ChapterPDF Available

The Energy Intensity of the Internet: Home and Access Networks


Abstract and Figures

Estimates of the energy intensity of the Internet diverge by several orders of magnitude. We present existing assessments and identify diverging de�finitions of the system boundary as the main reason for this large spread. The decision of whether or not to include end devices influences the result by 1-2 orders of magnitude. If end devices are excluded, customer premises equipment (CPE) and access networks havea dominant influence. Ofless influence is the consideration of cooling equipment and other overhead, redundancy equipment, and the amplifi�ers in the optical �bers. We argue against the inclusion of end devices when assessing the energy intensity of the Internet, but in favor of including CPE, access networks, redundancy equipment, cooling and other overhead as well as optical fi�bers. We further show that the intensities of the metro and core network are best modeled as energy per data, while the intensity of CPE and access networks are best modeled as energy per time (i.e., power), making overall assessments challenging. The chapter concludes with a formula for the energy intensity of CPE and access networks. The formula is presented both in generic form as well as with concrete estimates of the average case to be used in quick assessments by practitioners. The following chapter develops a similar formula for the core and edge networks. Taken together, the two chapters provide an assessment method of the Internet's energy intensity that takes into account diff�erent modeling paradigms for di�fferent parts of the network.
Content may be subject to copyright.
The Energy Intensity of the Internet:
Home and Access Networks
Vlad C. Coroama1, Daniel Schien2, Chris Preist2, and Lorenz M. Hilty3,4,5
1Measure-IT Research, Bucharest, Romania
2Department of Computer Science, University of Bristol, UK
3Department of Informatics, University of Zurich, Switzerland
4Empa, Swiss Federal Laboratories for Materials Science and Technology,
St. Gallen, Switzerland
5Centre for Sustainable Communications CESC, KTH Royal Institute of
Technology, Stockholm, Sweden
Abstract. Estimates of the energy intensity of the Internet diverge by
several orders of magnitude. We present existing assessments and iden-
tify diverging definitions of the system boundary as the main reason for
this large spread. The decision of whether or not to include end devices
influences the result by 1-2 orders of magnitude. If end devices are ex-
cluded, customer premises equipment (CPE) and access networks have
a dominant influence. Of less influence is the consideration of cooling
equipment and other overhead, redundancy equipment, and the ampli-
fiers in the optical fibers. We argue against the inclusion of end devices
when assessing the energy intensity of the Internet, but in favor of in-
cluding CPE, access networks, redundancy equipment, cooling and other
overhead as well as optical fibers. We further show that the intensities of
the metro and core network are best modeled as energy per data, while
the intensity of CPE and access networks are best modeled as energy per
time (i.e., power), making overall assessments challenging. The chapter
concludes with a formula for the energy intensity of CPE and access
networks. The formula is presented both in generic form as well as with
concrete estimates of the average case to be used in quick assessments
by practitioners. The following chapter develops a similar formula for
the core and edge networks. Taken together, the two chapters provide
an assessment method of the Internet’s energy intensity that takes into
account different modeling paradigms for different parts of the network.
Keywords: Internet, Energy Intensity, Energy Efficiency, Customer Premises
Equipment, Access Network
1 Introduction
Information and Communication Technologies (ICT) are increasingly perceived
as enablers of a reduction of anthropogenic greenhouse gas (GHG) emissions.
Studies providing evidence for this enabling effect come not only from academia
This Accepted Author Manuscript is copyrighted by Springer. The final publication will be
available via by end of August 2014. Suggested citation:
Coroama, V.C., Schien, D., Preist, C., Hilty, L.M.: The Energy Intensity of the Internet: Home and
Access Networks. In: Hilty, L.M., Aebischer, B. (eds.) ICT Innovations for Sustainability. Advances
in Intelligent Systems and Computing 310, Springer International Publishing (2014, in press)
This Accepted Author Manuscript is copyrighted by Springer. The final publication will be
available via by end of August 2014. Suggested citation:
Coroama, V.C., Schien, D., Preist, C., Hilty, L.M.: The Energy Intensity of the Internet: Home and
Access Networks. In: Hilty, L.M., Aebischer, B. (eds.) ICT Innovations for Sustainability. Advances
in Intelligent Systems and Computing 310, Springer International Publishing (2014, in press)
This Accepted Author Manuscript is copyrighted by Springer. The final publication will be
available via by end of August 2014. Suggested citation:
Coroama, V.C., Schien, D., Preist, C., Hilty, L.M.: The Energy Intensity of the Internet: Home and
Access Networks. In: Hilty, L.M., Aebischer, B. (eds.) ICT Innovations for Sustainability. Advances
in Intelligent Systems and Computing 310, Springer International Publishing (2014, in press)
2V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
[1–3], but also from organizations as diverse as ICT industry associations [4],
the European Commission [5], and the World Wildlife Fund [6]. Reductions are
usually estimated based on quantitative scenarios [7], yielding an abatement
potential which has to be adjusted downwards by an estimate of the ICT appli-
cations own footprint to calculate the net effect.
ICT can reduce energy consumption and related GHG emissions through
three mechanisms: i) from optimization effects in domains such as smart en-
gines, buildings, or logistics, ii) due to ICT-supported novel paradigms for the
generation and distribution of electricity (i.e., smart grids), and iii) due to sub-
stitution effects in which information and communication services partly replace
other more energy-intensive activities [4,8].
Two problems must be solved to quantify the net effect of ICT applications
in these cases: First, the energy savings induced by ICT must be assessed. As we
argue in [9] and [10], this is methodologically challenging: The baseline scenario,
among other factors, as it expands into the future, is inherently speculative.
Moreover, allocation issues are raised by the fact that ICT typically does not
induce efficiency on its own, but only in a suitable technological, political, or
organizational context. Secondly, the energy consumption of the ICT solution
involved must be determined. This is also technically challenging, and existing
literature reports diverse results. The current and the subsequent chapter explore
this issue, with a particular focus on Internet services.
The energy intensity of the Internet, expressed as energy consumed to trans-
mit a given volume of data, is one of the most controversial issues. Existing
studies of the Internet energy intensity give results ranging from 136 kWh/GB
[11] down to 0.0064 kWh/GB [12], a factor of more than 20,000. Whether and to
what extent it is more energy efficient to download a movie rather than buying
the DVD, for example, or more sustainable to meet via videoconference instead
of travelling to a face-to-face meeting are questions that cannot be satisfyingly
answered with such diverging estimates of the substitute’s impact.
The Internet’s energy consumption, and the energy intensity of the Internet
as a suitable metric, are the topics of both the current chapter and the next
one. The current chapter presents a review of existing studies and provides ex-
planations for the large spread in their results. The chapter then recommends a
definition of the system boundary that is most useful for decision-making and
concludes by zooming in on the peripheral parts of the network and assessing
their contribution to the energy intensity of the Internet. The next chapter [13]
will focus on the remaining components, i.e., the core of the Internet.
2 Definitions and System Boundaries
In the late 1960s and early 1970s, the term Internet literally indicated the in-
terconnection of a small number of local area networks on university campuses;
back then, “the Internet” comprised a few routers and cables. Today, the Inter-
net is the vast and heterogeneous infrastructure connecting billions of computers
worldwide using the TCP/IP family of communication protocols. The majority
The Energy Intensity of the Internet: Home and Access Networks 3
of these computers belong to private users who connect to the Internet through
their Internet service provider (ISP).
Figure 1 presents a high-level structure of the Internet. We comment only
briefly on this structure here; more detailed discussions can be found in the
following chapter in section 2.1, and in the network models presented by [14,
15]. Users’ devices (such as desktop, laptop or tablet PCs and smartphones)
connect through what is referred to as customer premises equipment (CPE),
which are mainly WiFi routers and modems, to their ISP. The ISP bundles the
data from several users in multiplexers. These vary depending on the subscription
technology; for the widely used DSL connectivity they are digital subscriber
line access multiplexers (DSLAM). Together with the cables connecting them to
the CPE, these multiplexers constitute what is called the access network. After
passing through an edge router, the traffic enters the metro and core parts of
the Internet where routers with increasing capacities bundle the traffic. On the
other side, the traffic is decomposed according to its destination; a large part
is directed to data centers, while a smaller part is directed to other users (not
represented in Figure 1).
CPE Metro Long Haul
Router Router
Data Center
Next ChapterThis Chapter
Fiber Optic
Ring Topology Fiber Optic
Mesh Topology
Fig. 1. Model of the Internet structure.
First attempts to understand the energy consumption of this distributed
and heterogeneous power-consumption system were undertaken a decade ago, in
2003-2004. Starting from statistical data and studies on the ICT equipment in
use, both [16] and [11] estimated the yearly power consumption of the Internet
in the US. Dividing this value by an estimate of the US Internet traffic for that
year resulted in estimates of the energy intensity of the Internet, i.e., the energy
consumed throughout the Internet per amount of data transferred.
The two studies differed in their definition of what constitutes the Internet.
