
14
14
13
12
12
Networks eq
56
61
67
74
81
Data center eq
44
47
51
56
60
Total (TWhr/yr)
338
356
375
390
408
AP(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
%
):@,#(6WY:[_((Unit numbers entering service; estimated manufacturing electricity per unit.
Worst Case
2013
2014
2015
2016
2017
KWhr/unit
(2010)
Desktops
153
156
158
161
164
215
Monitors
165
170
175
180
185
334
Laptops
284
332
384
443
511
167
Smartphones
900
1175
1390
1630
1911
60
Tablets
250
350
500
570
650
287
TV
261
264
267
270
273
500
TV STB
99
100
101
102
103
50
TV GC
52
53
53
54
55
150
A/V Receiver
78
79
80
81
82
100
DVD/Blueray
91
92
93
94
95
200
Networks eq (TWhrs)
431
528
649
797
981
Use stage 80%
Data center eq (TWhrs)
337
405
485
583
699
Use stage 80%
(
):@,#(6WY@[_(Manufacturing Electricity Consumption (TWh/yr) - worst case scenario.
2013
2014
2015
2016
2017
Desktops
22
22
21
20
19
Monitors
34
33
32
31
30
Laptops
30
34
38
42
47
Smartphones
31
41
47
54
61
Tablets
30
48
74
83
92
TV
80
76
72
69
65
TV STB
3
8
7
7
7
TV GC
6
12
11
11
11
A/V Receiver
6
12
11
11
11
DVD/Blueray
14
28
27
26
24
Networks eq
62
76
92
111
134
Data center eq
48
57
68
80
93
Total (TWhr/yr)
366
445
501
543
593
%
N"C$A,1<(-B$5',01-20216$7(4.89/12(4$
There is general consensus in the literature that networks are growing as is the amount of
electricity that they consumer [20], [36], [76], [79]. A more difficult question is to try and
quantify this growth in some sensible way. However the communications networking
infrastructure is so intertwined and no one body or organization has an overview of how it is
evolving. And a lot of data on network infrastructure is commercially sensitive, thus no
available publicly.
However there is a way to get a broad sense of network growth and that is to look at
projections for data traffic over the next few years. This information is quite readily available
and in fact a number of annual reports are available [28], [29]. These suggest growth rates of
30% for core network data and 70%+ for mobile data and were discussed previously in some
detail in section 3.3.2. Naturally we don’t expect the energy usage to rise a the same rates,
because improvements in terms of both technology and data management practices will lead
to an annual increase in operating efficiency.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
AV(
Another important point to make is that it is not simply the growth in device numbers that
drives the need for increased network capabilities. In parallel there are more network services
available and thus network usage per device is growing. This is shown in Table 11 which
compares the averaged monthly data usage for different categories of device as determined by
the CISCO Visual Networking Index (VNI) studies [28] from 2009, 2010, 2011 and 2012.
The projected growth rate for the period 2012-2017 is taken from the 2012 VNI.
What is particularly interesting from this comparison is that tablet devices and 4G smart-
phones are predicted to consume as much network data as laptops by 2017 and this is more
than twice the amount of data that is consumed by one of today’s laptops. Thus, in the context
of the network, thin clients such as smart-phones and tablets will be every bit as important in
terms of bulk network data and consequently network electricity consumption as desktops and
laptops are today.
Table 11: Data from the annual CISCO VNI reports - 2009, 2010, 2011 and 2012 - showing evolving
trends in terms of network data consumption (MB/month) by various categories of handheld devices.
From table 11 we can thus begin to understand the radical shift that is occurring and the
potential impact of thin clients on electricity consumption. While these devices use very little
electricity themselves they catalyze growth in communications networks, and in particular in
wireless access networks. From table 11 it can be seen that these devices will match laptops
in terms of their requirement for network bandwidth by 2017. And that, in turn, is twice the
bandwidth usage of today’s laptops.
H>B>6(K:+#*(1'(S:E(N:.:()$:44&-(\$1E.8(
The growth of network electricity consumption and its research by previous authors is
considered in sections 3.3 and 4.3. Our goal here is to determine an approach to extrapolate
our estimated base figure for 2012 over the period until 2017. The most useful and reliable
source of data we have for this period is that of network data traffic. As previously discussed
thus suggests a 30% year-on-year growth rate for core data [29] and in excess of 70% for
mobile data [28].
The next metric we need is the energy cost per unit of data. This was considered recently as
the core topic of Coroama et al [125] where they estimated the energy flow at 0.2 kWhr/GB,
or to convert it to units that make more sense for our scale of estimation we multiply above &
below the line by 10
9
to get 0.2 TWh/EB. As indicated by Coroama et al this is likely a
conservative estimate for the wired Internet so we should search for more exact estimates.
The ‘Corporate Responsibility Report 2010/11’ for Verizon [126] indicates 0.15 TWh/EB
in 2009 and 0.13 TWh/EB in 2010. Verizon has set targets to increase its network efficiency
to achieve below 0.08 TWh/EB by 2020. This is a relatively modern company and these
metrics appear to be relatively high-performing compared to others in the telecoms sector.
For example Telecom Italia reported values of their Eco-Efficiency indicator [127] that
give energy costs for data of 0.32 TWh/EB in 2007 and 0.24 TWh/EB in 2008 [128] but
suggest that further improvements on the 2008 figures will not be realistic. However in a
more recent 2012 sustainability report [129] this metric is again provided indicating its 2010
BW(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
value at 0.17 TWh/EB, 2011 value at 0.13 TWh/EB, and most recently the 2012 value is
0.115 TWh/EB.
Thus after some consideration we have decided to use an industry-wide value of 0.135
TWh/EB representing our best-case 2012 scenario; a slightly higher starting point of 0.14
TWh/EB represents our expected case and our worst case scenario begins at 0.145 TWh/EB.
As with all our estimates these are speculative, but the intention is to show how only a small
change in the overall efficiencies and the targets set by industry can result in wide variations
in the eventual outcomes. For our best case we assume 15% increase in efficiency on a year-
by-year basis – this is even more challenging than the goals set by Verizon for 2020. The
expected case uses a target of 10% annual improvement while our worst case only aims for
5% year-on-year improvement in efficiency. These are summarized in Table 12 below.
):@,#(6;_(Network Electricity Consumption (TWh/yr) estimated from network data traffic.
YoY Efficiency
Increase
2012
2013
2014
2015
2016
2017
Data Traffic (EB)
2600
3300
4100
5200
6600
8500
0.135 TWhr/EB
15%
351
379
400
431
465
509
0.14 TWhr/EB
10%
351
401
449
512
585
678
0.145 TWhr/EB
5%
351
423
500
602
726
888
Here we can see the importance of continual improvement, as even assuming an optimal
industry-wide starting point of 0.135 TWh/EB and 15% annual improvements in efficiencies
we find that network consumption of electricity will rise from the 2012 baseline of 351
TWh/yr to more than 500 TWh/yr. The substantial growth in data traffic is ultimately
responsible for this. For our expected projections the growth is even more substantial leading
to an increase to 678 TWh/yr or a 91% increase on the 2012 consumption of electricity. For
our worst case the 5% annual efficiency improvement is shown to be inadequate with more
than 150% increase in electricity consumption.
H>B>;(F&$#,#++(7--#++(X#.E1$G+((
There is a growing realization in the literature that the primary cost of modern
telecommunications networks does not in fact arise from the core wired or fibre-optic
network, but in fact from the access network – that portion of the network that enables
individual subscribers to connect and gain access to the main core infrastructure. To date this
has not been of great concern to the telecommunications industry as the amount of data that
originates through their wireless access networks is very small when compared with overall
network traffic.
For example, in 2012 mobile networks are expected to carry no more than 0.9 EB of data
per month, or just over a single EB in the entire year; but the global IP traffic, including data-
centers will be 1800 EB over the same year. Thus mobile data is only 0.6% of the network
traffic. By 2017 global IP traffic will be close to 9000 EB whereas mobile IP traffic will be a
mere 134 EB, or 1.5%.
V2-,',..$W(9,$%00,..$X$%$W2DD,4$7(.1Y$$
Of course the figures we give here relate to what ISPs call ‘mobile data’ implying that it is
data gathered by the ISPs own network. This does not include the significant volumes of data
that are introduced to the network via home Wifi routers. In fact quite a significant proportion
of home network access is via a Wifi router but as the electricity costs are paid by the
subscriber this does not impact on the service provider and thus is not included in their
metrics. Most modern thin clients provide features to restrict mobile data so that only a home
Wifi link is used to upload and download large data files. This practice is known as ‘data
offloading’ and it tends to hide the true costs of the wireless component of the access
network.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
B6(
Some studies have suggested that up to 50% of current network energy costs are due to the
wireless portion of the network [36]. This may indeed be true if we factor in home WiFi
routers and take into consideration the huge growth in new thin client devices. We also note
that some studies do not seek to make a distinction between the two network infrastructures
[17], [20], [126] although it is broadly recognized that the wireless access portion of the
network requires significantly more power [36], [45], [76], [77], [79].
