Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Estimating the Changing Environmental

Impacts of ICT-Based Tasks: A Top-Down


Approach
Paul Teehan, Milind Kandlikar, and Hadi Dowlatabadi

assessed, e.g. through life cycle analysis. Second, the


Abstract—We develop a simple method for attributing the percentage of these devices and infrastructure dedicated to
environmental impacts of tasks performed on ICT devices servicing the behavior is identified, typically in terms of time
according to the amount of time spent by the user pursuing them spent, for computers, or in terms of bandwidth, for network
and the amount of network bandwidth they require. We use infrastructure. The impact of the behavior is simply a
broad categories of tasks and ensure that the total impacts of all percentage of the impact of the devices and infrastructure.
tasks add to 100%. Using publicly available data, we develop a This can be compared against the impact of an alternative
case study of energy consumption in the U.S. and demonstrate
behavior in order to determine which is environmentally
that by 2015 web-based video will become a dominant driver of
ICT energy consumption, due to its very high bandwidth preferable. Several studies have followed this pattern,
requirements. Our approach is useful for the identification of examining downloaded music vs. physical retail CD purchase
broad trends in the aggregate behavior of the entire user-base. [2][3], online mail-order DVD rental vs. physical retail DVD
rental [4], online academic journals vs. paper journals [5], e-
Index Terms— Communication, energy measurement, mail vs. postal mail [6], and others. All studies found
environmental factors, information technology. environmental advantages to digital delivery methods.
We propose a slightly different approach to calculating the
impacts of technology-related tasks. Rather than estimate the
I. INTRODUCTION impact of an individual task from one user, we estimate the

I NFORMATION and communication technologies (ICT) impacts of all tasks at once, i.e. the total impact of all ICT
facilitate extraordinarily rapid change across many sectors products and infrastructure, and apportion these impacts
of our society. Two common examples of these changes are according to the portion of products and infrastructure
the emergence of virtual goods, such as downloaded music dedicated to each task. Unlike individual impact studies, we
and movies which may substitute for physical goods, and take care that these portions add to 100%. This approach
videoconferencing which may substitute for travel. The forces us to use broad categories of tasks less specific than the
environmental implications of these changes are the subject of previously cited studies, but it provides insight not available
much recent attention due to their potential impact on key through other methods. In particular, we can compare tasks to
environmental indicators such as global greenhouse gas see which ones have the highest impact in the aggregate;
emissions [1]. Assessing large-scale impacts of ICT on the furthermore, we can explore whether the total impacts of each
environment requires accurate quantification of the impacts of task are higher due to the consumer products used or due to
these new behaviors and the impact of ICT itself in enabling the backend infrastructure that supports them. Our model
these changes. also offers enough flexibility to forecast into the future and
Technology-related behaviors are normally assessed in two investigate how these relationships will change over time.
stages. First, the direct environmental impacts of all devices Here we focus on network-enabled devices for two reasons.
and infrastructure involved in supporting the behavior are First, networks, including the backend datacenters that support
them, represent a large and growing portion of the overall ICT
This work was supported in part by the Natural Sciences and Engineering
“footprint”, and thus should be explicitly included in any
Research Council of Canada and by the Canadian Institutes for Health analysis of the impacts of technology-related tasks. Second,
Research/Michael Smith Foundation of Health Research Strategic Training non-networked devices, such as televisions, printers, and so
Program Bridging Public Health, Engineering & Policy Research. on, are generally used for only one task, and thus can be
P. Teehan is with the Institute for Resources, Environment and
Sustainability at the University of British Columbia, 2202 Main Mall,
adequately analyzed using traditional life cycle analysis. The
Vancouver, BC, Canada V6T 1Z4. (phone: 604-816-4386; e-mail: relationship between end-user behavior and environmental
pteehan@interchange.ubc.ca). impacts for network-enabled devices is not an obvious one,
M. Kandlikar is with the Institute of Asian Research and the Liu Institute and while there are estimates of the aggregate impacts of some
for Global Issues at the University of British Columbia. (e-mail:
milind.k@ubc.ca).
of these networks (in particular, the internet [7][8]), these
H. Dowlatabadi is with the Institute for Resources, Environment and estimates are generally not mapped to the end-user devices
Sustainability and the Liu Institute for Global Issues at the University of that access the networks.
British Columbia. (e-mail: hadi.d@ubc.ca).