While [16] considered only networking equipment (i.e., Ethernet hubs, LAN and
WAN switches in offices and buildings, together with the routers of the Internet),
[11] also took into account devices in data centers such as servers or data storage
(see Figure 1). As we discuss in more detail later, this definitional discrepancy
4V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
across studies persists until today, and is one of the main causes for the large
spread of the published results.
Before going into detail, we note that we discuss estimates for direct energy
consumption in the form of electricity only. The energy supply chain (containing
the supply of primary energy, power plants transforming primary energy sources
to electricity, and grids bringing them to the consuming devices) is excluded
from the system under study. We also exclude the “grey” energy embedded in
ICT hardware, although the material flows caused by producing and disposing
of hardware are significant (see [17] and the chapter by Hischier et al. [18] later
in this volume).
All studies that will be discussed apply an average allocation rule, distribut-
ing the equipment energy consumption evenly among the total traffic volume
over a certain period of time. We found no study focusing on marginal instead
of average effects.
3 Existing Assessments and Methods Used
The two assessments mentioned above, [11] and [16], both used the same method-
ology: dividing an estimate of the overall US Internet energy consumption by
the estimated total Internet traffic in the US. Since then, several more studies
used the same approach, while other studies deployed different methodologies.
According to the basic methodological approach they use, existing studies can
be classified in two classes:
Top-down. According to [19], top-down analyses are based on two estimates:
1) the overall energy demand of either the entire Internet or a part of it (e.g., a
country or a continent), and 2) the total Internet traffic of that region.
Bottom-up. By contrast, bottom-up approaches model parts of the Internet
(i.e., deployed number of devices of each type) based on network design princi-
ples. Such a model combined with manufacturers’ consumption data on typical
network equipment leads to an estimate of the overall energy consumption [15],
which is then related to an estimate of the corresponding data traffic. Bottom-
up studies can also use direct observations made in one or more case studies.
These provide primary data on one or more of the following: deployed equipment,
network topology and routing, power consumption of specific devices, and data
volume passing through specific devices. Case studies present energy intensity
values for specific cases, typically followed by a discussion of how the results can
be generalized.
3.1 Top-Down Assessments
The earliest top-down assessments were introduced above: Gupta et al. published
their results in 2003 [16], and Koomey et al. in 2004 [11]. The two studies used
The Energy Intensity of the Internet: Home and Access Networks 5
the same statistical inventory data on ICT equipment in commercial buildings
in the US for the year 2000 [20]. Despite building on the same inventory, they
make different assumptions as to which devices belong to the Internet and thus
yield distinct consumption results.
The assessment by Gupta et al. [16], taking into account Internet routers,
WAN and LAN switches as well as Ethernet hubs, yields a yearly energy con-
sumption of 6.05 terawatt-hours per year (TWh/a) for all networking devices
in the US. In assessing the energy intensity of “the Internet” in the US, the
study narrows the focus even more. It leaves campus devices (i.e., LAN switches
and hubs) outside the calculation, and considers only the consumption of WAN
switches and Internet routers, estimated at just 1.25 TWh/a in 2000.
Koomey et al. [11], on the other hand, consider not only the campus net-
working devices, but also data center devices – servers (10.2 TWh/a) and data
storage (1.5 TWh/a) – as well as uninterruptible power supplies (5.8 TWh/a),
leading to a total of 23.65 TWh/a. Furthermore, they multiply this result by 2
to account for overhead such as cooling and ventilation, leading to a total of 47
TWh/a, 37 times higher than the value in [16].
The two studies also use an identical source to estimate the US Internet
traffic: data from the Minnesota Internet Traffic Studies (MINTS) by Andrew
Odlyzko and colleagues. These data, published e.g. in [21], estimate a US Internet
traffic in 2000 of 20,000-35,000 TB/month. Using the same traffic data, and
consumption data different by a factor of 37, one could expect from the two
studies energy intensity results differing by the same factor 37. This, however,
is not the case. Koomey et al. [11] use the lower end of the Internet traffic
data from [21] but complement it with traffic on other public data networks
and private lines, leading to a total traffic of 348,000 terabyte per year (TB/a).
The study thus yields an energy intensity of 136 kilowatt-hours per Gigabyte
(kWh/GB) [11]. As for the other study, we can only speculate that Gupta et al.
[16] misinterpreted the data in [21] as yearly instead of monthly traffic values.
The study thus calculates with the range of 20,000-35,000 TB/a, which leads to
an energy intensity of 0.128-0.225 Joule/Byte, or 38-67 kWh/GB. The correct
energy intensity for the system boundaries used by [16], however, should have
been twelve times lower because the yearly traffic estimate was actually twelve
times higher. The corrected values for [16] are 3.2-5.6 kWh/GB (see Table 1).
An update for Koomey et al. [11] was published a few years later. The new
assessment by Taylor and Koomey [22] referred to the year 2006. Estimating
again the US Internet energy consumption and using three existing estimates of
the US Internet traffic per year, the new study yielded as result the range 8.8-
24.3 kWh/GB [22]. This 2006 estimate was yet again updated for the year 2008
in an article by Weber et al. [23]. For this period, the authors assumed that total
Internet traffic increased by 50% per year, and that total Internet electricity use
grew at a yearly rate of 14%, which had been the average global growth rate of
data center electricity use between 2000 and 2005. These assumptions resulted
in an average Internet electricity intensity of about 7 kWh/GB for 2008 [23].
6V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
The study by Lanzisera et al. [24] is another well-known top-down estimate.
The analysis only includes networking equipment, excluding not only end de-
vices but also the transmission lines. Estimating the total of both the US and
the world networking equipment stock for 2008, the power of each device and
their individual usage patterns, the article computed an annual electricity con-
sumption of 18 TWh for all networking equipment in the US and of 50.8 TWh
for the world. The study did not relate this consumption to traffic values to com-
pute the Internet energy intensity. To make the result comparable with other
studies, we divide it by an estimate for Internet data traffic for 2008 in order to
calculate the energy intensity. According to Cisco’s “Visual Networking Index,”
“global IP traffic grew 45 percent during 2009 to reach an annual run rate of 176
exabytes per year” [25]. We therefore assume a traffic volume of 121 exabytes
(EB) for 2008. Using this value as a worldwide traffic estimate for 2008 yields
an energy intensity of 0.39 kWh/GB for the world average.
3.2 Bottom-Up Assessments
The model-based approach has been used by Kerry Hinton’s research group at
the University of Melbourne [12, 14, 15,26, 27], as well as in [28] and [29]. Some of
these studies are not directly comparable to the results of top-down assessments
because they have different focuses such as analyzing only a part of the Internet
transmission (e.g., [27]) or analyzing the Internet power consumption per sub-
scriber and not per amount of data [26]. A few of these results may, nevertheless,
be adapted to be made compatible with studies on Internet energy intensity. As
we present in detail in [30], the very first assessment from the Melbourne group
[26] yields an Internet energy intensity of 0.91-2.52 kWh/GB, depending on the
estimate of worldwide Internet traffic used.
[14] provides a direct estimate of the energy intensity of Internet data trans-
mission: 75 micro-Joule per bit (µJ/bit), equal to 0.179 kWh/GB, at the access
rates typical of 2008. As the authors point out, their result represents a lower
bound or optimistic estimate in terms of energy consumption, because the model
assumes only state-of-the-art equipment and ignores the fact that less energy-
efficient legacy network equipment is still in use. They further state that they
expect this energy intensity to drop in the near future to 2-4 µJ/bit with in-
creasing access rates.
[12] puts forward a value of 2.7 µJ/bit. This value corresponds to 0.0064
kWh/GB and represents the lowest value published thus far. This study, how-
ever, aimed to compare the energy demand of traditional computing with that
of cloud computing. As the energy consumption of the access network is largely
independent of the traffic and would leave the result untouched, the authors
legitimately ignored it: “The access network does not influence the comparison
between conventional computing and cloud computing. Therefore, it is omitted
from consideration and is not included in our calculations of energy consump-
tion” [12]. While this assumption stands to reason within the scope of the study,
it can lead to misinterpretation if taken out of context, as we will discuss in the
next section.
The Energy Intensity of the Internet: Home and Access Networks 7
Finally, Schien et al. [29] used a network model to analyze the download of
the UK newspaper “The Guardian,” as well as the download of a 640 second
video from the Guardian’s video section. The newspaper’s html homepage was
located on a server within the UK, while video and images were outsourced
to a Content Distribution Network (CDN) and mirrored on several continents
within the CDN’s network. Downloads from clients in Oceania, North America,
and Europe were studied. Results showed that because of the CDN architecture,
geographical distance played only a minor role; the energy intensities of the
downloads from different continents were similar. For both the homepage and
the video, the intensity was 8-9 Joules per megabit (J/mbit), which corresponds
to approximately 0.02 kWh/GB. The study, while considering access, metro, and
core parts of the Internet, did not account for the CPE.