The recent study from CEET [79] on wireless networks has addressed this issue and
provides estimates of the portion of wireless network energy consumption arising form home
WiFi. This is about 50% of the mobile access network in 2012 and should remain at about
that ratio to 2015.
54,-E6$5==202,406$*4D$I:5Z&C$A,1<(-B.$
From figure 11, page 16 of the VNI 2013 report [28] we note that by 2017 almost half of
mobile data will be carried by LTE/G4 networks. Thus LTE is envisaged to become the
access network of choice for many subscribers although many networks are still being rolled
out in a 2013/14 timeframe.
We next need to consider the energy per data unit for the wireless access infrastructure. It
is actually not easy to find a sensible value for this part of the network infrastructure. There
are studies that provide some insight but few that give practical data or measurements. The
most useful of these is the EARTH project funded by the European Union [130] which
contains the outcome of detailed modeling and practical studies on LTE networks and in table
11 of deliverable 2.4 [131] provides practical estimates of two different LTE network
scenarios – one with 20% of heavy users and an efficiency of 1.37 TWh/EB and a second
scenario with 50% of heavy users and an efficiency of 0.73 TWh/EB. These figures are
derived using the 2010 power model for LTE networks.
Other studies exist but not at the level of detail of the EARTH project. In fact there is a
very significant body of literature surrounding LTE and various aspects of power
consumption and efficiency relating to LTE. What is clear is that LTE can be efficient but it
needs to operate at very high throughput to achieve efficiency, whereas in more practical
cases where it is employed it is significantly under-utilized. In the EARTH study it is
mentioned that even with many heavy users on the system it operated on average with no
more than 10% of transport frames carrying data. Another study [132] presents the work of
the EARTH project in a more concise format.
A study by Huang et al [75] on working LTE networks in the US found significantly better
downlink and uplink data rates of 13 Mpbs and 6 Mbps respectively when compared with 3G
and Wifi networks. But, despite several new power saving improvements, LTE networks
were found to be as much as 23 times less power efficient compared with WiFi, and also less
power efficient than 3G. In particular the long high-power tail of LTE was found to be a key
contributor. In another study by CEET [79] the significance of the wireless access network is
again highlighted. These authors also find that LTE/G4 is about to become the main source of
electricity consumption in the access network and when combined with Wifi and other mobile
wireless networking technologies will grow much faster than the network component of data
centers.
H>B>A(9#3:$:.&'%(.8#(F&$#,#++(7--#++ (01 " 31 '# '.(
Our next challenge is to figure a sensible approach to achieve this separation. There are
multiple issues here as some of the wireless access consumption is, in our opinion, hidden by
home wireless installations. However it is better to analyze and quantify where we have
useful data so we adopted this approach.
To separate the wireless access component we used the data projections from CISCO [28]
for growth in mobile data. These show monthly estimates of mobile data growing from 0.9
EB in 2012 to 11.2 EB in 2017. Thus annual totals increase from 10.8 EB to 134 EB in 2017
B;(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
– an annual growth rate in excess of 70%[20]. If we remove this element from the core data
we can treat it based on what appear to be sensible efficiency estimates based on the literature
specializing on wireless and LTE/G4 networks.
To match our three scenarios – best/expected/worst cases – we used the higher value
estimate from the EARTH project of 1.37 TWh/EB as a starting point; for expected case the
lower EARTH value of 0.73 TWh/EB was employed and for best-case analysis we used a
lower starting point of 0.5 TWh/EB – these are shown in Table 13 below. An annual
improvement rate of 5% was applied starting with 2012 values. We used a ration of 1.46 for
the relative efficiency of 3G networks, deduced from [75]. Thus the energy usage of 3G is
estimated at c. 66% that of our LTE and mobile data is split between 3G and LTE as
estimated by [28]. We did not assume a higher improvement rate that 5% on the basis that
LTE/G4 networks are a new technology and it is to be expected that many large-scale
deployments will perform less efficiently than the ‘ideal’ cases studied by EARTH.
(
):@,#(6A_(Network Electricity Consumption per Exabyte (TWh/EB) estimates for
best, expected and worst case network growth scenarios.
One difference that will be noted by the observant reader is that our estimates for the core
network projections differ slightly from those in section 5.4.1 – this is because we decided to
align our three growth scenarios for the core network with a fixed growth percentage. This
provides an alternative metric to compare our estimates with other studies. We found in the
best case analysis that overall core network growth is likely to proceed at 7.5% - this may
seem higher than some studies but we have used the 30% year-on-year growth projections for
network data, combined with an optimistic 15% year-on-year improvement in efficiency. For
our expected growth case the match is to a 14% annual growth rate with our worst-case
analysis achieving a 20.5% growth rate with only a 5% annual efficiency improvement.
Projections for all three growth scenarios are presented in tables 14(a), (b) and (c) below.
The total annual mobile data projected for each year is shown in table 14(a) followed by the
projected electricity consumption of the core and wireless access networks. We note that
CEET [79] project a low value of 17.8 TWh/yr for 2015 which matches our best-case and
expected estimates; their worst-case is 25.5 TWh/yr which aligns nicely with our expected
growth scenario. It is worth commenting that in the early part of the 21st century many large
telcos had been quite successful in achieving annual improvements in energy efficiency of up
to 20% in core networks but this has slowed to a 10% improvement in recent years [78].
Perhaps the most notable aspect of all three scenarios is that wireless network contributions
will be more than doubled from their 2015 values by 2017. The overall contribution of
wireless access is still relatively small (<10%) compared to the core network expansion, but
based on our assumptions it is projected to grow at rates of 2-3 times that of the core.
):@,#(6BY:[_(Network Electricity Consumption per Exabyte (TWh/yr) estimates for
wireless and core from 2012-2017. Data in Exabytes is also shown. Best case.
Low Growth
Expected Growth
High Growth
LTE
0.5 TWh/EB
0.73 TWh/EB
1.37 TWh/EB
2012
2013
2014
2015
2016
2017
Non-LTE (EB)
9.3
15.7
25.9
40.0
56.8
73.9
LTE (EB)
1.5
3.5
7.7
16.4
32.0
60.5
Wireless
3.7
6.4
10.9
17.8
27.4
40.8
Core
350.3
376.5
404.8
435.1
467.8
502.8
Total
354.0
383.0
415.7
453.0
495.2
543.7
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
BA(
):@,#( 6BY@[_( Network Electricity Consumption per Exabyte (TWh/yr) estimates
broken into wireless and core from 2012-2017 – for expected case analysis.
2012
2013
2014
2015
2016
2017
Wireless
5.5
9.4
15.9
26.0
40.0
59.6
Core
348.5
397.3
453.0
516.4
588.7
671.1
Total
354.0
406.7
468.9
542.4
628.7
730.7
(
):@,#( 6BY-[_( Network Electricity Consumption per Exabyte (TWh/yr) estimates
broken into wireless and core from 2012-2017 – for worst-case analysis.
2012
2013
2014
2015
2016
2017
Wireless
10.2
17.6
29.8
48.7
74.9
111.6
Core
343.8
413.9
498.4
600.0
722.4
869.8
Total
354.0
431.5
528.2
648.7
797.4
981.4
N"N$G*1*$7,41,-$5',01-20216$7(4.89/12(4$
Data Centers have been discussed in some detail in sections 3.4 and 4.4. As was previously
commented this is the most difficult component to find useful data. Koomey’s approach [31]
leverages data on the shipment of servers, but even this is likely to be incomplete. Some
operators use custom builds of hardware and these may not appear under conventional server
shipments.
Nevertheless we have started from Koomey’s 2010 baseline. Our original approach was to
select separate growth rates from the 2010 baseline, but this led to different 2012 estimates.
Again, a key goal of our work is to have a common starting point from 2012 for each
category of electricity consumption. Thus, if some of our baseline estimates are shown to be
inaccurate it is very straightforward to adjust the model upwards, or downwards from 2012 as
appropriate. Bearing this in mind we decide to extrapolate from 2010 using our expected
scenario and use this to establish a common 2012 baseline. This is illustrated in Table 15
where we obtained a 2012 baseline of 281 TWh/yr.