Authorized licensed use limited to: University of Ioannina. Downloaded on November 09,2022 at 17:42:57 UTC from IEEE Xplore. Restrictions apply.
II. MODEL allocated a share of production and disposal impacts consistent
with their choice of method. One also included a portion of
A. Formulation the impacts of lighting and heating or cooling the room in
Suppose we are interested in the aggregate impact of all which the computer is used [4], but this approach is not
users performing computational task t on device d. Tasks common.
could include browsing the web, watching videos, playing In the interests of ensuring all impacts are accounted for, and
games, etc., while devices could include desktop PCs, laptop that the sum impacts of all tasks adds to 100%, we can either
PCs, or mobile internet devices (i.e. smartphones). We define define “overhead” to be a task, or we can distribute the
the total environmental impact of such a task, E(t,d), to be the overhead amongst the other tasks. Overhead is itself
sum of the impacts from using the devices and the impacts ambiguously defined; in addition to idle electricity, it might
from using the network that services the task, as follows: include time spent running autonomous background processes,
or even idle CPU cycles while the user process waits for input
E(t , d ) = E local (t , d ) + E network (t , d ) . or is otherwise paused. We feel that such ambiguities are best
(0.1) avoided, and that the burden factor approach is more useful,
since it maps impacts according to user intent. Accordingly,
The impact has been split into two components which can be the impact share of a task t on device d is defined as follows:
considered independently. Making the simplifying
assumption that all the devices of a particular type d1 are Time(t , d )
identical in manufacturing and performance, the local sharelocal (t , d ) =
component must account for the physical impacts of the ∑ Time(all tasks, d ) . (0.4)
device, which is a fraction of the physical impact of one such
device, E(d1), multiplied by the device installed base, N (d1). The impact of internet infrastructure due to a specific task is
The fraction should reasonably represent the portion of the instead best defined relative to the amount of data traffic, or
device that is used for task t. The form is as follows: bandwidth, the task is responsible for. Studies of individual
tasks have measured the impact of this data traffic by
E local (t , d ) = N(d ) ⋅ E(d ) ⋅ sharelocal (t , d ) . calculating the energy efficiency of the internet in terms of
(0.2)
kWh per Gb by estimating the total energy consumption of the
Likewise, the network component must account for the internet and dividing by the total amount of data traffic
physical impacts of the network, which we represent with [9][10]. The energy costs of transmitting a fixed amount of
E(n), multiplied by an appropriate fraction to isolate the data have decreased dramatically over time [10]. However,
portion of the network that is dedicated to serving task t on all the amount of data transmitted per user has grown
devices d: tremendously [11], and overall internet energy continues to
grow modestly [8]. This signals the potential existence of a
significant rebound effect, in which users download more
E network (t , d ) = E(n) ⋅ share network (t , d ) . content as the price of bandwidth decreases (in monetary
(0.3)
terms and in time to delivery), which may overwhelm any
The physical impacts E can be measured using traditional environmental advantages of digital content delivery. In
methodologies such as life-cycle analysis, depending on the addition, it suggests that the bandwidth demands of the
type of impact that is of interest; in this paper we measure average user’s mixture of tasks are in flux, and growing. In
annual energy consumption. Note that E is determined order to explore these trends and identify which tasks are
independent of the mixture of tasks and is assumed to be an generating this traffic, we need to work with aggregate figures
average impact. This frees us to use estimates of E from representing the entire user-base, and we again need to
different sources. normalize the share with respect to the entire internet traffic
load. The form is as follows:
B. Calculating Shares
In devices like computers which support multiple tasks, the Traffic(t , d )
impact of a given task is usually defined to be proportional to sharenetwork (t , d ) =
the amount of time the user spends performing the task. ∑ Traffic(all network) . (0.5)
However, there is some ambiguity in how to account for
overhead, including manufacturing and disposal impacts and All analysis under this model proceeds by first identifying
use-phase energy when the device is idle or off. One method the total amount of time spent on each device and the total
is to estimate the actual resources consumed strictly in amount of traffic carried by the network, and dividing these
servicing the task, i.e. the electricity used to power the device into their component parts.
for only the amount of time spent performing the task, leaving
idle electricity separate; [2] [3] and [6] use this approach. The
second is to charge the overhead to each task according to that
task’s portion of total active use time (rather than total device
lifetime). The latter approach was used in [4] and [5], in
which this portion was called a “burden factor”. All studies

