IEEE Circuits and Systems Magazine - Q1 2023 - 61

End users should use low-power devices for video streaming,
e.g., smartphones and laptops, instead of TV sets
and desktop PCs.
optical fiber) or mobile networks (2G/3G/4G/5G). To obtain
power consumption values for these networks, it
is common to consider the overall power consumption
of each network node involved in the transmission [1],
[33], [34], [36].
To quantify the energy efficiency of data transmission, a
common method is to determine the energy intensity, i.e.,
the energy consumption per transmitted bit in
power consumption per throughput in


W
bps
bit  or the
J

 [3], [33], [37],
where the unit bps represents " bits per second. " Here, the
power and the energy comprise the electromagnetic radiation
power plus the additional power needed to operate
the devices. Over the last decades, the energy and power
intensities have decreased by several orders of magnitude
[33]. For example, the earliest estimates suggested a power
intensity of more than 3 kW
only 36 W
Mbps before the year 2000, but
Mbps in 2016 [1], [38].
Recently, it was proposed to extend this simplified
view, which just sets the overall energy consumption
in relation to the transmitted data, by also taking into
account other factors such as the number of users, subscribers,
and lines [1]. It was argued that the traditional
approach neglected that most of the power consumption
is independent of the actual amount of transmitted
data. This power is referred to as a constant offset power
in the following, which is always consumed independently
of the usage of the network. For example, a base
station for a mobile network always consumes a certain
offset power, even if no subscriber is logged in. Although
additional power is consumed during data transmission,
this additional power is often much smaller than the
constant offset power [1]. Hence, it is argued that the
constant part of the power should be associated with
the number of subscribers instead of a user's data rate.
Using this approach, it was found that especially the
power consumption of high data rate applications (file
download and video streaming) is much lower than reported
before [1].
As a consequence, in Section III, we construct our
power model based on a separation of a bitrate-dependent
variable power and the constant offset power.
D. Content Delivery Networks
Certain online video services are operated using a socalled
content delivery network (CDN) [17]. The concept
of a CDN considers a network of spatially distributed
FIRST QUARTER 2023
DCs, which communicate over the Internet. A CDN is especially
useful if content shall be stored close to the end
users, for example, in on-demand video streaming [17].
A general assumption for CDNs is that there is a main
server storing all videos and that there are additional
surrogate servers hierarchically structured, which
store copies of the most requested videos [17]. Due to
additional management and regular updates of stored
videos, such CDNs consume additional energy, which is
discussed in detail in [17], [39], and [40].
E. End Users
The energy consumed by end users must be handled
differently than the energy consumed by DCs and transmission
networks. The reason is that DCs and transmission
networks usually serve a very large number of end
users (several thousands of end users for mobile networks
in cities or up to millions of end users for Internet
exchange nodes), where in contrast, end-user devices
usually only serve a small number of users (one to a
couple of end users), but several millions of end-user
devices must be considered.
Some overall estimates for the worldwide energy consumption
of end-user devices are given in [1] and [3]. In
these studies, the main devices are displaying devices
such as smartphones, tablet PCs, and television (TV)
sets including the display. Utilities such as routers and
local network switches, which were already discussed
in Section II-C, are also considered. Typically, it is taken
into account that routers are running continuously with
approximately constant power consumption, whereas
displaying devices only consume power during video
streaming [1]. Typical energy and power consumption
values for these devices can be found in device manuals
and were analyzed in detail in different studies [1], [10],
[16], [41], [42].
Earlier studies found that the total energy consumption
of all end-user devices performing online video
tasks is approximately the same as that of all DCs and
all transmission networks individually [3], [9]. However,
recent studies suggest that in fact, end-user devices
contribute most to the overall power consumption during
streaming [1], where it is taken into account that
cloud services support many more applications than
just online video applications. Hence, the use of power
efficient end-user devices, e.g., the use of smartphones
IEEE CIRCUITS AND SYSTEMS MAGAZINE
61
IEEE Circuits and Systems Magazine - Q1 2023

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