From 2012 we then decided to use the fixed rates determined from our analysis of the core
network data to model data center growth. This is a relatively crude approach, but at this point
it is the best we have and it seems quite reasonable that data center growth would follow that
of the core network. Again we restate the assumption that network infrastructure within and
between data centers is part of core network - thus we are only considering the data
processing, storage and HVAC infrastructures of data centers.
Applying this approach to our baseline yields the projections provided in table 15 below.
):@,#(6H_(Network Electricity Consumption projections for Data Centers in (TWh/yr); based on
fixed growth rates determined from growth in core network capacity in section 5.4.3 above.
2010
2011
2012
2013
2014
2015
2016
2017
Low Growth (7.5%)
223.2
280.9
302.0
324.7
349.0
375.2
403.3
Expected Growth (14%)
223.2
254.4
280.9
320.3
365.1
416.2
474.5
540.9
High Growth (20%)
223.2
280.9
337.1
404.5
485.5
582.5
699.1
A first observation is that growth is not as aggressive as for networks. There is no
equivalent element of data center function that can be equated to the LTE/G4 access network.
Thus data centers are likely to evolve in a more straightforward way compared to the
networks that supply them. In fact data centers may benefit from a range of emerging
strategies such as strategic choice of geographic location to minimize HVAC costs; use of
GPU technology to mitigate the computational costs of video processing; economies of scale
as smaller data-centers coalesce into larger regional clusters. Note, however, that many
BB(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
economies within the data center require greater data mobility and will thus tend to drive
network traffic. The reality is that data center and network are becoming increasingly
entwined in a ‘back-end’ that is no longer transparent to the consumer. We will return to this
concept of the hidden back-end in our concluding discussion.
J>(<$1]#-.&1'+(41$()1.:,(0!M50)(!,#-.$&-&./(0 1 ' +2 "3.&1'((
Now we must first thank the reader for their patience as we worked through the various
parts of this complex puzzle. It has been quite a journey, but we can now put the pieces back
together and get a better understanding of the combined electricity consumption due to
today's digital lifestyle.
["#$7( 9)24,D$+-(L,012(4.$=(-$I(<S$5K/,01,D$*4D$W2E;$&-(<1;$
Q0,4*-2(.$
Here we present in table 16(a) the combined totals for the different growth scenarios of our
model. These are annual electricity consumption totals given in terrawatt-hours per annum
(TWh/yr). We note that if all ‘best case’ criteria are met there could in fact be an overall
reduction in electricity consumption generated by CE-ICT devices. Indeed some of the trends
that might be essential to this transition are already in evidence. We already commented on
the recent drop in sales of personal computers [41] [43] and it was also mentioned that TV
panel sales have been stagnant for the last few years [37], [38]. Both device categories are the
most energy-hungry of CE-ICT devices thus a shift in device sales to tablets and a reduction
in TV panel sales could help meet the reduced direct consumption targets of our best-case
scenario. Our best case scenario suggests that consumption can be constrained to less than a
very reasonable annual 2% growth rate - this is below the overall growth rate of world
electricity consumption and would be achieved through a combination of improved
operational efficiencies and new technologies.
):@,#( 6JY:[_( Projections of total electricity consumption in (TWh/yr); based on the
best/expected/worst case scenarios for devices (direct consumption and manufacturing energy),
networks and data centers, as outlined throughout section 5.
TWh/yr
2012
2013
2014
2015
2016
2017
Best Case
1,817
1,832
1,858
1,895
1,929
1,982
Expected
1,817
1,923
2,051
2,200
2,358
2,547
Worst Case
1,817
2,045
2,317
2,643
2,998
3,422
On the other hand our worst case scenario suggests that a near doubling of electricity
consumption over the next 5 years could also be plausible! This suggests an overall growth
rate in the region of 13% per annum, driven primarily by unconstrained expansion in the
networking and data center industries and a return to growth in consumer markets for CE-ICT
devices. Now while this latter scenario seems less probable given the current state of the
World economy it is interesting to note that global electricity consumption rebounded by 5%
in 2010 compared with a 1.9% decline in 2009, driven mainly by the developing economies,
particularly China [23].
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
BH(
Figure 3(a): Graph of electricity consumption in TWh/yr for best/expected/worst case scenarios.
Perhaps a more useful presentation is provided by table 16(b) which shows the percentage
of total global electricity that is attributable to CE-ICT ‘gadgets’ – here we have assumed a
3% annual growth in total electricity consumption, a figure that has been adopted by other
authors [20]. This is from a baseline of 23,192 TWh/yr in 2010, the most recent figures
available to us [23]. In this case our best case scenario shows the consumption of electricity
by CE-ICT decline as a proportion of global electricity consumption from 7.4% in 2012 to
6.9% in 2017. For our expected growth scenario the increase is to 8.9%, a significant rise but
not unexpected given the many disruptive technologies at work in the CE-ICT sector.
):@,#( 6JY@[_( Projections of the combined %age of total global electricity consumption in
(TWh/yr); based on the best/expected/worst case scenarios as outlined throughout section 5.
Total Electricity
21,039
21,670
22,320
22,990
23,680
24,390
Best Case
7.4%
7.2%
7.1%
7.0%
7.0%
6.9%
Expected
7.4%
7.6%
7.9%
8.2%
8.5%
8.9%
Worst Case
7.4%
8.1%
8.9%
9.8%
10.8%
12.0%
If however, things do not go to plan we have our outlier, but plausible, worst-case scenario.
Here we see demand for PCs & TV panels is revitalized by the demands of emerging
economies and the new middle classes of Brazil, India, Russia and China. In turn LTE/G4
networks create new demands for network services and introduces broadband Internet to large
urban population centers in Africa and Asia, but the communications industry fails to mitigate
the large electricity requirements of these networks in their quest for short-term profits. In
turn new data centers and core network backbones are hurriedly installed to handle the
accelerated global demand for network and cloud services. In turn electricity consumption
will jump by 40% in a five-year timeframe and the proportion that can be associated with CE-
ICT becomes 12% of global electricity.
5I@@@!
5I2@@!
3I@@@!
3I2@@!
:I@@@!
:I2@@!
3@53!
3@5:!
3@5-!
3@52!
3@59!
3@5B!
CJKL!6MKJ!
$NOJPLJQ!
'RSKL!6MKJ!
BJ(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
Figure 3(b): Percentage of global electricity consumption due to CE-ICT for best/expected/worst
case scenarios.
["!$7(9/*-*123,$%4*'6.2.$
At the beginning we outlined the scope of this work was to try and quantify the total
proportion of global electricity consumption that can be ascribed to the use of CE-ICT
devices. To this end we categorized 4 principle forms of energy that contribute to total energy
usage: (i) direct usage; (ii) manufacturing energy (LCA); (iii) network infrastructure; (iv) data
centers. It is instructive to consider the relative contribution of these components and, more
interestingly to observe how they are likely to change over the next five years.
To this end figure 4 shows the relative contributions for our 2012 baseline data in each
category. It is pretty clear from this figure that direct consumption by devices is just a shade
less than half of the total contribution at 47%. Each of the three remaining components are
nearly equal with data centers at 15%, manufacturing at 18% and networks at 20%.
Figure 4: Graph of the 4 main components of electricity consumption (2012 baseline).
Next in figure 5(a), (b) and (c) we take a look at the same data for 2017. What is
remarkable here is that there is very little difference between the ratios of these components
regardless of the growth scenario. The direct energy usage by devices ranges from 31-34%;
networks range form 26%-29% and data centers from 21%-25% with LCA lying in the range
of 16%-19%.
94@T!
B4@T!
=4@T!
>4@T!
5@4@T!
554@T!
534@T!
3@53!
3@5:!
3@5-!
3@52!
3@59!
3@5B!
CJKL!6MKJ!
$NOJPLJQ!6MKJ!
'RSKL!6MKJ!
8JUVPJKW!
-BT!
/JLXRSYKW!
3@T!
8MLM!
6JZLJSKW!
52T!
<6;W!5=T!
8JUVPJK!
/JLXRSYK!
8MLM!6JZLJSK!
<6;!
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
BO(
We can make a number of observations on this projected 2017 data:
• direct consumption by devices is less than 1/3 of electricity; compare with 1/2 in 2012.
• data centers + networks combined now represent 1/2 of electricity usage
• LCA remains approximately at the same level of contribution
So over the next 5 years we will see the combined contributions of networks & data centers
switch place with direct electricity usage. In other words our CE-ICT devices will consume
more electric power indirectly, outside the home than in the home itself. And this trend looks
set to continue for the foreseeable future. It is also interesting that this transition is
independent of the growth scenario.