Authorized licensed use limited to: University of Ioannina. Downloaded on November 09,2022 at 17:42:57 UTC from IEEE Xplore. Restrictions apply.
III. STUDY DESIGN
TABLE I
A. Scope and approach INSTALLED BASE AND ENERGY CONSUMPTION ESTIMATES

To demonstrate the insights available with this model, we Value (Year) CAGR
analyze the environmental impact of common tasks on three
major computing platforms: desktop PCs, laptop PCs US installed base (millions)
(including netbooks), and mobile devices, i.e. smartphones. Desktop PC/Monitor 160a/160b (2010) -4% (2010-2015)a
Computing tasks are divided into six broad categories: offline
Laptop PCc 130a (2010) +11% (2010-2015)a
tasks (e.g. word processing), online gaming, instant messaging
(IM), peer-to-peer file sharing (p2p), web video, and other Mobile/smartphone 87 (2009) [12] +34% (2009-2014) [12]
non-video web browsing, including e-mail.
We choose to study energy consumed by these devices and Unit energy consumption [kWh/year]
the supporting network in the U.S., primarily for reasons of Desktop PC/monitor
convenience and data availability. A future broader analysis use 237/85 (2007) [13] No data
could include other life-cycle impacts such as total carbon manufacturingd 592e (2004) [14] No data
footprint or total toxic emissions, so long as the impacts are Laptop PC
expressed numerically and are consistently measured across use 72 (2007) [13] No data
all consumer devices and across all devices comprising the manufacturing 300f No data
network. Public phone networks and television broadcast Mobile/smartphone
networks could also be included; other devices, including use 5 (2007) [15] No data
televisions and set-top-boxes, game consoles, e-book readers, manufacturing 14g No data
and other emerging devices could be analyzed similarly. For
simplicity, and due to limited data availability, we leave these Aggregate network energy consumption [TWh/year]
extensions for future study. Internet
In order to determine how the environmental impacts of Servers (use only)h,i 44.9 (2006) [10] 12% (2000-2006) [10]
each task are distributed between local and network devices, Cooling overheadh 30.7 (2006) [16] 14% (2000-2006) [16]
we need to fill in each term in the above equations, including Total 75.6 13%
installed base for each product, total energy consumption for a
Author’s estimates derived from data in [17].
each product and for the network, a list of tasks and the share b
Assumed each desktop PC also has a monitor attached, and laptops do not
of each device attributable to each task, and the share of the have monitors attached.
c
network attributable to each task. We have not yet had the Includes notebooks, netbooks, other similar devices
d
Energy due to end of life phase is negligible relative to use and
opportunity to measure any of these ourselves, and so rely on manufacturing phases [5]
available published studies, the majority of which come from e
Based on 6400MJ manufacturing energy over a 3 year lifetime, includes PC
market research firms. When possible we seek recent figures + CRT monitor [14]
f
in order to limit uncertainty due to rapid technical change. g
No data; assumed of 50% of the desktop+monitor
In order to project trends into the future, we also seek the Based on 130MJ manufacturing energy over a 2.5 year lifetime [18]
h
Study [10] updates and expands [16] to include network infrastructure e.g.
compound annual growth rate (CAGR) of each figure. routers and hubs, but removed datacenter cooling infrastructure. We believe
Estimates that occurred in the past are projected to the present the cooling infrastructure should be included, and thus add it back in to the
with the best available figures; all estimates are then forecast results in [10].
i
from 2010 to 2015. The following section lists all data and Manufacturing phase excluded; assumed to be small relative to server use-
assumptions used in this analysis. phase energy (no data).