In [31], this work was extended to include CPE and end devices, and to
explore uncertainty and variability in assessments of digital services. In contrast
to this chapter and to [30], which try to represent the existing variability in
previous assessments, [31] estimates how uncertainty in energy intensity affects
the overall result. Combining earlier results in a triangular distribution, the
study arrived at a mean energy intensity for metro and core networks of 0.038
kWh/GB. For access network and CPE together, and excluding end devices, it
provided a mean energy estimate of 0.019 kWh/GB.
In [32], Coroam˘a and Hilty present an assessment of a 40 megabit per second
(Mbps) videoconferencing transmission of the case study introduced in [33]. For
a system boundary that included network devices and optical fibers but no end
devices, and making pessimistic assumptions in terms of energy consumption
where specific data were not available, the study yielded an energy intensity of
0.2 kWh/GB for 2009. As we argued in [32], many characteristics of the study
(such as an above-average number of hops) justify considering its result above-
average in terms of energy intensity. This implies that the case-study result,
when generalized, should be considered an upper bound for the average energy
The setting of this case study was such that no CPE or access network was
distinguishable. The conference was held between a large conference center in
Switzerland and a university campus in Japan. Both sites were directly connected
to the metro network in the same way that Figure 1 depicts data centers. Yet,
the edge routers on each side of the connection behaved similarly as CPE and
the access network behave in the typical setting: they had a low load far from
their capacity, and with an energy consumption that had to be allocated entirely
or to a large extent to the case study.
4 Factors Influencing the Results
Results of the surveyed studies span a very wide range: from the 136 kWh/GB
of [11] down to the 0.006 kWh/GB from [12], there is a spread of four orders
of magnitude. In this section, we show how the distinct assumptions about the
system boundary and further factors affected the results of the individual studies.
8V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
Table 1 summarizes the characteristics and the results of the studies pre-
sented. Special emphasis is given to their system boundaries. Our analysis re-
vealed the following factors to be the most important influences on the result:
The reference year of the study,
the inclusion of data center devices within the system boundary,
the inclusion of customer premises equipment and access network, and
the inclusion of overheads such as cooling and redundancy equipment.
Each influencing factor is discussed in the following subsections.
Table 1. Estimates of the energy demand of Internet transmissions. The columns be-
low “system boundary” show which parts of the Internet and of the end devices (as
introduced in Figure 1) were accounted for by the individual studies. CPE is the cus-
tomer premises equipment. Core stands for the metro and long haul Internet together
– all studies consider both of them. Fibers are the optical transmission lines, and DCs
stands for the equipment in data centers.
Study Method System boundary Data on Energy intensity
CPE Access Core Links DCs [year] [kWh/GB]
[16] Top-down X 2000 38 – 67
[16], corrected Top-down X 2000 3.2 – 5.6
[11] Top-down X X X 2000 136
[22] Top-down X X X X 2006 8.8 – 24.3
[23] Top-down X X X X 2008 7
[24] Top-down X X X — — 2008 0.39
[26] Model X X X 2007 0.7 – 2.1
[14] Model X X X X — 2008 >0.179
[12] Model X X X — 2011 0.006
[29] Model X X X — 2009 0.02
[31] Meta-Analysis X X X X 2009 0.057
[32] Case study X X X X 2009 <0.2
Reference Year. An important part of the large differences can be explained
by the year of reference of the individual studies, ranging from 2000 [16,11]
to 2011 [12]. The ICT sector is characterized by fast innovation cycles, and
the equipment is becoming ever more energy efficient, needing less energy per
amount of data being processed or transmitted. Taking [22]’s estimate that the
energy intensity of the Internet decreases by 30% each year, this technological
progress alone leads to a reduction by a factor of 50 over the period of 11 years
covered by the studies.
System Boundary: Data Centers. The most important determining factor
is the system boundary, in particular, whether or not data center devices (i.e.,
The Energy Intensity of the Internet: Home and Access Networks 9
data storage and server-type devices running in server rooms or data centers)
are viewed as part of the Internet. This decision has a large impact on the result:
As shown in Table 1, for 2008, which is referred to by several studies, the two
studies not including data centers result in energy intensities of 0.39 [24] and
0.179 kWh/GB [14] – factors of 18 and 39 below the 7 kWh/GB of the study
that includes data centers [23], respectively.
The original statistical data from [20] (used by both [11] and [16]) supports
this observation. While the consumption of core Internet devices was estimated
at just 1.25 TWh/a, the consumption of storage devices and servers together
was estimated at 11.7 TWh/a.
System Boundary: CPE and Access Networks. From the equipment in-
ventories in [16] and [11] (presented in Table 1 of each paper), it is clear they
do not consider customer premises equipment or access networks. This stands
to reason: both rely on 2000 inventory data for ICT in US offices, not in homes.
And the Internet had not yet exploded, reaching every home as it does today.
In 2000, ISPs, access networks, and customers’ modems and routers were not
nearly as prevalent or important as they later quickly became. [22] and [23], on
the other hand, explicitly include access networks and exclude CPE. Most of the
other studies account for both, except [27] and [29] which do not include CPE
in their assessment and [12] which does not consider the access network.
Not considering CPE or access networks, however, has a great impact on
the result. When the otherwise dominating data centers are excluded from the
calculation, numerous studies point out that CPE and access networks dominate
the energy intensity of the Internet over core and metro networks. Figure 2, for
example, shows the cumulated power consumption of the case study presented
in [32]. The peripheral parts of the network clearly dominate the overall power
Both [14] and [15] also point out that for the peak access rates typical of
2008 and 2010, respectively, access networks dominate the power consumption
of the Internet. Core and metro networks would become the dominating part
only if access rates were to grow extensively and reach peak access rates in
excess of 100 Mbps (for the energy efficiency of 2010 networking technology).
For networking technology that becomes more efficient along the foreseeable
trends, typical peak access rates would have to be over 1 Gbps for the core and
metro networks to surpass the power consumption of the access networks. For
the moment, as typical access rates are much lower, access networks use more
energy than the core of the Internet and their exclusion profoundly changes the
The consumption of customer premises equipment is within the same order
of magnitude as the access network but, as a general tendency, is slightly higher:
according to [14], the access network needs 2.8W of power, while the CPE needs
between 4-10W, depending on the technology used. [31] calculate with 2W per
subscriber for the DSLAM of the access network and with 10 W for the CPE
(5W for the modem and another 5W for the WiFi router). As with the access
10 V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
network, the decision of whether or not to include the CPE decisively influences
the result.
0 5000 10000 15000 20000 25000 30000
Cumulated power (excl. PUE) [W]
Distance from Davos [km]
Fig. 2. Cumulated power demand for the videoconference case study [33] along a dis-
tance of 27,000km. The power demand of local network components, albeit relatively
small, is allocated to a relevant extent to the case study’s traffic due to the low average
load of the local components. These local components therefore clearly dominate the
overall power consumption of the transmission. The large core Internet nodes and the
transoceanic “data highways,” while utilizing a relatively large amount of power, typi-
cally have a switching/transporting capacity on the order of hundreds of GB/s, or even
TB/s. Their contribution to the overall demand amounts to less than 1 Watt/Mbps in
the case study. By contrast, the power demand of transcontinental links does contribute
significantly to the overall consumption, due to their relatively low bandwidths.
Overheads: Cooling and Redundancy Equipment. The facilities (rooms,
buildings) hosting ICT networking equipment and datacenters induce a power
overhead due to non ICT-related consumption such as cooling or lighting. The
measure widely used to account for this consumption is the Power Usage Effec-
tiveness (PUE). The PUE is computed as a facility’s total power divided by the
power needed to run the ICT equipment only [34]. As the former includes the
latter, the PUE is larger than, or equal to, 1. The closer the PUE is to 1, the less
power is “wasted” for activities other than information processing. The average
PUE for datacenters nowadays is slightly lower than 2 [35] with a decreasing
tendency; for Internet routers it was around 1.7 in 2009 [36]. Most studies do
consider a PUE between 1.78 [29] and 2 [12, 14, 26, 32] in their calculations. No-
The Energy Intensity of the Internet: Home and Access Networks 11
table exceptions are [16] and [24] that do not account for the PUE, i.e., they
present results one would obtain for a PUE = 1.
Another overhead is induced by redundancy devices. Referring to routers,
[31] notes that “devices are operated with at least dual redundancy in order
to cope with failure.” [14], too, notes that a minimum of two uplinks are used
for redundancy in metro and core networks. Several of the bottom-up studies
(both model-based and case studies) account for the redundancy equipment via
a redundancy factor of 2 [12, 14, 29, 32, 31].
5 Discussion
As the system boundary plays such a decisive influence on the end result, this
section discusses possible best practices in defining the system boundary of the
Internet for energy-related studies. It then shows how the access network and
CPE differ from the more central parts of the network, and concludes with an
assessment of the energy intensity of the former.