Figure 5(a): Graph of the 4 main components of electricity consumption (2017 best-case scenario).
Figure 5(b): Graph of the 4 main components of consumption (2017 expected-case scenario).
8JUVPJKW!
:5T!
/JLXRSYKW!
39T!
8MLM!
6JZLJSKW!
32T!
<6;W!5=T!
8JUVPJK!
/JLXRSYK!
8MLM!6JZLJSK!
<6;!
8JUVPJKW!
:-T!
/JLXRSYKW!
3>T!
8MLM!
6JZLJSKW!
35T!
<6;W!59T!
8JUVPJK!
/JLXRSYK!
8MLM!6JZLJSK!
<6;!
BP(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
Figure 5(c): Graph of the 4 main components of consumption (2017 worst-case scenario).
["?$G2.08..2(4 $
Our study highlights some trends that have been commented on by earlier researchers.
The clearest trend in 2013 is the decline in the desktop computing market as growth in
sales of thin-client devices, tablets & smartphones, continues unabated. But this is not the
only trend that can be observed from our analysis.
To help with organizing our final discussion we have grouped it into three sections: firstly
we discuss some of the encouraging trends that suggest our best-case scenario may be viable;
secondly, we discuss emerging challenges - these are the factors that will push consumption
in the direction of our worst-case analysis; finally, we give some consideration to external
factors - the global economy, the potential impact of developing economies and global shifts
in demographics, social structures and individual behaviors all of which could strongly
influence how CE-ICT evolves over the next decade.
J>A>6(<1+&.&D#(5"3:-.+(14(!"#$%&'%()#-8'1,1%&#+(?(9/'#$%&#+(
:;,$54,-E6$5==202,41$:T$+*4,'$
We have already commented on this in section 4.1.3 noting that very significant
improvement has been made and that by 2014 practically all TV panels will consumer less
that 100W. This is important because TV panels are the single most significant source of
direct electricity consumption among CE-ICT appliances and industry efforts to improve the
underlying technology are to be commended.
The key variable here are the replacement rate of older TV panels, a variable that has not
been researched in the literature. This variable is, in turn, tied in with the lifecycle of TV
panels. In section 5.2.3 we took a look at the potential impact of different device lifecycles
and it is clear that this can be substantial. Many TV panels will continue to function beyond
their planned lifecycle and may be re-sold, or re-used within the household of the original
owners. Government policies and industry recycling initiatives can encourage a higher
replacement of older devices and there is a need for more studies in this area to determine
how best to optimize the transition to lower power appliances.
What we can say with some certainty is that the contribution of TV panels to electricity
consumption will enter a downward trend over the next decade and there is some scope to
accelerate the rate of this decline. At some point towards the end of the decade we should see
at least a 50% decline even if the installed base of TV panels continues to grow.
8JUVPJKW!
:3T!
/JLXRSYKW!
3>T!
8MLM!
6JZLJSKW!
3@T!
<6;W!5>T!
8JUVPJK!
/JLXRSYK!
8MLM!
6JZLJSK!
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
BV(
:;,$G,92.,$(=$1;,$G,.B1(/Y$
The market for desktop PCs has been stagnant for some time, primarily because of the
reduced costs and greater convenience of laptop computers but from early 2013 there has
been a pronounced decline in this market [43]. However we are not convinced that consumers
are substituting desktop computers with tablets because the use-modes are different; desktop
or laptop computers are still needed for 'work' in the home office, whereas tablets are
primarily entertainment devices for use on the couch [27].
A continuation of the transitioning to laptops is, however, likely to continue and the
desktop computer will need to reinvent itself if it is to survive the next decade. As in the case
of TV panels the transitioning from today's desktop to laptops is very likely to eventually lead
to at least a 50% decline in direct electricity consumption by home computing appliances over
the next decade, even if there is growth in the installed base of these devices. But within the
timeframe of the current study a growth in demand from developing economies will
counterbalance any declines in the developed world. Our best-case estimate sees PC sales
remaining stagnant, but with 5% annual improvement in efficiency reducing electricity
consumption by c. 25%.
A,<$A,1<(-B$%-0;21,018-,.$
One challenge we will discuss shortly is that of mobile access networks. Modern devices
almost invariably use a wireless connection to connect to the network. Next generation
mobile wireless access networks have much potential inefficiency built-in and could become
a very expensive 'power-hog'. On the other hand consumer are coming to expect ubiquitous
connectivity.
In response various researchers have suggested new hybrid wireless/optical approaches to
provide broadband access networks. A recent summary is given by Shi et al [133].
J>A>;(`!"#$%&'%`(08:,,#'%#+(:'*(.8#&$(<1.#'.&:,(5" 3 :- .+(
:;,$&-(<1;$(=$H()2',$%00,..$A,1<(-B.$
There remain, however, significant challenges in the networking and data center portions
of the energy consumption equation. Network operators across the world, from China to
Europe and the Americas are in the process of rolling out the next generation of mobile access
networks. Our reviews of the literature on LTE/G4, as this technology is commonly known,
suggest that the energy efficiency of these networks will be highly dependent on well thought
out deployments and the use of sophisticated load balancing/shifting policies. This is a largely
unknown quantity for the communications industry and if things do not go smoothly the
networking component of electricity consumption could become an industry-wide power hog.
However the industry is aware of this and has strong motivations to achieve successful
deployments of LTE/G4 as future profitability will depend on these.
This was discussed at some length in section 5.4.2 and despite many studies the
practicalities of these networks remain largely unknown. Two things are clear - firstly most
LTE/G4 networks will operate at much less than capacity and thus at poor overall efficiency;
secondly, the local network topology, traffic balancing between different cell sizes and
operational policies will determine the practical efficiencies of these networks rather than the
digital simulations that are used for most of today's estimations.
There is already a large body of literature concerned with the problems of optimizing these
networks. In this paper we simply identify their energy-efficient deployment and management
as a significant challenge for service providers over the nest 5 years. If successful these access
networks will bridge almost half of all global mobile data traffic onto the core network by
2017 [28].
For data centers we envisage less challenges as the industry is already working on
innovative modular solutions and low-power CPU technology is being driven strongly by the
HW(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
needs of the mobile device market. However it is important that improved sources of data are
made available. There is an admirable policy in the communications industry to make
available key metrics on energy efficiency of networks The data center industry needs to
introduce similar strategies in order to capture public confidence. Google is one of the more
forthcoming operators of data centers, but they represent a relatively small proportion of this
industry. In fairness it is still a young industry and will have significant challenges to face,
particularly in terms of public image [115], over the next few years.
7'(8D$7(9/8124E$
Originally a concept of the 1960s [25] cloud computing transitioned from a business-to-
business solution into a consumer service industry during 2009-2011 with the emergence of a
range of new services for consumers. These included data backup and storage from Dropbox,
Sugarsync, Amazon cloud drive, Skydrive and Google drive and services offering added
value sharing and management of personal content such as pictures (Flickr, Snapfish,
Photobucket) and video (youtube).
Now cloud-computing services drive demand for both network infrastructure and data
centers. There are arguments that some efficiencies are better implemented on a large scale
and cloud-computing will facilitate these - for example it makes sense to share one copy of a
song, or movie between 100,000 users rather than have each user store and access a personal
copy. On the other hand a cloud computing storage service encourages a user to upload and
share their personal content that might have stayed on a home computer. Consider a video of
an Xmas school play that is uploaded to youtube and then shared with 20 sets of parents,
generating many gigabytes of data traffic. And frankly, many of these new services are quite
compelling for consumers.
The challenge for the cloud-computing industry is not how to put the genie back in the
bottle - it is already too late for that! - but rather how to stop the genie growing too quickly.
:;,$>41,-4,1$(=$:;24E.$*4D$1;,$\0(44,01,D\$H20-(<*3,$(3,4$
Network-connected appliances are becoming more commonplace. Our CE-ICT devices are
mostly connected to the Internet, but even legacy TV sets can now be linked to a low-cost
"smart-TV" box that bridges your TV onto your home network. However the real
breakthrough into what is known as the "Internet of Things" (IoT) will come when our
kitchen appliances start to become 'connected' [134]. Expert projections indicate that the
number of networked appliances could reach 50-100 billion over the next 5 to 10 years [135],
[136].
However the concept of network-connected appliances is not exactly new [137–140] and
there are anecdotal discussions about the networked refrigerator every time a new 'smart-
home' concept is unveiled to the public, but this deluge of devices has yet to manifest itself. It
is true that given the low cost of modern wireless connectivity there has never been a better
time for the IoT to gain momentum but the authors remain a little skeptical of the benefits,
except in some limited examples. Thus we record IoT as a potential emerging challenge, but a
'potential' one that has not made significant progress in the last 20 years.
