IEEE Circuits and Systems Magazine - Q1 2023 - 60

research directions targeting an increased energy efficiency.
Finally, Section V concludes this article.
II. Global Energy Consumption in Online Video
In this section, we first discuss how the energy consumption
of online video services ultimately leads to climate
change in Section II-A. Afterward, in Sections II-B to II-E,
we provide an overview of the current literature on the
energy consumption and energy modeling for providerside
data centers, transmission networks, and end-user
devices as shown in Fig. 1.
A. Energy, Greenhouse Gases, and Climate Change
In science, it is well accepted that GHG emissions cause
climate change, which is supported by many scientific
studies, simulations, and surveys [19], [20], [21]. In this
respect, carbon dioxide (CO2) is generally considered to
be the most important factor, however, also other gases
such as methane and nitrous oxide need to be considered
[20]. To simplify considerations, a unified metric
was proposed, which translates all GHG emissions
caused by any gas to the so-called CO2-equivalents,
which, in the following, will be referred to by the unit
[CO2E] [22].
During the use of online video, the resulting GHG
emissions are caused by the power consumption of
all involved devices. In the literature, the conversion
from power to emissions is usually done using the CO2
or carbon intensity of energy production, which is the
amount of CO2 emissions in grams per energy unit
gCOE
kWh


2
power [23]. This unit mainly depends on the type of
power plant used to generate electricity. For example,
modern wind parks have a very small CO2-intensity that
is close to 0 gCOE
have a very high intensity of more than 350 gCOE
kWh
2
kWh
2
Average carbon intensities for various countries can be
found online [23], [25].
B. Data Centers
There is a large body of literature on the energy consumption
and the processing efficiency of data centers
(DCs) [12], [26], [27], [28], [29], [30], [31]. Most of this research
targets general DCs that provide different kinds
of services such as cloud services, storage, processing
power for complex calculations, block chains, and so on.
Various studies have shed light on the overall power
consumption of nation-wide DCs, where detailed data
is available especially for the USA [12] and Europe [26],
[27]. The main findings from these studies are that the
overall power consumption of DCs is constantly increasing
over the years and that throughout the last decade,
60
IEEE CIRCUITS AND SYSTEMS MAGAZINE

, where the energy is the time integral of the
the power consumption of all DCs had a global share of
the total energy consumption of 1% to 2%. In these studies,
the focus lies on the power consumption of the main
hardware components, i.e., servers, storage units, and
networking devices. Additionally, the power consumption
related to infrastructure such as lighting and climatization
is also considered [27], [28].
To determine the energy efficiency of these DCs, the
most common approach is to use the so-called powerusage
efficiency (PUE) or its inverse, the data center
infrastructure efficiency (DCE) [28], [29], [30]. These figures
of merit describe the relation between the power
consumption of all computationally active components
of a DC and the total power consumption including management
and climatization. DCs are energy efficient if
their PUE is close to one, which means that almost all
the power is consumed by active components. Uzaman
et al. [29] report that today, the DCs of the big tech companies
have PUEs between 1.08 and 1.45. As the PUE
does not provide information on the true power consumption
and the related GHG emissions, further studies
investigated the actual GHG emissions in detail [31].
An exhaustive overview of energy and power modeling
approaches for DCs is given in [32]. In this survey,
power models are listed for many different components
such as processors, storage, and software. Also, the
total power consumption of an entire DC is discussed.
For our proposed generic power consumption model for
online video applications, in Section III, we will select
appropriate models from the aforementioned papers
[12], [26], [27], [28], [29], [30], [31], [32] while taking into
account the PUE.
. In contrast, coal-fired power plants
[24].
C. Transmission Networks
Since the beginning of the '90s, the energy consumption
of communication networks has been investigated in
detail [1], [13]. Power and energy consumption data are
available for many countries, operators, and technologies
[1], [33], [34], [35]. In general, we may distinguish
between the core network, in which data is transmitted
globally between Internet exchange nodes, and access
networks located close to the end user. For the former,
it is common to consider the number of hops, where
one hop represents the transmission from one Internet
node to the next. Examples of such exchange nodes
can be commercial Internet exchange points (CIX) or
regional routers [33]. Common to nodes in the core network
is that they can potentially serve any Internet user
worldwide.
Access networks, which are located close to the
end user, comprise fixed broadband access networks
and mobile access networks [1]. Both provide Internet
access to end users using either fixed lines (copper or
FIRST QUARTER 2023
IEEE Circuits and Systems Magazine - Q1 2023

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q1 2023