5.1 System Boundary
End Devices. [30] argues extensively against the inclusion of end devices (both
user end devices such as personal or laptop computers, as well as servers and
storage in data centers) within the system boundary for the energy intensity of
the Internet. As we have shown above for data centers, including such devices
can dramatically change the results. This, however, is inadequate not only as a
matter of semantics; The concept of the Internet by definition does not include
end devices but only the infrastructure connecting them. There are also practical
arguments against including end devices, since that would yield results ill-suited
and potentially misleading on most questions:
The consumption in access, metro and core networks is largely independent
of both the end devices and the application generating the traffic. Mean-
ingful averages can thus be defined and estimated. By contrast, end devices
such as desktop computers, smartphones or web servers have very different
power demands. Different applications imply different sets of end devices:
web browsing a server at a data center and a desktop, laptop or tablet com-
puter; peer-to-peer file exchange two computers; and high-end videoconfer-
encing two large LCD screens in combination with codec devices.
Moreover, even for identical devices at both ends, distinct applications can
induce very different consumption levels per amount of transferred data.
While a peer-to-peer file exchange, for example, can use a bandwidth of
several Mbps, a Skype voice call gets by with a bandwidth of only 60 kbps.
Assuming exclusive usage of the two client devices, the low-bandwidth case
induces a much higher energy consumption per bit at the terminal nodes due
to the low utilization. This could lead to the seriously misleading conclusion
that “the Internet” uses more energy per amount of data for applications
12 V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
with lower bandwidth demand, in this case for a Skype voice call, than for
a highly demanding file exchange.
Under these circumstances, with different devices and different applications
inducing varying consumption in the end devices, it is unclear how these con-
sumption levels could be aggregated into meaningful averages. It seems more
meaningful always to assess network energy and the energy of end devices sep-
arately, and to add them up when needed – for example, for the assessment of
the energy needs of a specific service [32].
CPE and Access Networks. The consumption of customer premises equip-
ment and access networks, on the other hand, should always be considered.
Unlike end devices, these devices have no stand-alone meaning. They only exist
to connect end devices and thus semantically belong to the Internet.
Moreover, as shown in several studies [14, 15, 32], CPE and access network
dominate the energy consumption of the Internet over metro and core network.
Although this might not be true for services with a very high bandwidth usage
(see next chapter [13]) and might permanently change in the future with in-
creasing access rates [15], it is certainly the rule for the moment. It is advisable
always to include these factors, even if they make an equal contribution in the
cases under scrutiny. Although dropping such factors may simplify comparisons
such as in [12], it will also change the absolute values which, if taken out of
context, may lead to misunderstandings.
Overhead: PUE and Redundancy. The cooling and other types of overheads
included in the PUE, as well as the redundancy equipment, provide support or
safety functions in the Internet. Hence, their consumption should be accounted
As one of their few advantages over bottom-up assessments, top-down studies
inherently include the redundancy equipment, as this is included in the stock
inventories these are based on. Bottom-up analyses must account for them ex-
plicitly – as mentioned above, a factor of 2 for redundancy and a factor of 1.7-2
for the PUE are the values most studies use.
It must be noted, however, that both PUE and redundancy only apply to
access, edge, metro and core equipment. CPE have no redundancy nor are they
cooled. Whether redundancy and overheads are considered thus has a relatively
low influence on end results that include CPE, and there is a risk of overesti-
mation if one includes redundancy and PUE for the CPE as well (as we did in
Fibers. Whether the power of the amplifiers along the fiber cables is considered
(as it is in most studies) or not (as in [16, 24]), also has only a marginal impact
on the result. The relatively high power of dozens of kilowatts of transoceanic
fibers gets divided by such a large amount of traffic that the contribution per
The Energy Intensity of the Internet: Home and Access Networks 13
amount of data becomes negligible [32]. The consumption along fibers only be-
comes relevant for fibers with a low load such as the US transcontinental links
in [32]. Such a case leads to a considerable allocation to relatively low amounts
of traffic. Often, however, this is not the case. [14] argues that “the core opti-
cal transport (wavelength division multiplexed links) accounts for only a small
fraction of the total energy consumed by the Internet.” But even when relevant,
the consumption along the fibers was still smaller than the consumption of the
access network [32].
5.2 The Challenge of Defining the Energy Intensity of the Internet
A basic methodological problem with the energy intensity of the Internet is that
for some devices the energy consumption scales with the traffic volume and for
other devices it scales along different dimensions, especially time of usage [31,
30]. The former is true of most networking devices in the metro and core network
[31], while the energy consumption of devices in the access network and CPE (as
well as end devices, which have been excluded from the analysis) usually scales
with the time of usage and is largely traffic-independent [27].
To account for these differences, [31] recommends allocating the energy con-
sumption of a device based on an approach that takes into account the limiting
factor of the device – i.e., the factor that, if increased, would first limit the qual-
ity of service. If the limiting factor is in practice very hard to reach, the approach
allocates energy according to the dimension that, if changed, results in the most
significant change in energy consumption [31]. For the overall network energy
intensity, [30] thus suggests using a combined approach that encompasses inten-
sities defined as both energy per data and energy per time (i.e., power) where
appropriate. We will elaborate on this approach in subsection 5.3.
Existing studies, however, do not differentiate between categories of devices
when defining a metric for the energy intensity. Instead, each study defines the
energy intensity for all devices along the same dimension: Most studies define
the network energy intensity as energy per amount of data [16, 11, 22, 23, 29,32].
With a partial focus on CPE and access networks, [31] defines the energy inten-
sity for those devices as energy per time. Some studies with a focus on access
networks, noting that access networks’ devices are always on and their consump-
tion is thus both traffic- and time-independent, define the energy intensity as
energy per subscriber [26, 27]. Finally, top-down studies can avoid the problem
altogether by computing only the overall energy of the Internet and not relating
it to any other dimension for a measurement of energy intensity (e.g., [24]).
We conclude this chapter by putting forward a formula for the computation
of the energy intensity of the access network and customer premises equipment.
The next chapter in this volume [13] addresses the metro and core networks,
and proposes a formula for the energy intensity of those parts of the Internet.
That energy intensity is defined as energy per data. Summing these two leads to
the first formula for the energy intensity of the Internet that combines an energy
per time component with an energy per data component and thus models reality
more closely than previous work.
14 V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
5.3 The Energy Intensity of CPE and Access Networks
In this subsection, we develop the formula for the energy intensity of CPE and
access networks. We build on the analysis of [31], which analyzes the energy
consumption of online multimedia services. For the consumption in access net-
works and CPE, that article puts forward the following formula (formula 9 in
the article):
P U EN et(1)
EAN is the energy consumption in the access network (including CPE in
that article’s terminology) for the consumption of a given service,
tSthe time of service consumption,
PCP E and NCP E the power of all CPE taken together and the number of
users connected to them, respectively,
PT U and NT U the power of the access network devices and the number of
users connected to them, respectively, and
P U EN et the PUE of the DSLAM, which typically requires cooling.
We start from this formula to define the energy intensity per unit time (i.e.,
power) of customer premises equipment and access networks. For more clarity,
we consider access network and CPE separately and add them back together in
the end. As we are interested in the energy intensity and not the total amount
of energy, the time factor tsdisappears. Additionally, the energy intensity of the
Internet is the average value for one Internet communication and thus always
includes exactly one set of typical customer premises equipment (NCP E = 1) –
what “typical” means in this context will be addressed shortly.
With these observations, and renaming PTU ,NT U , and P U EN et to PAN ,
NAN , and pueAN for more consistency, the intensities of the access network and
the CPE, iAN and iCP E , are:
pueAN (2)
iCP E =PC P E (3)
The trivial formula 3, however, ignores one important aspect that was not
considered in [31] either: the energy used by CPE while idle, i.e., while not
providing any service. This energy has to be somehow distributed among the
services provided during a certain period [37]. We choose to distribute the idle
energy consumption among the services provided over a given period of time
proportionally to the time those services are active.
For the entire cycle over which meaningful averages can be built (i.e., a day,
a week, a year), we define
tOn the total time in which the equipment is on,
The Energy Intensity of the Internet: Home and Access Networks 15
tUse the total time in which the CPE is in use, i.e., in which it is used for
data transmission, and
tIdle the total idle time, when the CPE is on but not used (tI dle =tOn tU se).
The CPE consumes power for the time tOn but only provides services during
tUse . Distributing the entire power on the services provided during tUse needs
the factor tOn/tU se for extrapolation. With this, formula 3 becomes
iCP E =tOn
PCP E ,(4)
which, because tOn =tU se +tIdle , can also be written as
iCP E =1 + tI dle
tUse PC P E (5)
The energy intensity of the access network and CPE taken together is
iCP E &AN =1 + tIdle
tUse PC P E +PAN
pueAN =tOn
pueAN (6)
While such a generic formula is interesting from a theoretical point of view,
for added practical relevance, the formula should ideally be parametrized with
meaningful average values. These values should be based on the equipment and
usage patterns considered typical and will change over time.
For the access network, the fraction PAN/NAN represents the energy intensity
per subscriber. For ADSL2+, this was assessed as 3.4W by [14] and 2W by [31].
As [31] is the more recent analysis, we use its value. We further assume a PUE
of 2, following the majority of studies.