B. Data sources and assumptions Our analysis of the bandwidth demand of different tasks
Table I below displays estimated installed base for all three relies on data from a large-scale internet traffic study
product categories, their average unit energy consumption, and conducted by Cisco Systems called the Visual Networking
the energy consumption of the internet and associated Index [19] [20], which estimates the total monthly traffic
infrastructure, including cooling. The results are drawn from a broken down into several categories, and forecast five years
range of publicly available sources as indicated in the notes. into the future. We unfortunately cannot reproduce the data
We include the manufacturing energy, distributed over the due to lack of space, but it is available from the authors upon
lifetime of the device, as part of the annual energy request, or from the original source. Notably, most categories
consumption. Energy consumed during the end-of-life phase have very high annual growth rates. Overall consumer internet
is negligible relative to manufacturing and use-phase energy traffic grows by 40% per year; within this category, online
[5] and is thus excluded. There do not appear to be any video grows by 60% per year. Mobile data traffic is small but
studies estimating manufacturing energy for servers, but we doubling annually; mobile online video grows fastest, at 131%
expect this will be small relative to their use-phase energy per year. Generally speaking, online video is becoming a
consumption, since servers have high operating power, long dominant portion of all internet traffic. We explore the energy
lifetimes, and are usually on all the time. There is no data implications of this shift later in the paper.
regarding unit energy consumption trends for the consumer Knowing both overall network energy and the composition
products. In our experience the trends are relatively static of network traffic, we can calculate the portion of the network
over short time scales (<10 years). energy consumed in servicing these tasks. To calculate the

Authorized licensed use limited to: University of Ioannina. Downloaded on November 09,2022 at 17:42:57 UTC from IEEE Xplore. Restrictions apply.
impact due to use of consumer devices, we also need to know
U.S. annual energy consumption (GWh)
the amount of time spent on each task. Unfortunately, such
data appears to be relatively scarce. Only one study, based in 0 0 0 0 00
00 00 00 00 00
France in 2009 [21], has appropriate measurements, which are 0 20 40 60 80 10
shown in Table II. We assume the results are equivalent for
mobile web
the U.S. The study used automated measurements of
participant’s computers, which is essential in order to avoid mobile VoIP
double-counting due to multi-tasking. We assume the usage mobile v id
patterns are equivalent for laptops. Unfortunately we have not
mobile p2p
found any similar studies targeting mobile devices. We
assume that the time spent on a given task on mobile device is mobile game
proportional to the traffic generated by that task. This
assumption is questionable, but not critically important, since laptop web v id
the overall local energy due to mobile devices is quite small
laptop web non-v id
and not likely to be of concern. There is also no data
regarding potential growth rates for any of these tasks. We laptop p2p
cautiously assume that time spent on internet video will laptop of f line
increase 10% per year while other tasks remain constant.
laptop IM
TABLE II laptop game
DEVICE USAGE BREAKDOWN BY USER ATTENTION TIME

Task category Device sharea CAGR desktop web v id

desktop web non-v id


PC offline 33% No data
PC total online 67% No data desktop p2p
PC p2p 3% No data
PC gameb 4% No data desktop of f line
PC IM 8% No data
desktop IM
PC web total 52% No data
PC web vid 3.4% [22]c No data d desktop game
PC web non-vid 48.6% No data