["C$7(40'8D24E$@,9*-B.$
As noted in our introduction the CE-ICT industry is experiencing a radical and global shift
brought on by a confluence of disruptive technologies. Cloud computing, high-speed wireless
access networks and rapidly evolving thin client technology are emerging to meet changing
consumer demands and new usage patterns for CE-ICT. In turn we are seeing the emergence
of a wealth of new ICT products and services together with new patterns of consumption and
socio-economic behaviors and an evolving network infrastructure to support these changes.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
H6(
A key goal of this work has been to outline a framework for evaluating future electricity
growth patterns and determine various factors and emerging trends that can influence the
future evolution of CE-ICT.
Thus, while we have presented a broad range of possible future outcomes, these are
supported with quite detailed sets of data and assumptions and compared with equivalent
work from other researchers to provide, where practical, a consensus. It should thus be
possible to review our work and assumptions retrospectively and identify where our approach
has aligned usefully with future growth patterns and where it has not.
The three growth scenarios used to project 2013-2017 data represent a very wide range of
possibilities, but the goal of this work is not so much to provide an accurate future prediction
- in reality there are too many unknowns. Instead we have tried to consider some alternative
scenarios and illustrate how the future may unfold. And while our hopes lie with an overall
decline in electricity consumption on the lines of our best-case scenario, the alternative worst-
case scenario shows annual compound growth rates in electricity usage of more than 12% for
the next 5 years. This can serve as a motivation for all of us in different industry sectors to
continue to focus research efforts to improve efficiency and reduce the impact of new
technologies across the sector.
S#4#$#'-#+
[1] G. Shapiro, “America’s Comeback Starts with American Innovators [Soapbox],” IEEE Consumer
Electronics Magazine, vol. 1, no. 1, pp. 19–24, Jan. 2012.
[2] H. Mouftah and B. Kantarci, “Energy-efficient cloud computing–A green migration of traditional IT,” in
Handbook of Green Communications,, 2012, pp. 295–329.
[3] S. Srikantaiah, A. Kansal, and F. Zhao, “Energy aware consolidation for cloud computing,” in Proc
USENIX workshop on power aware computing and systems in conjunction with OSDI, 2008, pp. 1–5.
[4] A. Berl, E. Gelenbe, and M. Di Girolamo, “Energy-efficient cloud computing,” The Computer Journal,
vol. 53, no. 7, pp. 1045–1051, 2010.
[5] A. Greenberg and J. Hamilton, “The cost of a cloud: research problems in data center networks,” ACM
SIGCOMM Computer Communication Review, vol. 39, no. 1, pp. 68–73, 2008.
[6] D. Mcqueen, “The momentum behind LTE adoption [sGPP LTE,” IEEE Communications Magazine, vol.
47, no. 2, pp. 44–45, Feb. 2009.
[7] M. A. Marsan and M. Meo, “Energy efficient wireless Internet access with cooperative cellular
networks,” Computer Networks, vol. 55, no. 2, pp. 386–398, 2011.
[8] P. M. Corcoran and J. Desbonnet, “Wireless home network infrastructure for wearable appliances,” in
2002 Digest of Technical Papers. International Conference on Consumer Electronics, 2002, pp. 104–105.
[9] P. M. Corcoran, J. Desbonnet, P. Bigioi, and I. Lupu, “Home network infrastructure for
handheld/wearable appliances,” Consumer Electronics, IEEE Transactions on, vol. 48, no. 3, pp. 490–
495, 2002.
[10] D. Maga, M. Hiebel, and C. Knermann, “Comparison of two ICT solutions: desktop PC versus thin client
computing,” The International Journal of Life Cycle Assessment, pp. 1–11, 2012.
[11] P. Kudtarkar, T. DeLuca, V. Fusaro, P. Tonellato, and D. Wall, “Cost-effective cloud computing: a case
study using the comparative genomics tool, roundup,” Evolutionary Bioinformatics Online, vol. 6, p. 197,
2010.
[12] M. Mihailescu and Y. Teo, “Dynamic resource pricing on federated clouds,” in Cluster, Cloud and Grid
Computing ( CCGrid), 2010 10th IEEE/ACM International Conference on, 2010.
[13] R. Buyya, A. Beloglazov, and J. Abawajy, “Energy-efficient management of data center resources for
cloud computing: A vision, architectural elements, and open challenges,” arXiv preprint, 2010. [Online].
Available: http://arxiv.org/abs/1006.0308. [Accessed: 22-May-2013].
[14] A. Beloglazov and R. Buyya, “Energy efficient resource management in virtualized cloud data centers,”
in Proceedings of the 2010 10th IEEE/ACM International Conference on Cluster, Cloud and Grid
Computing, 2010, pp. 826–831.
[15] J. Baliga and R. Ayre, “Green cloud computing: Balancing energy in processing, storage, and transport,”
PProceedings of the IEEE,, vol. 99, no. 1, pp. 149–167, 2011.
[16] P. M. Corcoran, “Cloud Computing and Consumer Electronics: A Perfect Match or a Hidden Storm?,”
IEEE Consumer Electronics Magazine, vol. 1, no. 2, pp. 14–19, Apr. 2012.
H;(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
[17] M. Pickavet, W. Vereecken, S. Demeyer, P. Audenaert, B. Vermeulen, C. Develder, D. Colle, B. Dhoedt,
and P. Demeester, “Worldwide energy needs for ICT: The rise of power-aware networking,” in 2008 2nd
International Symposium on Advanced Networks and Telecommunication Systems, 2008, pp. 1–3.
[18] K. Hinton, J. Baliga, and M. Feng, “Power consumption and energy efficiency in the internet,” Network,
…, 2011.
[19] J. Baliga, K. Hinton, R. Ayre, and R. S. Tucker, “Carbon footprint of the Internet,” Telecommunications
Journal of Australia, vol. 59, no. 1, pp. 5.1–5.14, Feb. 2009.
[20] S. Lambert and W. Van Heddeghem, “Worldwide electricity consumption of communication networks,”
Optics Express, vol. 20, no. 26, pp. B513–B524, 2012.
[21] B. Global, “BP statistical review of world energy,” 2010-12-10)[2011-02-12]. http://www, bp. conv/ …,
2010.
[22] “Electricity in a Climate-Constrained World - Books - OECD iLibrary.” [Online]. Available:
http://www.oecd-ilibrary.org/energy/climate-and-electricity-annual-2012_9789264175556-en. [Accessed:
28-Apr-2013].
[23] P. Benoit and R. Baron, “Introduction,” in Electricity in a Climate Constrained World- Data and
Analysis, IEA, 2013, pp. 7–11.
[24] “IEA - Publication:- Gadgets and Gigawatts: Policies for Energy Efficient Electronics.” [Online].
Available: http://www.iea.org/publications/freepublications/publication/name,3807,en.html. [Accessed:
28-Apr-2013].
[25] D. Parkhill, The Challenge of The Computer Utility. Addison-Wesley Publishing Company., 1966.
[26] “Gartner Says Worldwide Mobile Phone Sales Declined 1.7 Percent in 2012,” Gartner Newsroom.
[Online]. Available: http://www.gartner.com/newsroom/id/2335616. [Accessed: 11-Jun-2013].
[27] “A Bitter Pill or a Better Tablet? - IEEE - The Institute, PETER CORCORAN 21 October 2011.”
[Online]. Available: http://theinstitute.ieee.org/technology-focus/technology-history/a-bitter-pill-or-a-
better-tablet. [Accessed: 22-May-2013].
[28] “Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2012–2017
[Visual Networking Index (VNI)].” [Online]. Available:
http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11-
520862.html. [Accessed: 28-Apr-2013].
[29] “Cisco Global Cloud Index: Forecast and Methodology, 2011–2016 [Global Cloud Index (GCI)].”
[Online]. Available:
http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns1175/Cloud_Index_White_
Paper.html. [Accessed: 28-Apr-2013].
[30] J. Koomey, “Worldwide electricity used in data centers,” Environmental Research Letters, vol. 3, no.
3:034008, pp. 1–8, 2008.
[31] J. Koomey, “Growth in data center electricity use 2005 to 2010,” Oakland, CA: Analytics Press., 2011.
[32] S. Garimella, T. Persoons, J. Weibel, and L. Yeh, “Technological drivers in data centers and telecom
systems: Multiscale thermal, electrical, and energy management,” Applied Energy, vol. 107, pp. 66–80,
2013.