For the power of the customer premises equipment (PCP E ), a few older
studies consider only modems [14], while the more recent studies consider both
modems and WiFi routers [24, 31, 37]. We follow [24] who assumes that only a
few users use a modem without a WiFi router and that their number is compa-
rable to those with multiple WiFi routers or WiFi repeaters. This is equivalent
to assuming that 100% of users use both a modem and a WiFi router, either as
two separate devices or integrated into an Integrated Access Device (IAD). Tak-
ing into account the US distribution of IADs versus modems with WiFi routers,
[24] puts forward 7.1W as average CPE consumption for DSL and 9.5W for
cable. The November 2013 Energy Star requirements for small network equip-
ment [38] call for a base power of at most 5.5W for ADSL and 6.1W for cable
IADs, respectively, allowing another 0.8W for fast Ethernet and WiFi, and 0.5W
for the telephony functionality of DSL modems. Considering these numbers as
well, both ADSL and cable IADs are required to be just below 7W. Allowing
for the slightly higher consumption of two separate devices as well as for legacy
equipment, we use PCP E = 8W.
Finding data for the idle and usage times of modems is far more challenging
and is by far the greatest source of uncertainty. A 2011 BBC study [39] found
16 V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
that set-top boxes (the modems which deliver both cable TV and Internet con-
nectivity) are on for 15.57 hours/day, but it did not address their usage time. A
2007 study found that in Europe, DSL modems are idle for 20 hours/day and in
use for the remaining 4 hours/day [40]. Although this study is older, we use its
assumptions. The study distinguishes between the on and idle state of modems,
and we feel that the on-time of 24 hours/day from [40] better reflects reality
than the 15.57 hours/day from the 2011 study [39]. With these assumptions,
tOn/tU se = 6.
With all these specific values, formula 6 leads to a “currently typical” value
for the energy intensity (per time) of the access network and CPE of
iCP E &AN = 6 8W+ 2W2 = 52W(7)
The average value from formula 7 already includes the PUE of the access network
and the idle consumption of the CPE, and can thus be used for quick assess-
ments of energy consumptions in access networks and by consumer premises
equipment. As mentioned above, however, the formula has a relatively large de-
gree of uncertainty, especially due to the uncertainty of the idle time of the
6 Conclusion
We have shown that the energy intensity of customer premises equipment and
access networks has to be assessed differently from the intensity of metro and core
networks. We proposed a formula for the intensity of the former, both generically
and parametrized with typical data for 2014. The next chapter [13] complements
this work with a formula for metro and core networks. Taken together, the
two chapters provide an assessment method for the Internet energy intensity
that appropriately uses different allocation approaches for different parts of the
network. Parametrized with typical values for 2014, this method can be used by
practitioners for quick assessments of various Internet-based services.
1. Laitner, J.A.: Semiconductors and information technologies: The power of produc-
tivity. Journal of Industrial Ecology 14(5) (2010) 692–695
2. Mattern, F., Staake, T., Weiss, M.: ICT for green – how computers can help us to
conserve energy. In: Proceedings of e-Energy 2010, Passau, Germany, ACM (April
2010) 1–10
3. Hilty, L.M., Ruddy, T.: Sustainable development and ICT interpreted in a natural
science context: the resulting research questions for the social sciences. Information,
Communication & Society 13(1) 7–22
4. GeSI: Smart 2020: Enabling the low carbon economy in the information age. A
report by the climate group on behalf of the global esustainability initiative (gesi)
The Energy Intensity of the Internet: Home and Access Networks 17
5. European Commission: Addressing the challenge of energy efficiency through infor-
mation and communication technologies. Communication from the Commission to
the European Parliament, the Council, the European Economic and Social Com-
mittee and the Committee of the Regions 241 final, Brussels (2008)
6. Pamlin, D., Pahlman, S.: Outline for the first global it strategy for co2 reductions.
Technical report, World Wildlife Fund (2008)
7. Erdmann, L., Hilty, L.M.: Scenario analysis: Exploring the macroeconomic impacts
of information and communication technologies on greenhouse gas emissions. Jour-
nal of Industrial Ecology 14(5) (2010) 826–843
8. Hilty, L.M.: Information Technology and Sustainability. Essays on the Relationship
between ICT and Sustainable Development. Books on Demand, Norderstedt (2008)
9. Hilty, L.M., Coroama, V.C., de Eicker, M.O., Ruddy, T., Mller, E.: The role of
ICT in energy consumption and energy efficiency. Technical Report ICT-ENSURE:
European ICT Environmental Sustainability Research; call identifier FP7-ICT-
2007-2 (2009)
10. Coroama, V.C., Hilty, L.M.: Energy consumed vs. energy saved by ICT – a closer
look. In Wohlgemuth, V., Page, B., Voigt, K., eds.: Environmental Informatics
and Industrial Environmental Protection: Concepts, Methods and Tools. (2009)
11. Koomey, J., Chong, H., Loh, W., Nordman, B., Blazek, M.: Network electricity
use associated with wireless personal digital assistants. Journal of Infrastructure
Systems 10(3) (2004) 131–137
12. Baliga, J., Ayre, R.W., Hinton, K., Tucker, R.S.: Green cloud computing: Balanc-
ing energy in processing, storage, and transport. Proceedings of the IEEE 99(1)
(January 2011) 149–167
13. Schien, D., Coroama, V.C., Hilty, L.M., Preist, C.: The energy intensity of the
internet: Edge and core networks. In Hilty, L.M., Aebischer, B., eds.: ICT Innova-
tions for Sustainability. Springer, Heidelberg, Germany (2014)
14. Baliga, J., Ayre, R., Hinton, K., Sorin, W.V., Tucker, R.S.: Energy consumption
in optical IP networks. Journal of Lightwave Technology 27(13) (2009) 2391–2403
15. Hinton, K., Baliga, J., Feng, M.Z., Ayre, R.W., Tucker, R.S.: Power consumption
and energy efficiency in the Internet. IEEE Network 25(2) (March 2011) 6–12
16. Gupta, M., Singh, S.: Greening of the Internet. In: Proceedings of the 2003 Con-
ference on Applications, Technologies, Architectures, and Protocols for Computer
Communications (SIGCOMM ’03). SIGCOMM ’03, New York, NY, USA, ACM
(2003) 19–26
17. Teehan, P., Kandlikar, M.: Comparing embodied greenhouse gas emissions of mod-
ern computing and electronics products. Environmental Science & Technology
47(9) (2013) 3997–4003
18. Hischier, R., Coroama, V.C., Schien, D., Achachlouei, M.A.: Grey energy and
environmental impacts of ICT hardware. In Hilty, L.M., Aebischer, B., eds.: ICT
Innovations for Sustainability. Springer, Heidelberg, Germany (2014)
19. Chan, C.A., Gygax, A.F., Wong, E., Leckie, C.A., Nirmalathas, A., Kilper, D.C.:
Methodologies for assessing the use-phase power consumption and greenhouse gas
emissions of telecommunications network services. Environmental Science & Tech-
nology 47(1) (2012) 485–492
20. Roth, K.W., Goldstein, F., Kleinman, J.: Energy consumption by office and
telecommunications equipment in commercial buildings – volume i: Energy con-
sumption baseline. Technical report, US Department of Energy, Office of Energy
Efficiency and Renewable Energy, Washington DC (2002)
18 V.C. Coroama, D. Schien, C. Preist, and L.M. Hilty
21. Coffman, K.G., Odlyzko, A.M.: Growth of the Internet. In Kaminow, I., Li, T.,
eds.: Systems and Impairments. Volume IV B of Optical Fiber Telecommunications.
Academic Press (2002) 17–56
22. Taylor, C., Koomey, J.: Estimating energy use and greenhouse gas emissions of
Internet advertising. Technical report, IM C 2(2008)
23. Weber, C.L., Koomey, J.G., Matthews, H.S.: The energy and climate change
implications of different music delivery methods. Journal of Industrial Ecology
14(5) (2010) 754–769
24. Lanzisera, S., Nordman, B., Brown, R.E.: Data network equipment energy use and
savings potential in buildings. Energy Efficiency 5(2) (2012) 149–162
25. Cisco: Hyperconnectivity and the approaching zettabyte era. Technical report,
Cisco, San Jose, CA, USA (2010)
26. Baliga, J., Hinton, K., Tucker, R.S.: Energy consumption of the Internet. In:
Proceedings of the Joint International Conferences on Optical Internet and the
32nd Australian Conference on Optical Fibre Technology, COIN-ACOFT. (2007)
27. Baliga, J., Ayre, R., Sorin, W.V., Hinton, K., Tucker, R.S.: Energy consumption
in access networks. In: Proceedings of the Optical Fiber Communication Confer-
ence and The National Fiber Optic Engineers Conference (OFC/NFOEC), Optical
Society of America (2008)
28. Kilper, D.C., Atkinson, G., Korotky, S.K., Goyal, S., Vetter, P., Suvakovic, D.,
Blume, O.: Power trends in communication networks. IEEE Journal of Selected
Topics in Quantum Electronics 17(2) (2011) 275–284
29. Schien, D., Preist, C., Yearworth, M., Shabajee, P.: Impact of location on the
energy footprint of digital media. In: Proceedings of the 2012 IEEE International
Symposium on Sustainable Systems and Technology (ISSST), IEEE (2012) 1–6
30. Coroama, V.C., Hilty, L.M.: Assessing Internet energy intensity: A review of meth-
ods and results. Environmental Impact Assessment Review 45 (February 2014)
31. Schien, D., Shabajee, P., Yearworth, M., Preist, C.: Modeling and assessing vari-
ability in energy consumption during the use stage of online multimedia services.