Laptop [all tasks] Assume same as desktop PC 2010 dev ice 2010 network
2015 dev ice 2015 network
Mobile [all tasks] No datae No data
a
All device share figures from [21] unless otherwise noted
b
[21] reported total 8% of time for all gaming and appears to show roughly Fig. 1 Total U.S. energy usage by device and task, including local and
50% of gaming is internet-enabled. network energy, grouped by device
c
[21] did not track web video. [22] reports web video time is 12% of total
internet time which would be 3.4% of total device share using model from
[21]. The graph depicts immediate evidence of the disparity
d
No data for CAGR of device usage by task. Assume PC web video grows between tasks. Some tasks, like instant messaging, consume
by 10% per year in terms of hours per week while all other tasks remain comparatively little energy, while other tasks, like web
constant. browsing (both video and non-video) consume a large amount
e
Assume devise usage is proportional to network data usage. of energy. Tasks on which the user spends a lot of time
consume more energy, as do bandwidth-intensive tasks. The
IV. ANALYSIS highest-energy tasks in 2010 are desktop web-browsing (non-
video) and desktop offline use, due to the large amount of time
A. Energy consumption by task
spent pursuing these tasks – about 80% of all active use,
Fig. 1 shows the annual energy consumption in the U.S. due to according to the data in Table 3 – and the relatively high unit
all computational tasks across all three computational devices, energy consumption of the average desktop computer, which
using 2010 and 2015 figures calculated with the CAGRs listed includes idle energy and manufacturing energy under our
in the above tables. The tasks are grouped according to the model. In 2015, desktop web-browsing (non-video) is still
type of device, with network energy and device energy forecast to be the largest consumer of energy, but the total
(including manufacturing and idle overhead) compared on the amount has decreased with the falling desktop installed base.
same bar. The figures are aggregates for the entire U.S. Web video energy for both desktops and laptops increases
installed base. dramatically by 2015 due to tremendous increases in network
energy to service these tasks, which corresponds to a higher
demand for bandwidth. Mobile video’s energy use is
moderate despite tremendous growth. Other mobile tasks are
negligible in terms of their energy consumption.

Authorized licensed use limited to: University of Ioannina. Downloaded on November 09,2022 at 17:42:57 UTC from IEEE Xplore. Restrictions apply.
These results are sensitive to the growth rate of the overall Note, however, that we cannot compare directly to such
energy due to the internet. Using a 13% growth rate as in Fig. studies, since our model describes only an average mixture of
1, which indicates a doubling about every six years, internet all web-based tasks. Following the model in [13], the
energy due to laptop web video (for example) is projected to desktop’s active power is 100W (70W for the computer, and
grow from 9200 GWh to 54000 GWh. If instead we 30W for an LCD monitor), while the laptop’s active power is
optimistically assume the internet energy were to remain 25W. We exclude mobile devices from this graph due to our
constant from 2010 to 2015, then the internet energy due to uncertainty about their usage patterns. In addition, we assume
laptop web video would increase from 9200 GWh to only overheads due to manufacturing and idle time in servers are
18000 GWh, due only to increase in this task’s relative share small relative to use-phase energy.
of all network traffic.
B. Aggregate trends
Laptop
The overall energy consumption due to local devices and Internet energy
their corresponding network demands are compiled in Figure Ov erhead:
2. Device use includes idle and manufacturing overhead. Desktop manuf acturing
Note that in 2010, local energy consumption due to desktops Ov erhead: idle

comprises nearly half of the total energy consumption, but the A c tiv e energy
0 20 40 60 80 100 120 140 160
share gradually decreases as the installed base shrinks in favor Energy per 10 minutes web non-v ideo task (Wh), 2010 data
of laptops. Laptop local energy is smaller but grows
proportionally. Local energy due to mobile devices is Fig. 3 Energy consumed, including pro-rated overhead, during ten minutes of
active use web-browsing (non-video).
negligible. The share of internet energy grows steadily, with
the largest expansion due to demand from laptops; this is due
The relatively small amount of energy consumed in actually
to expected increases in the laptop installed base and
completing the task highlights the need to consider
corresponding increasing demand for internet video.
manufacturing and idle overheads, as these overheads account
for the majority of the energy consumption. Aggressive
Total energy due to U.S. installed base