[33] J. Koomey, “Estimating total power consumption by servers in the US and the world,” 2007. [Online].
Available: http://hightech.lbl.gov/documents/DATA_CENTERS/svrpwrusecompletefinal.pdf. [Accessed:
27-Apr-2013].
[34] A. S. G. Andrae and O. Andersen, “Life cycle assessments of consumer electronics — are they
consistent?,” The International Journal of Life Cycle Assessment, vol. 15, no. 8, pp. 827–836, Jul. 2010.
[35] B. Urban, V. Tiefenbeck, and K. Roth, “Energy consumption of consumer electronics in US homes in
2010,” 2011.
[36] “Global Sustainability Initiative, SMARTer 2020 Report: The Role of ICT in Driving a Sustainable
Future,” via Web, 2012. [Online]. Available:
http://gesi.org/assets/js/lib/tinymce/jscripts/tiny_mce/plugins/ajaxfilemanager/uploaded/SMARTer 2020 -
The Role of ICT in Driving a Sustainable Future - December 2012._2.pdf. [Accessed: 05-Jun-2013].
[37] Displaysearch.com, “Global TV Demand Expected to be Flat in 2013, According to NPD DisplaySearch -
DisplaySearch.” [Online]. Available:
http://www.displaysearch.com/cps/rde/xchg/displaysearch/hs.xsl/121023_global_tv_demand_expected_to
_be_flat_in_2013.asp. [Accessed: 28-Apr-2013].
[38] Displaysearch.com, “Global LCD TV shipments fall for the first time in 2012; outlook cautious for 2013,
says DisplaySearch,” Web, 2013. [Online]. Available:
http://www.displaysearch.com/cps/rde/xchg/displaysearch/hs.xsl/130321_global_lcd_tv_shipments_fall_f
or_the_first_time_in_2012.asp. [Accessed: 11-May-2013].
[39] V. Upadhyayula, “Screening Level Cradle to Grave Life Cycle Assessment of a Conceptual Carbon
Nanotube Field Emission Display Device.,” nsti.org. [Online]. Available:
http://www.nsti.org/procs/Nanotech2012v3/10/T7.194. [Accessed: 11-May-2013].
[40] “iPad 3 4G Teardown - Page 3 - iFixit.” [Online]. Available:
http://www.ifixit.com/Teardown/iPad+3+4G+Teardown/8277/3. [Accessed: 05-Jun-2013].
[41] “IDC worldwide quarterly PC tracker,” 2013. [Online]. Available:
http://www.idc.com/tracker/showproductinfo.jsp?prod_id=1. [Accessed: 29-May-2013].
[42] P. Teehan and M. Kandlikar, “Sources of variation in life cycle assessments of desktop computers,”
Journal of Industrial Ecology, vol. 16, no. s1, pp. s182–s194, 2012.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
HA(
[43] “The personal computer slump | The Economist.” [Online]. Available:
http://www.economist.com/node/454922. [Accessed: 07-May-2013].
[44] E. G. Hertwich and C. Roux, “Greenhouse gas emissions from the consumption of electric and electronic
equipment by Norwegian households.,” Environmental science & technology, vol. 45, no. 19, pp. 8190–6,
Oct. 2011.
[45] B. Sikdar, “A study of the environmental impact of wired and wireless local area network access,”
Consumer Electronics, IEEE Transactions on, 2013.
[46] K. De Decker, “The monster footprint of digital technology,” Low-Tech Magazine, 2009. [Online].
Available: http://www.lowtechmagazine.com/2009/06/embodied-energy-of-digital-technology.html.
[Accessed: 22-May-2013].
[47] F. Wabwoba, “Barriers to Implementation of Green ICT in Kenya,” International Journal of Science and
Technology, vol. 2, no. 12, 2012.
[48] S. Cayford and E. Wilson, “Waste 2.0,” 2009. [Online]. Available:
http://www.hhh.umn.edu/centers/stpp/pdf/paper_2009-05-22.b.pdf. [Accessed: 05-Jun-2013].
[49] A. Andrae and O. Andersen, “Life cycle assessment of integrated circuit packaging technologies,” The
International Journal of Life Cycle Assessment, vol. 16, no. 3, pp. 258–267, 2011.
[50] “ETSI TS 103 199 V1.1.1!: Environmental Engineering (EE); Life Cycle Assessment (LCA) of ICT
equipment, networks and services; General methodology and common requirements.” European
Telecommunications Standards Institute.
[51] K. Wiens and P. Corcoran, “Repairability Smackdown: How Do the Latest Tablet Models Stack Up?,”
Consumer Electronics Magazine, IEEE, vol. 2, no. 1, pp. 42–50, 2013.
[52] R. Hischier and I. Baudin, “LCA study of a plasma television device,” The International Journal of Life
Cycle Assessment, vol. 15, no. 5, pp. 428–438, 2010.
[53] “Ecoinvent Database: Desktop Computer, without screen, at plant/GLO U,” 2012.
[54] “Ecoinvent database: LCD flat screen, 17 inches, at plant/GLO S,” 2012.
[55] “iPhone 4S Environmental Report.” [Online]. Available:
http://images.apple.com/euro/environment/reports/docs/iPhone4S_product_environmental_report_sept20
12.pdf. [Accessed: 17-Jun-2013].
[56] P. Tuzzolino and T. Tanaka, “Involvement of France Telecom Group for a Sustainable Future,” in Design
for Innovative Value Towards a Sustainable Society, Springer, 2012, pp. 732–736.
[57] “France Telecom. WWF-Orange Methodology Mobiles Version 3.” 2012.
[58] “iPad mini Environmental Report.” [Online]. Available:
http://images.apple.com/euro/environment/reports/docs/iPadmini_PER_oct2012.pdf. [Accessed: 29-May-
2013].
[59] “Sustainability Communication 2008: Thomson Technicolor,” 2008. [Online]. Available:
http://www.technicolor.com/uploads/associated_materials/sustainability_2008_4d88cbf24153161607145
5.pdf. [Accessed: 17-Jun-2013].
[60] Q. Yang, “Life cycle assessment in sustainable product design,” SIMTech technical reports, Volume 8
Number 1, 2007. [Online]. Available: http://www.simtech.a-star.edu.sg/TechnicalReports/tech-
reportV8N1/STR_V8_N1_11_LCE.pdf. [Accessed: 29-May-2013].
[61] A. (Huawei) Andrae, “Comparative Micro Life Cycle Assessment of Physical and Virtual Desktops in a
Cloud Computing Network with Consequential, Efficiency, and Rebound Considerations,” Journal of
Green Engineering, vol. 3, no. 2, pp. 193–218., 2013.
[62] P. Teehan and M. Kandlikar, “Comparing embodied greenhouse gas emissions of modern computing and
electronics products,” Environmental Science & Technology, vol. 47, no. 9, pp. 3997–4003, 2013.
[63] A. Andrae, “European LCA Standardisation of ICT: Equipment, Networks, and Services,” in Towards
Life Cycle Sustainability Management, Springer Netherlands, 2011, pp. 483–493.
[64] “Global Sustainability Initiative: GHG Protocol Product Life Cycle Accounting 8 and Reporting Standard
9 ICT Sector Guidance,” 2011. [Online]. Available:
http://www.ghgprotocol.org/files/ghgp/public/Telecommunications Network Services guide.pdf.
[Accessed: 05-Jun-2013].
[65] Dell’Oro Group., “Mobile RAN Report Five Year Forecast 2013 – 2017.,” vol. 12, no. 1, 2013.
[66] C. Goldey and E. Kuester, “Lifecycle assessment of the environmental benefits of remanufactured
telecommunications product within a ‘green’ supply chain,” in Sustainable Systems and Technology
(ISSST), 2010 IEEE International Symposium on, 2010, pp. 1–6.
[67] J. Gozalvez, “Wireless Connections Surpass 6 Billion Mark,” Vehicular Technology Magazine, IEEE,
vol. 7, no. 4, pp. 13–17, 2012.
[68] C. Honée and D. Hedin, “Environmental Performance of Data Centres: A Case Study of the Swedish
National Insurance Administration: EGG2012, Berlin,” in Electronics Goes Green 2012+, ECG 2012-
Joint International Conference and Exhibition, Proceedings, 2012.
[69] “Gartner Says Worldwide Server Shipments Declined 0.2 Percent; Revenue Increased 5.1 Percent in
Fourth Quarter of 2012.” [Online]. Available: http://www.gartner.com/newsroom/id/2351518. [Accessed:
21-Jun-2013].