Journal of Industrial Ecology 17(6) (2013) 800–813
32. Coroama, V.C., Hilty, L.M., Heiri, E., Horn, F.M.: The direct energy demand of
Internet data flows. Journal of Industrial Ecology 17(5) (2013) 680–688
33. Coroama, V.C., Hilty, L.M., Birtel, M.: Effects of Internet-based multiple-site con-
ferences on greenhouse gas emissions. Telematics and Informatics 29(4) (November
2012) 362–374
34. Rawson, A., Pfleuger, J., Cader, T.: Green grid data center power efficiency metrics:
PUE and DCIE. Technical Report White Paper #6, The Green Grid (2008)
35. Stansberry, M., Kudritzki, J.: 2012 Data Center industry survey. Technical report,
Uptime Institute (2012)
36. Moth, J., Norris, M.: G ´
EANT3 carbon footprint. In: Terena Networking Confer-
ence. (2010)
37. Schien, D., Shabajee, P., Wood, S.G., Preist, C.: A model for green design of
online news media services. In: Proceedings of the 22nd International Conference
on World Wide Web (WWW’13), Republic and Canton of Geneva, Switzerland,
International World Wide Web Conferences Steering Committee (2013) 1111–1122
38. EPA: Energy star program requirements for small network equipment. Technical
report, US Environmental Protection Agency (November 2013)
39. Chandaria, J., Hunter, J., Williams, A.: A comparison of the carbon footprint of
digital terrestrial television with video-on-demand. Technical Report WHP 189,
British Broadcasting Corporation, London, UK (March 2011)
The Energy Intensity of the Internet: Home and Access Networks 19
40. Nissen, N.F.: EuP preparatory study lot 6: Standby and off-mode losses – task
3: Consumer behaviour and local infrastructure. Technical Report TREN/D1/40,
Fraunhofer Institute for Reliability and Microintegration, IZM, Berlin, Berlin, Ger-
many (October 2007)
... We here mainly focus on the network. Note that some earliest works estimating the energy consumption of the Internet included datacenters while others did not, but it is now more common to consider them apart [6]. Early works [7], [8] were bottomup approaches. ...
... Baliga's model has inspired many later works [6], [11]- [14]. In particular, Hinton et al. [13] presented an extension to assess the energy consumption of optical networks for different services and scenarios. ...
Assessing the energy consumption or carbon footprint of data distribution of video streaming services is usually carried out through energy or carbon intensity figures (in Wh or gCO2e per GB). In this paper, we first review the reasons why such approaches are likely to lead to misunderstandings and potentially to erroneous conclusions. To overcome those shortcomings, we propose a new methodology whose key idea is to consider a video streaming usage at the whole scale of a territory, and evaluate the impact of this usage on the network infrastructure. At the core of our methodology is a parametric model of a simplified network and Content Delivery Network (CDN) infrastructure, which is automatically scaled according to peak usage needs. This allows us to compare the power consumption of this infrastructure under different scenarios, ranging from a sober baseline to a generalized use of high bitrate videos. Our results show that classical efficiency indicators do not reflect the power consumption increase of more intensive Internet usage, and might even lead to misleading conclusions.
... Indeed, the packet traffic analysis presented in this work demonstrates that the frequency at which the local infrastructure connects with the cloud infrastructure, the volume of data generated and transmitted, and the protocol overhead of transmissions in regular and/or user-driver operations are all fundamental aspects to be considered in the modeling of reference flows and impact estimation of full IoT systems. This last aspect warns Life Cycle Assessment (LCA) practitioners from be cautious about the postures of Malmodin et al. [31] and Coroama et al. [60] who suggest that, for estimating the electricity intensity of ICT (which is different from estimating its ecological impact), end devices-such as sensor systems-should be consider only from a use time perspective. ...
Full-text available
As we reach the limits of our current technologies and the number of connected devices grows, scientists put more efforts to estimate and reduce the ecological damage of the Internet of Things. Unfortunately, recent literature related to Life Cycle Assessment and eco design of IoT systems suffers from a major inconvenience so far: it does not put sensor data in the focus of attention. This paper aims to point out explicitly the essential role of this aspect for modeling reference flows, and demonstrate its relevance for appropriate environmental assessment and practical eco design. Also, it aims to illustrate that such modeling process must happen in a comprehensive way. For this, our work relies on a case study addressing smart metering, and we proceed as follows: based on available documentation and inspired by certain aspects of different technologies, we deduce the maximal capacities of key electronic components and construct an unfavorable data flow scenario, to get a rough idea of the reference flow and the long-term impact of our system during its use phase. Results from this procedure are later contrasted with results obtained from a packet traffic analysis, in which local and internet data flow are examined carefully. At the end, we verify the importance of sensor data theoretically and empirically, and we conclude that the reference flow and the impact contributors of a system could be affected not only by the local data transit but also by the complex interactions between edge devices and cloud resources. All our findings are discussed to nourish the state-of-the-art around the environmental impact of using full IoT systems and their sustainable design.
... Green technologies and ICT solutions improve energy efficiency (Chourabi2012; Chavan & Chavan, 2020) and achieve targeted environmental governance through realtime monitoring of the entire city. The use of technology facilitates applications, such as e-commerce and e-banking that enable remote working and online transactions instead of physical mobility, resulting in less traffic congestion, lower energy consumption and less waste, i.e. less travel time and printed marketing materials (Coroama et al., 2015;Romm, 2002). At the same time, internet use enables higher levels of education and stimulates civic participation. ...
Full-text available
Smart cities are rapidly gaining momentum around the world as new opportunities to shift towards a sustainable urban future. Despite their promising solutions, there is so far little evidence on how smart city initiatives can achieve urban sustainability. This study explores the impact of smartness on urban sustainability by introducing the concept of smart complementarity and investigating whether the systemic governance of smart complementarities has a positive impact on sustainability. To this end, we employ a panel vector autoregression model with panel data for 26 smart cities from 2009 to 2019 to test the complex dynamic relationships between smart city drivers and dimensions of sustainability. Such analysis seems crucial for smart city governance and whether the vision of a smart agenda is compatible with urban sustainability. The findings suggest that while smart city initiatives have improved economic sustainability, they have compromised environmental and social sustainability. This points to the need for a coordinated approach to address the complexity of urban systems and sustainability challenges.
... With a similar approach, Costenaro et al. [32] found the same estimation to be 5.12 kWh/GB. In their works, Schiem et al, proposed a model that takes into account users' devices and calculated that a gigabyte of transmitted data implies the consumption of 0.052 kWh [31,70]. Coroama et al. directly measured the energy consumption of transmitting data through the network by using real-world data and found an overall cost of 0.1993 kWh / GB [30]. ...
Full-text available
In the past decade, global warming made several headlines and turned the attention of the whole world to it. Carbon footprint is the main factor that drives greenhouse emissions up and results in the temperature increase of the planet with dire consequences. While the attention of the public is turned to reducing carbon emissions by transportation, food consumption and household activities, we ignore the contribution of CO2eq emissions produced by online activities. In the current information era, we spend a big amount of our days browsing online. This activity consumes electricity which in turn produces CO2eq. While website browsing contributes to the production of greenhouse gas emissions, the impact of the Internet on the environment is further exacerbated by the web-tracking practice. Indeed, most webpages are heavily loaded by tracking content used mostly for advertising, data analytics and usability improvements. This extra content implies big data transmissions which results in higher electricity consumption and thus higher greenhouse gas emissions. In this work, we focus on the overhead caused by web tracking and analyse both its network and carbon footprint. By leveraging the browsing telemetry of 100k users and the results of a crawling experiment of 2.7M websites, we find that web tracking increases data transmissions upwards of 21%, which in turn implies the additional emission of around 11 Mt of greenhouse gases in the atmosphere every year. We find such contribution to be far from negligible, and comparable to many activities of modern life, such as meat production, transportation, and even cryptocurrency mining. Our study also highlights that there exist significant inequalities when considering the footprint of different countries, website categories, and tracking organizations, with a few actors contributing to a much greater extent than the remaining ones.
... It was also observed that the data and information flow managed by specific capacities of sensing and edge devices, under specific operational parameters (e.g., transmission bandwidths, spreading factors or distance range) are as relevant as the use time for environmental impact estimation. This finding challenges the conclusions of Malmodin, J. et al. [11] and Coroama, V. C. et al [249], who claim that for end-user ICT equipment, the use time is more relevant because the energy consumption and embodied carbon footprint is not to the same extent related to transmitted data volume. ...