100%
power management could reduce these overheads, as would
90%
more efficient manufacturing techniques. Note, however, that
80%
Device: mobile
the estimates we use for manufacturing overhead are from
70%
Device: laptop 2004 [14]; this overhead may have since decreased. Network
60% Device: desktop energy during web browsing is comparably small, likely
50% because many forms of web browsing, such as email or
40% Internet: mobile reading news, can absorb large amounts of user attention
30% Internet: laptop while requiring only a small amount of data transfer.
Internet: desktop
20%
Internet: (other)
10%
0% D. Network efficiency
2010 2011 2012 2013 2014 2015 According to our model and US internet traffic estimates in
Fig. 2 Aggregate trends in energy consumption: 100% graph including U.S. [11], this ten minutes of web browsing generates about 5
local device energy and U.S. internet energy megabytes of traffic, which is responsible for 18Wh of
network energy, visible in Figure 2. The energy burden of this
C. Device efficiency data transfer is 3.6kWh/Gb. We have used the same data
The local energy consumption figures are likely larger than sources as [10] in calculating this figure, but with more recent
other estimates due to our method of apportioning overheads estimates; their 2006 estimate was 8kWh/Gb. Thus, the
among active use tasks. According to [21], the average energy intensity of data transfers has dropped considerably
computer is actively used for about 9 hours per week; since then. Note, however, that it is unlikely that the average
similarly, a Nielsen study reports 15 hours per week of active network energy consumption per ten minutes of browsing has
use [22]. On the other hand, according to [13], desktop PC’s dropped as well.
are in high-power active mode about 34% of the time, or about Figure 4 shows the trends in network traffic and network
57 hours per week, suggesting that each hour of activity is energy due to demand from laptop computers. The left axis
accompanied by three to five hours of inactivity while still in shows energy in kWh, while the right axis shows traffic in Gb.
high-power active mode. Likewise, laptops are in active mode The projections show that network efficiency will continue to
27% of the time, or about 45 hours per week [13], suggesting grow – the ratio between the two curves is approaching unity,
two to four hours of inactivity per hour of active use. 1kWh per Gb – but only because the energy growth rate is
Fig. 3 makes explicit the relationship between energy exceeded by the traffic growth rate. The environmental
consumed during use and the idle and manufacturing energy impact per unit of user time will continue to grow in step with
overhead. It also compares network energy, assuming the the total energy consumption of the network.
computer is used for non-video web browsing, using 2010
data. Ten minutes of continuous active use was assumed, in
order that the results might be comparable to studies of
individual tasks, which often require similar lengths of time.

Authorized licensed use limited to: University of Ioannina. Downloaded on November 09,2022 at 17:42:57 UTC from IEEE Xplore. Restrictions apply.
vol. 11, 2007, pp. 77-91.
[5] D.L. Gard and G.A. Keoleian, “Digital versus Print: Energy
Performance in the Selection and Use of Scholarly Journals,” Journal
400 400
of Industrial Ecology, vol. 6, 2002, pp. 115-132.
Annual energy per laptop, U.S. (kWh)

Annual traffic per laptop, U.S. (Gb)