[70] M. StutZ, S. O’Connell, and J. Pflueger, “Carbon Footprint of a Dell Rack Server,” in Electronics Goes
Green 2012+, ECG 2012-Joint International Conference and Exhibition, Proceedings, 2012.
[71] K. Kumar and Y. Lu, “Cloud computing for mobile users: Can offloading computation save energy?,”
Computer, 2010.
HB(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
[72] A. Beloglazov, J. Abawajy, and R. Buyya, “Energy-aware resource allocation heuristics for efficient
management of data centers for cloud computing,” Future Generation Computer Systems, vol. 28, no. 5,
pp. 755–768, 2012.
[73] Q. Zhang, L. Cheng, and R. Boutaba, “Cloud computing: state-of-the-art and research challenges,”
Journal of Internet Services and Applications, vol. 1, no. 1, pp. 7–18, 2010.
[74] P. Buchanan, V. Yampolsky, and B. Buchanan, “Comparison of the Power Consumption and Carbon
Footprint of a Cloud Infrastructure against Standard Desktops.” [Online]. Available:
http://www.buchananweb.co.uk/cloud.pdf. [Accessed: 21-Jun-2013].
[75] J. Huang, F. Qian, A. Gerber, and Z. Mao, “A close examination of performance and power
characteristics of 4G LTE networks,” in Proceedings of the 10th international conference on Mobile
systems, applications, and services, 2012, pp. 225–238.
[76] J. Baliga and R. Ayre, “Energy consumption in wired and wireless access networks,” Communications
Magazine, IEEE, vol. 49, no. 6, pp. 70–77, 2011.
[77] M. Deruyck, W. Vereecken, E. Tanghe, W. Joseph, M. Pickavet, L. Martens, and P. Demeester, “Power
consumption in wireless access network,” in 2010 European Wireless Conference (EW), 2010, pp. 924–
931.
[78] D. C. Kilper, G. Atkinson, S. K. Korotky, S. Goyal, P. Vetter, D. Suvakovic, and O. Blume, “Power
Trends in Communication Networks,” IEEE Journal of Selected Topics in Quantum Electronics, vol. 17,
no. 2, pp. 275–284, Mar. 2011.
[79] “CEET Whitepaper:The Power of Wireless Cloud.” [Online]. Available:
http://www.ceet.unimelb.edu.au/pdfs/ceet_white_paper_wireless_cloud.pdf. [Accessed: 22-May-2013].
[80] W. Vereecken, W. Heddeghem, M. Deruyck, B. Puype, B. Lannoo, W. Joseph, D. Colle, L. Martens, and
P. Demeester, “Power consumption in telecommunication networks: overview and reduction strategies,”
IEEE Communications Magazine, vol. 49, no. 6, pp. 62–69, Jun. 2011.
[81] M. Deruyck, E. Tanghe, W. Joseph, and L. Martens, “Modelling and optimization of power consumption
in wireless access networks,” Computer Communications, vol. 34, no. 17, pp. 2036–2042, 2011.
[82] “Global LTE Traffic will Grow by Over 200% in 2013.” [Online]. Available:
http://www.abiresearch.com/press/global-lte-traffic-will-grow-by-over-200-in-2013. [Accessed: 29-Apr-
2013].
[83] “China’s Going To Build Massive 4G Mobile Networks—Snubbing European Companies In The
Process.” [Online]. Available: http://www.businessinsider.com/chinas-to-start-building-4g-networks-
2013-4. [Accessed: 29-Apr-2013].
[84] J. Zhang and G. D. la Roche, Femtocells: technologies and deployment. Wiley, New York, 2010, pp. 1–
13.
[85] H. Claussen, L. Ho, and L. Samuel, “Self-optimization of coverage for femtocell deployments,” in
Wireless Telecommunications Symposium, 2008. WTS 2008, 2008, pp. 278–285.
[86] D. Knisely, “Standardization of femtocells in 3GPP,” Communications Magazine, IEEE, vol. 47, no. 9,
pp. 68–75, 2009.
[87] L. Wang, Y. Zhang, and Z. Wei, “Mobility management schemes at radio network layer for LTE
femtocells,” in Vehicular Technology Conference, 2009. VTC Spring 2009. IEEE 69th, 2009, pp. 1–9.
[88] G. Gur, S. Bayhan, and F. Alagöz, “Cognitive femtocell networks: an overlay architecture for localized
dynamic spectrum access,” Wireless Communications, IEEE, vol. 17, no. 4, pp. 62–70, 2010.
[89] Z. Lu, T. Bansal, and P. Sinha, “Achieving User-Level Fairness in Open-Access Femtocell based
Architecture,” Mobile Computing, IEEE Transactions on, no. 99, 2012.
[90] B. Kaufman, J. Lilleberg, and B. Aazhang, “Femtocell Architectures with Spectrum Sharing for Cellular
Radio Networks,” arXiv preprint, 2013. [Online]. Available: http://arxiv.org/abs/1301.7245. [Accessed:
12-Jun-2013].
[91] A. Fehske, G. Fettweis, J. Malmodin, and G. Biczok, “The global footprint of mobile communications:
The ecological and economic perspective,” IEEE Communications Magazine, vol. 49, no. 8, pp. 55–62,
Aug. 2011.
[92] S. Aleksic, “Analysis of power consumption in future high-capacity network nodes,” Optical
Communications and Networking, IEEE/OSA Journal of, vol. 1, no. 3, pp. 245–258, 2009.
[93] A. Tzanakaki and K. Katrinis, “Dimensioning the future Pan-European optical network with energy
efficiency considerations,” Optical Communications and Networking, IEEE/OSA Journal of,, vol. 3, no.
4, pp. 272–280, 2011.
[94] Y. Zhang and P. Chowdhury, “Energy efficiency in telecom optical networks,” Communications Surveys
& Tutorials, IEEE,, vol. 12, no. 4, pp. 441–458, 2010.
[95] J. Baliga, R. Ayre, and K. Hinton, “Energy consumption in optical IP networks,” Journal of Lightwave
Technology, vol. 2713, no. 2391–2403, 2009.
[96] W. Van Heddeghem and F. Idzikowski, “Power consumption modeling in optical multilayer networks,”
Photonic Network Communications, vol. 24, no. 2, pp. 86–102, 2012.
[97] M. Murakami and K. Oda, “Power consumption analysis of optical cross-connect equipment for future
large capacity optical networks,” in Transparent Optical Networks, 2009. ICTON’09. 11th International
Conference on, 2009, pp. 1–4.
[98] Y. Wu and L. Chiaraviglio, “Power-aware routing and wavelength assignment in optical networks,” in
Optical Communication, 2009. ECOC’09. 35th European Conference on, 2009, pp. 1–2.
!"#$%&'%()$#'*+(&'(!,#-.$&-&./(01'+2"3.&1'(41$(01'+2"#$(50)(
HH(
[99] M. Deruyck, W. Vereecken, E. Tanghe, W. Joseph, M. Pickavet, L. Martens, and P. Demeester,
“Comparison of power consumption of mobile WiMAX, HSPA and LTE access networks,” in
Telecommunications Internet and Media Techno Economics (CTTE), 2010 9th Conference on, 2010, pp.
1–7.
[100] C. Lange, D. Kosiankowski, R. Weidmann, and A. Gladisch, “Energy Consumption of
Telecommunication Networks and Related Improvement Options,” IEEE Journal of Selected Topics in
Quantum Electronics, vol. 17, no. 2, pp. 285–295, Mar. 2011.
[101] L. Correia, D. Zeller, O. Blume, D. Ferling, Y. Jading, I. Gódor, G. Auer, and L. Van Der Perre,
“Challenges and enabling technologies for energy aware mobile radio networks,” Communications
Magazine, IEEE, vol. 48, no. 11, pp. 66–72, 2010.
[102] C. Doctorow, “Big data: welcome to the petacentre,” Nature, vol. 455, no. 7209, pp. 16–21, 2008.
[103] R. Buyya, C. Yeo, S. Venugopal, J. Brobeerg, and I. Brandic, “Cloud computing and emerging IT
platforms: Vision, hype, and reality for delivering computing as the 5th utility,” Future Generation
computer systems, 25(6), 599-616., vol. 25, no. 6, pp. 599–616, 2009.
[104] L. Barroso, J. Dean, and U. Holzle, “Web search for a planet: The Google cluster architecture,” Micro,
IEEE, vol. 23, no. 2, pp. 22–28, 2003.