The accelerated adoption of the Internet of Things by our modern societies has increased significantly the production of connected devices and data in recent years. In the face of the potential impacts of this tendency, researchers put more efforts on measuring the environmental impact of IoT systems, proposing tools to reduce this impact and offering innovative solutions. However, Life Cycle Assessment (LCA) literature focused on IoT systems shows that few authors cover the full architecture. On the other hand, the eco design tools found in literature suffer from shortcomings and some of the most innovating solutions are projected promising, but also can cause collateral damage. Besides of all this, the research on impact estimation struggles with the absence of LCA data, and practice of eco design is hampered by the impracticability of applying exhaustive LCA modeling, within the typical design workflow of devices. It is in this context that this thesis aims to build a practical design methodology oriented to estimate the environmental impact of full IoT systems, and minimize this impact from the early steps of the development of new prototypes. To achieve this goal, this work starts from the idea that substantial information for an IoT application can be obtained from the efficient collection and organization of sufficient, yet meaningful raw data. In this manner, this thesis is developed on the basis of two points of reflection. The first one establishes two inexorable and indissociable concepts “function-capacity” that facilitate the definition of reference flows. Based on that, a framework for impact estimation is built. The second one promotes the approach of “right-provisioned-devices” that guides the selection of suitable components under three interdependent criteria (physical, technical and circular), considering a preliminary design step of data and information flow. Based on that, another framework for eco design is built. Both frameworks complement each other and compose a unique methodology for the eco innovation of IoT systems, applicable from basic information available to designers. In this work, this methodology has been implemented and illustrated in two parts. Firstly, the framework for impact estimation was implemented by a bottom-up, transversal life cycle model, which aims to illustrate the theoretical and empirical estimation of the reference flow and long-term impact of an IoT system oriented to smart metering. Secondly, the framework for eco design was implemented and illustrated by a preliminary design step of data and information flow of a prototype of a self-powered EH sensor system developed at the System Division of CEA-Leti; and by a LCA-based evaluation step, that involves two of its versions. This work concludes with 22 guidelines that must be adopted with a critical and global approach. That is, they should be challenged, refined or complemented in the context of other case studies; and by using the proposed methodology in a continuous, coherent and automated manner, particularly with the adaptation of Information Systems.
... A growing body of research attempts to measure the ecological footprint of the Internet. Using various methods, these works provide estimations about the energy intensity of data centers (Koomey, 2008;Masanet et al., 2020), the core network (Coroama et al., 2015), fixed Internet data transmission (Aslan et al., 2018) and mobile network transmission (Pihkola et al., 2018). Due to the great complexity of networks and the heterogeneity of the data, most researchers acknowledge that the estimations vary greatly. ...
YouTube is currently the most widely used platform for music streaming. Users listen to music videos rather than watch them. This is environmentally suboptimal since video data require more energy than audio data to be hosted and transmitted. Why are consumers using a video platform to stream music? In this paper, we sketch a framework for analyzing digital practices as consumption practices and their transformation in the context of the ecological transition. We interviewed 29 online music consumers from varied backgrounds. Drawing on practice theory, we conceptualize online music use as a combination of sociotechnical configurations articulating listening devices, types of attention to music, and the social contexts of daily life. We analyze how different platforms, especially YouTube, are embedded in specific configurations. We first establish that configurations in which videos are actually watched are rare. Though users are aware of the carbon footprint of streaming, this representation does not inform their listening configurations. We describe three types of online music practices according to the role YouTube plays in, that correlate with music passion: YouTube can be framed as a free and open listening platform (especially to casual listeners), as an efficient soundtracking device in many contexts, as a useful complementary listening and music sharing device. The paper extends the literature on green consumption to digital consumption, analyzing relations to infrastructures in a regime of abundance, and contributes to the sociology of online music consumption, showing how platform choices are linked with music passion and embedded in social contexts.
... Die Autoren errechnen, dass der Energieverbrauch der gesamten IKT-Infrastruktur hierfür etwa 165 kg CO2e verursachte.Das Nachhaltigkeitsteam der Universität Zürich(Warland & Hilty, 2016) hat die THG-Emissionen einer vierstündigen Videokonferenz zwischen Zürich und Paris geschätzt und errechnete 1,2 kg CO2e pro Person, unter Berücksichtigung der Herstellung, des Betriebs und der Entsorgung der benötigten Netze und Endgeräte. Für ihre Berechnungen verwendeten sie Resultate zweier anderer Studien, welche den THG-Fussabdruck einer Videokonferenz in HD-Qualität auf 160-290 g CO2e pro Stunde schätzten(Coroamă et al., 2015;Hischier et al., 2015). einer Studie berechnetenOng et al. (2014) die THG-Emissionen einer 5-stündigen Videokonferenz-Sitzung mit insgesamt 4 Teilnehmern für das Jahr 2010. ...
Full-text available
The use of digital products and services has continued to increase, especially in recent years, due to the COVID-19 pandemic in both work and private life. For example, people are using video conferencing systems and cloud services more than ever to work from home, ordering more and more products online, and accessing an inexhaustible selection of videos and music titles through streaming platforms. As the use of digital products and services leads to profound changes in working and private life, the question arises to what extent these contribute to a reduction or increase in the emission rate of greenhouse gases and are thus rather an opportunity or a hurdle for the achievement of climate protection goals. Research to date shows that a differentiated approach is necessary here and that blanket estimates of the climate impact of digitalization are not helpful.
Smart cities have emerged to cope with economic and environmental challenges posed by modern-day complex city systems. This study hypothesized that overcoming such challenges with a smart city framework entails a systematic approach to increasing synergies and eliminating trade-offs between different smart city indicators. This study applied network analysis and used data from 33 smart cities worldwide for the period 2005–2019 to ascertain these synergies and trade-offs by introducing a modified framework of smart cities based on the community identification of interconnected smart city drivers. The findings of this study reveal that very few trade-offs exist in the network of smart city indicators compared to the identified synergies. Furthermore, the network statistics show that innovation, business sophistication, ICT infrastructure, and government effectiveness are key drivers of a smart city.
Full-text available
Internet access has reached ca. 60% of the global population, with an average individual spending > 40% of the waking life on the Internet. We assess the environmental impacts of digital content consumption against the Earth’s ecological budget, finding that web surfing, social media, video and music streaming, and video conferencing could consume on average ∼40% of the per capita carbon budget consistent with limiting global warming to 1.5°C, as well as 55% of the per capita carrying capacity for mineral and metal resources depletion and > 10% for other five impact categories. Electricity decarbonisation would mitigate the climate impacts of Internet consumption substantially, but other impacts due primarily to the mining activities linked to electronic devices would remain of concern. A synergistic combination of rapid decarbonisation, electronic devices’ energy efficiency improvement, lifetime extension, and recycling, and behavioural change is paramount to prevent the increasing Internet demand from hindering sustainable lifestyles.
Full-text available
Despite the fact that Information and Communication Technologies (ICTs) are responsible for only a small part of worldwide greenhouse gas emissions – current estimations attribute around 2 % of man‐made emissions to ICT – this sector is the one with the fastest growing emissions. As a result, there is an increasing concern about the environmental impact of ICT, especially the climate change potential induced by ICT‐related energy consumption. At the same time, there is a growing perception that ICT can also substantially reduce the environmental impacts of other sectors, in particular by increasing their energy efficiency. ICT can help all economic sectors to become more energy efficient – since ICT allows existing processes to be optimized or enables entirely new, more energy‐efficient processes. The energy that could be saved by ICT‐induced energy efficiency is estimated to be several times larger than the overall energy consumption of ICT itself. The European Commission recognizes this potential and hopes that Europe will go a long way toward achieving its target of 20 % greenhouse gas reduction by 2020 by deploying ICT for energy efficiency. The present study looks at the field spawned by these two main issues at the intersection between ICT and energy: ICT’s own energy consumption and ICT’s potential to induce energy efficiency across the economy. In its approach to these issues, the study looks both at today’s situation, as well as future opportunities and risks. The study discusses the following research questions: a) estimates of the current energy consumption of ICT, b) prospective future developments in this energy consumption, and c) future energy efficiency potentials induced by ICT in various economic sectors. ICT‐related energy efficiency potentials already realized are not covered by this study, because it is virtually impossible to retrospectively allocate advances in energy efficiency to the various changes that created them. Two methodologies have been used for the study: literature review and expert interviews. For the former, we have reviewed: i) recent quantitative studies (since 2005) with a focus on “ICT energy consumption”, “ICT for energy efficiency”, “Green I(C)T”, “ICT and climate change” or “ICT and sustainability” in general; ii) other documents describing projects, programs or initiatives aimed at reducing ICT‐related energy consumption or increasing ICT‐related energy efficiency; iii) Life Cycle Assessment (LCA) studies on ICT products and services, and iv) studies on the potential of smart power networks (smart grids). We decided to include the more specific topics iii and iv because they are not sufficiently covered by items i and ii. These sources were then evaluated and the relevant content structured according to relevance (Chapter 2), state of the art (Chapter 3), and research programmes (Chapter 5) in order to give the reader insight into the motivation driving this research area, current knowledge, and the focus of ongoing research, respectively. Chapter 4 presents the results of expert interviews based on a questionnaire which we developed to fill the knowledge gaps identified in the literature review. The aim of the interviews was to collect ideas beyond the current state of quantitative knowledge and to identify research questions ‐‐ not to do a representative survey. The experts were only asked about future developments (research questions b and c). For ICT’s future energy consumption (b), the experts were asked to estimate for different categories of technologies (such as “data centres”, or “embedded ICT”) how their respective global energy consumption totals would evolve in both a business‐as‐usual scenario (alongside the foreseeable technological, political, and market developments) and in an “energy‐optimistic” scenario, in which energy‐reducing measures would be rigourously applied. As for question c above (ICT’s potential for energy efficiency), the experts were presented with the possible application areas in which the deployment of ICT was expected to lead to better energy efficiency, and were asked to estimate their relative importance. Furthermore, the experts were asked to determine which ICT categories were relevant for inducing energy efficiency in other economic sectors. In our analysis of the current situation and the future potential of both “ICT energy consumption” and “ICT for energy efficiency”, three main results become evident: • A thorough overview of the state‐of‐the‐art literature for all three questions considered, with emphasis on the LCA methodology and including a formal definition of ICT‐related energy efficiency as well as a conceptual framework of the effects of ICT on energy efficiency. • An overview of existing research programmes, project clusters, and institutions involved in them, both in the EU and beyond. • The results of expert interviews regarding future ICT energy consumption and future applications of ICT for energy efficiency. In addition to comparing business‐as‐usual with energy optimistic scenarios, thus revealing where the largest energy‐saving potentials for ICT lie, the experts have also – as a novelty – related the consumption of individual technologies to their respective potentials for inducing energy efficiency. At a more detailed level, the results show that some application fields (such as “TVs and set‐top boxes”) are expected to drastically increase their energy consumption without contributing to energy efficiency in any way, while others (such as “embedded ICT”), although increasing their collective energy consumption as well, are expected to play a crucial role in energy efficiency across the economy. We hope that this fresh thinking will help to introduce a more differentiated view of ICT into the public discourse and political decision making.