350 350 [6] M. Zurkirch and I. Reichart, “Environmental Impacts of
Telecommunication Services Two Life-Cycle Analysis Studies,”
300 300 Greener Management International, vol. 32, 2000, pp. 70-88.
[7] J. Baliga, K. Hinton, and R. Tucker, “Energy Consumption of the
250 250 Internet,” Joint International Conference on Optical Internet and
32nd Australian Conference on Optical Fibre Technology, 2007, pp.
200 200
1-3.
150 150 [8] J.G. Koomey, “Worldwide electricity used in data centers,”
Environmental Research Letters, vol. 3, 2008, p. 034008.
100 100 [9] J. Koomey, H. Chong, W. Loh, B. Nordman, and M. Blazek,
“Network Electricity Use Associated with Wireless Personal Digital
50 50 Assistants,” Journal of Infrastructure Systems, vol. 10, 2004, pp. 131-
137.
0 0 [10] C. Taylor and J. Koomey, Estimating energy use and greenhouse gas
2010 2011 2012 2013 2014 2015 emissions of Internet advertising., 2008.
[11] A. Odlyzko, “MINTS - Minnesota Internet Traffic Studies,” 2009.
Energy Traffic
[12] IDC, U.S. Converged Mobile Device Installed Base 2009-2013
Forecast, 2009.
Fig. 4 Annual network energy (left axis, light) and network traffic (right axis, [13] K.W. Roth and K. McKenney, Energy Consumption by Consumer
dark) projections, laptop computers in the U.S. Electronics in U.S. Residences, Cambridge, MA: TIAX LLC, 2007.
[14] E. Williams, “Energy Intensity of Computer Manufacturing: Hybrid
Assessment Combining Process and Economic Input−Output
Methods,” Environmental Science & Technology, vol. 38, Nov. 2004,
V. DISCUSSION pp. 6166-6174.
[15] J.A. McAllister and A.E. Farrell, “Electricity consumption by battery-
powered consumer electronics: A household-level survey,” Energy,
The analysis has shown that the majority of energy vol. 32, Jul. 2007, pp. 1177-1184.
consumption from most ICT tasks is due to the local devices, [16] EPA, Report to Congress on Server and Data Center Energy
with the key exceptions of online video and peer to peer file Efficiency Public Law 109-431, U.S. Environmental Protection
Agency ENERGY STAR Program, 2007.
sharing. Online video in particular will require large and [17] eTForecasts, Worldwide PC Forecast, 2009.
increasing amounts of energy, the vast majority of which will [18] S.D. Frey, D.J. Harrison, and E.H. Billett, “Ecological Footprint
be due to the network. This is not surprising given current Analysis Applied to Mobile Phones,” Journal of Industrial Ecology,
internet traffic forecasts. vol. 10, 2006, pp. 199-216.
[19] Cisco Systems Inc., Cisco Visual Networking Index: Forecast and
Desktop-based tasks consume a great deal of energy, most Methodology, 2008-2013, Cisco Systems Inc., 2009.
of which is overhead. The recent emergence of cloud [20] Cisco Systems Inc., Cisco Visual Networking Index: Global Mobile
computing, a computational model in which tasks are Data Traffic Forecast Update, Cisco Systems Inc., 2009.
offloaded to server farms, alongside the emergence of new [21] T. Beauvisage, “Computer usage in daily life,” Proceedings of the
27th international conference on Human factors in computing
portable network-driven products that access these resources, systems, Boston, MA, USA: ACM, 2009, pp. 575-584.
should have a beneficial impact on consumer energy [22] Nielsen, A2/M2 Three Screen Report, 2009.
consumption, since these portable devices require much less
energy than desktops and laptops. It remains to be seen
whether this impact will be offset by continuing increases in
backend network infrastructure energy; more recent data is
needed.
This exercise highlights the usefulness of a top-down
approach. Studies of individual tasks may show the
environmental advantages of a particular behavior performed
once, but the overall impacts of these behaviors can only be
analyzed in the aggregate. Otherwise, the trends described in
this paper would remain hidden.

REFERENCES
[1] The Climate Group, Global e-Sustainability Initiative, and
McKinsey&Company, Smart2020: Enabling the low carbon economy
in the information age, 2008.
[2] C.L. Weber, J.G. Koomey, and H.S. Matthews, The Energy and
Climate Change Impacts of Different Music Delivery Methods,
Microsoft Corporation and Intel Corporation, 2009.
[3] V. Turk, V. Alakeson, M. Kuhndt, and M. Ritthoff, The environmental
and social impacts of digital music - A case study with EMI, 2003.
[4] D. Sivaraman, S. Pacca, K. Mueller, and J. Lin, “Comparative Energy,
Environmental, and Economic Analysis of Traditional and E-
commerce DVD Rental Networks,” Journal of Industrial Ecology,

Authorized licensed use limited to: University of Ioannina. Downloaded on November 09,2022 at 17:42:57 UTC from IEEE Xplore. Restrictions apply.

You might also like