[105] K. Le and R. Bianchini, “Cost-and energy-aware load distribution across data centers,” Proceedings of
HotPower, 2009. [Online]. Available:
http://seelab.ucsd.edu/virtualefficiency/related_papers/27_hotpower09.pdf. [Accessed: 30-Apr-2013].
[106] E. Carrera, E. Pinheiro, and R. Bianchini, “Conserving disk energy in network servers,” in Proceedings of
the 17th annual international conference on Supercomputing 86-97, 2003, pp. 86–97.
[107] M. Poess and R. Nambiar, “Energy cost, the key challenge of today’s data centers: a power consumption
analysis of TPC-C results,” Proceedings of the VLDB Endowment, vol. 1, no. 2, pp. 1229–1240, 2008.
[108] J. Koomey, C. Belady, and M. Patterson, “Assessing trends over time in performance, costs, and energy
use for servers,” Lawrence Berkeley National Laboratory, Stanford University, Microsoft Corporation,
and Intel Corporation, Tech. Rep., 2009. [Online]. Available:
http://www3.intel.com/assets/pdf/general/servertrendsreleasecomplete-v25.pdf. [Accessed: 30-Apr-2013].
[109] M. Armbrust, A. Fox, and R. Griffith, “A view of cloud computing,” Communications of the ACM, vol.
53, no. 4, pp. 50–58, 2010.
[110] W. Park, A. Phadke, N. Shah, and V. Letschert, “Efficiency improvement opportunities in TVs:
Implications for market transformation programs,” Energy Policy, 2013.
[111] “PR Web. Blu-ray Players to Reach 105 Million Units Shipped in 2015, Says In-Stat.,” 2013. [Online].
Available: http://www.prweb.com/releases/in-stat_npd_group/blu-ray_players/prweb8824799.htm.
[Accessed: 18-Jun-2013].
[112] IDC, “Smartphones Expected to Outship Feature Phones for First Time in 2013,” prUS23982813, 2013.
[Online]. Available: http://www.idc.com/getdoc.jsp?containerId=prUS23982813. [Accessed: 07-May-
2013].
[113] “Connected World: Global set top box shipments to reach 228 million units in 2012,” 2012. [Online].
Available: http://www.connectedworld.tv/articles/global-set-top-box-shipments-to-reach-228-million-
units-in-2012/7617/. [Accessed: 18-Jun-2013].
[114] IDC, “IDC Anticipates a Video Game Console Rebound As New Platforms Arrive.” [Online]. Available:
http://www.idc.com/getdoc.jsp?containerId=prUS23901313. [Accessed: 18-Jun-2013].
[115] “How Clean is Your Cloud? | Greenpeace International.” [Online]. Available:
http://www.greenpeace.org/international/en/publications/Campaign-reports/Climate-Reports/How-Clean-
is-Your-Cloud/. [Accessed: 28-May-2013].
[116] Y. Jin, Y. Wen, Q. Chen, and Z. Zhu, “An Empirical Investigation of the Impact of Server Virtualization
on Energy Efficiency for Green Data Center,” The Computer Journal, 2013. [Online]. Available:
http://comjnl.oxfordjournals.org/content/early/2013/02/20/comjnl.bxt017.full.pdf+html. [Accessed: 28-
Apr-2013].
[117] A. Beloglazov and R. Buyya, “Energy efficient allocation of virtual machines in cloud data centers,” in
Cluster, Cloud and Grid Computing (CCGrid), 2010 10th IEEE/ACM International Conference on, 2010,
pp. 577–578.
[118] G. Koutitas, “Challenges for energy efficiency in local and regional data centers,” 2010. [Online].
Available: http://researchwebshelf.com/uploads/104_article1_1.pdf. [Accessed: 28-Apr-2013].
[119] M. Lin and A. Wierman, “Dynamic right-sizing for power-proportional data centers,” in INFOCOM,
2011 Proceedings IEEE, 2011, pp. 1098–1106.
[120] F. Ahmad and T. Vijaykumar, “Joint optimization of idle and cooling power in data centers while
maintaining response time,” ACM Sigplan Notices, vol. 45, no. 3, pp. 243–256, 2010.
[121] B. Chun and G. Iannaccone, “An energy case for hybrid datacenters,” ACM SIGOPS Operating Systems
Review, vol. 44, no. 1, pp. 76–80, 2010.
[122] A. Szalay, G. Bell, and H. Huang, “Low-power amdahl-balanced blades for data intensive computing,”
ACM SIGOPS Operating Systems Review, vol. 44, no. 1, pp. 71–75, 2010.
[123] J. Enos and C. Steffen, “Quantifying the impact of GPUs on performance and energy efficiency in HPC
clusters,” in Green Computing Conference, 2010 International, 2010, pp. 317–324.
[124] L. Johnsson, “Efficiency, Energy Efficiency and Programming of Accelerated HPC Servers: Highlights of
PRACE Studies,” in GPU Solutions to Multi-scale Problems in Science and Engineering, Springer Berlin
Heidelberg, 2013, pp. 33–78.
HJ(
<#.#$(=>(01$-1$:'(?(7'*#$+(7'*$:#(
[125] V. Coroama, L. Hilty, E. Heiri, and F. Horn, “The Direct Energy Demand of Internet Data Flows,”
Journal of Industrial Ecology, 2013. [Online]. Available:
http://www.google.com/events/howgreenistheinternet2013/pdfs/DirectEnergyDemand.pdf. [Accessed:
12-Jun-2013].
[126] “Corporate Responsibility Report 2010/11, Verizon Communications,” 2011. [Online]. Available:
http://responsibility.verizon.com/assets/docs/verizon_cr_report_2010-2011.pdf. [Accessed: 12-Jun-2013].
[127] F. Cucchietti, G. Griffa, and L. Radice, “Eco-efficiency indicator: an operator’s energy performance
indicator,” in Telecommunications Energy Conference, 2007. INTELEC 2007. 29th International, 2007,
pp. 743–748.
[128] “Eco-efficiency Indicator for Telecom_Italia_S.P.A. (TI).” [Online]. Available:
http://www.wikinvest.com/stock/Telecom_Italia_S.P.A._(TI)/Eco-efficiency_Indicator. [Accessed: 13-
Jun-2013].
[129] “Sustainability Report | Telecom Italia Group,” 2012. [Online]. Available:
http://www.telecomitalia.com/tit/en/sustainability/sustainability-report.html. [Accessed: 13-Jun-2013].
[130] “EARTH D4.3 - Final Report on Green Radio Technologies.” [Online]. Available: https://bscw.ict-
earth.eu/pub/bscw.cgi/d70472/EARTH_WP4_D4.3.pdf. [Accessed: 28-Apr-2013].
[131] “EARTH Deliverable D2.3 - Energy efficiency analysis of the reference systems, areas of improvements
and target breakdown.” [Online]. Available: https://bscw.ict-
earth.eu/pub/bscw.cgi/d71252/EARTH_WP2_D2.3_v2.pdf. [Accessed: 28-Apr-2013].
[132] G. Auer, V. Giannini, and C. Desset, “How much energy is needed to run a wireless network?,” Wireless
Communications, IEEE Transactions on, vol. 18, no. 5, pp. 40–49, 2011.
[133] L. Shi, “Saving energy in long-reach broadband access networks: architectural approaches,”
Communications Magazine, IEEE, vol. 51, no. 2, pp. S16–S21, 2013.
[134] V. Rozite, “How can we make an Internet-surfing microwave oven go to ‘sleep’?,” in Electricity in a
Climate Constrained World- Data and Analysis, IEA, 2013, pp. 20–24.
[135] C. Valhouli, “The Internet of things: Networked objects and smart devices,” 2010.
[136] L. Atzori, A. Iera, and G. Morabito, “The internet of things: A survey,” Computer Networks, vol. 54, no.
15, pp. 2787–2805, 2010.
[137] P. Corcoran, “Mapping home-network appliances to TCP/IP sockets using a three-tiered home gateway
architecture,” Consumer Electronics, IEEE Transactions on, vol. 44, no. 3, pp. 729–736, 1998.
[138] P. M. Corcoran, “A Three-tier, TCP/IP Gateway Architecture To Link Home Networks With The
Internet.,” in Consumer Electronics (ICCE98), IEEE International Conference on, 1998, pp. 242–243.
[139] P. M. Corcoran, J. Desbonnet, and K. Lusted, “CEBus Network Access via the World-Wide-Web,” in
International Conference on Consumer Electronics, 1996, p. 236.
[140] J. Desbonnet and P. M. Corcoran, “System architecture and implementation of a CEBus/Internet
gateway,” IEEE Transactions on Consumer Electronics, vol. 43, no. 4, pp. 1057–1062, 1997.