Full-text available
Environmental assessments of digital services seeking to take into account the Internet’s energy footprint typically require models of the energy intensity of the Internet. Existing models have arrived at conflicting results. This has lead to increased uncertainty and reduced comparability of assessment results. We present a bottom-up model for the energy intensity of the Internet that draws from the current state of knowledge in the field and is specifically directed towards assessments of digital services. We present the numeric results and explain the application of the model in practice. Complementing the previous chapter that presented a generic approach and results for access networks and customer premise equipment, we present a model to assess the energy intensity of the core networks, yielding the result of 0.052kWh/GB.
Full-text available
Direct energy consumption of ICT hardware is only “half the story.” In order to get the “whole story,” energy consumption during the entire life cycle has to be taken into account. This chapter is a first step toward a more comprehensive picture, showing the “grey energy” (i.e., the overall energy requirements) as well as the releases (into air, water, and soil) during the entire life cycle of exemplary ICT hardware devices by applying the life cycle assessment method. The examples calculated show that a focus on direct energy consumption alone fails to take account of relevant parts of the total energy consumption of ICT hardware as well as the relevance of the production phase. As a general tendency, the production phase is more and more important the smaller (and the more energy-efficient) the devices are. When in use, a tablet computer is much more energy-efficient than a desktop computer system with its various components, so its production phase has a much greater relative importance. Accordingly, the impacts due to data transfer when using Internet services are also increasingly relevant the smaller the end-user device is, reaching up to more than 90% of the overall impact when using a tablet computer.
Full-text available
Assessing the average energy intensity of Internet transmissions is a complex task that has been a controversial subject of discussion. Estimates published over the last decade diverge by up to four orders of magnitude — from 0.0064 kilowatt-hours per gigabyte (kWh/GB) to 136 kWh/GB. This article presents a review of the methodological approaches used so far in such assessments: i) top–down analyses based on estimates of the overall Internet energy consumption and the overall Internet traffic, whereby average energy intensity is calculated by dividing energy by traffic for a given period of time, ii) model-based approaches that model all components needed to sustain an amount of Internet traffic, and iii) bottom–up approaches based on case studies and generalization of the results. Our analysis of the existing studies shows that the large spread of results is mainly caused by two factors: a) the year of reference of the analysis, which has significant influence due to efficiency gains in electronic equipment, and b) whether end devices such as personal computers or servers are included within the system boundary or not. For an overall assessment of the energy needed to perform a specific task involving the Internet, it is necessary to account for the types of end devices needed for the task, while the energy needed for data transmission can be added based on a generic estimate of Internet energy intensity for a given year. Separating the Internet as a data transmission system from the end devices leads to more accurate models and to results that are more informative for decision makers, because end devices and the networking equipment of the Internet usually belong to different spheres of control.
Full-text available
The direct energy demand of Internet data flows can be assessed using a variety of method-ological approaches (top-down, bottom-up, or hybrid/model based) and different definitions of system boundaries. Because of this diversity, results reported in the literature differ by up to two orders of magnitude and are difficult to compare. We present a first assessment that uses a pure bottom-up approach and a system boundary that includes only transmis-sion equipment. The assessment is based on the case study of a 40 megabit per second videoconferencing transmission between Switzerland and Japan, yielding a consumption of 0.2 kilowatt-hours per transmitted gigabyte for 2009, a result that supports the lowest of the existing estimates. We discuss the practical implications of our findings.
Conference Paper
The use of information and communication technology and the web-based products it provides is responsible for significant emissions of greenhouse gases. In order to enable the reduction of emissions during the design of such products, it is necessary to estimate as accurately as possible their carbon impact over the entire product system. In this work we describe a new method which combines models of energy consumption during the use of digital media with models of the behavior of the audience. We apply this method to conduct an assessment of the annual carbon emissions for the product suite of a major international news organization. We then demonstrate its use for green design by evaluating the impacts of five different interventions on the product suite. We find that carbon footprint of the online newspaper amounts to approximately 7600 tCO2e per year, of which 75% are caused by the user devices. Among the evaluated scenarios a significant uptake of eReaders in favor of PCs has the greatest
In this study, we use an improved, more accurate model to analyze the energy footprint of content downloaded from a major online newspaper by means of various combinations of user devices and access networks. Our results indicate that previous analyses based on average figures for laptops or desktop personal computers predict national and global energy consumption values that are unrealistically high. Additionally, we identify the components that contribute most of the total energy consumption during the use stage of the life cycle of digital services. We find that, depending on the type of user device and access network employed, the data center where the news content originates consumes between 4% and 48% of the total energy consumption when news articles are read and between 2% and 11% when video content is viewed. Similarly, we find that user devices consume between 7% and 90% and 0.7% and 78% for articles and video content, respectively, depending on the type of user device and access network that is employed. Though increasing awareness of the energy consumption by data centers is justified, an analysis of our results shows that for individual users of the online newspaper we studied, energy use by user devices and the third-generation (3G) mobile network are usually bigger contributors to the service footprint than the datacenters. Analysis of our results also shows that data transfer of video content has a significant energy use on the 3G mobile network, but less so elsewhere. Hence, a strategy of reducing the resolution of video would reduce the energy footprint for individual users who are using mobile devices to access content by the 3G network.
The direct energy demand of Internet data flows can be assessed using a variety of methodological approaches (top-down, bottom-up, or hybrid/model based) and different definitions of system boundaries. Because of this diversity, results reported in the literature differ by up to two orders of magnitude and are difficult to compare. We present a first assessment that uses a pure bottom-up approach and a system boundary that includes only transmission equipment. The assessment is based on the case study of a 40 megabit per second videoconferencing transmission between Switzerland and Japan, yielding a consumption of 0.2 kilowatt-hours per transmitted gigabyte for 2009, a result that supports the lowest of the existing estimates. We discuss the practical implications of our findings.
There is a growing consensus that ICT can contribute to the reduction of anthropogenic greenhouse gas (GHG) emissions, both by increasing the efficiency of existing processes and by enabling substitution effects to usher in more energy efficient patterns of production and consumption. While, however, many studies based on theoretical reduction potentials have been presented, in practice it has only been possible to cite a few examples of such reductions thus far.This article presents the results of a field experiment for one particular domain in which ICT can be substituted for more carbon-intensive technologies: using advanced videoconferencing technology to reduce intercontinental conference travel and thus travel-related GHG emissions. We organized a large resource management conference simultaneously on two continents and assessed the emissions caused by the attendees’ travel and by the additional ICT equipment utilized to connect the two venues. We further assessed, based on a survey, the emissions in the alternative scenarios of holding the conference at either one of the places, and the satisfaction of the participants with the two-site conference format.The results show that reductions of 37% and 50% in travel-related GHG emissions were attained as compared to the single-site alternatives, although more people took part than in any of these alternatives. At the same time, the attendees’ experience was clearly positive, showing that the multiple-site format can serve as an acceptable alternative to the traditional one-site format of holding an international conference.