IEEE Circuits and Systems Magazine - Q1 2023 - 69

values highly depend on the chosen parametrization
of the online video services. Hence, we cannot conclude
that a certain service is more energy intensive
than another.
The overall energy consumption Es
is then split
up into the energy consumption caused by the enduser
devices Es,UT, the DCs on the provider side Es,VP,
and the transmission networks Es,NW (cf., Table 5). We
can see that the main energy consumption for ondemand
video, IPTV, and teleconferencing services
is caused by end-user devices, which account
for
97%, 95%, and 95% of the total energy consumption,
respectively. For the social network, the energy consumption
of the video providers is the most important
component (98%). It is worth mentioning that the
energy consumption of the transmission networks is
rather small for all services (below 100 GWh) and in
each case corresponds to less than 5% of the total
energy consumption.
For a profound discussion and analysis, we further
split up the energy consumption down to the device
level. On the end-user side, we study the impact of
the type of end-user device on the energy consumption,
where we consider smartphones, tablet PCs,
laptops, TVs, and desktop PCs as listed in Tables 1
and 2. Hence, using Eq. (3) we take subsets from the
set of devices Ds, where each subset only contains
one type of device. The resulting energy consumption
values are illustrated in Fig. 3, where missing
bars indicate that in our parametrization, the service
is not requested from the corresponding device (cf.,
Appendix B). We can see that most energy is consumed
by TVs and desktop PCs, which is expected
because their power consumption during operation is
the highest. If devices with lower power consumption
are used (e.g., smartphones or tablet PCs for a social
network or teleconferencing), also the overall energy
consumption is decreased by at least one order of
magnitude. Hence, a significant amount of energy can
be saved by simply switching from power-intensive
desktop PCs or TVs to energy-efficient devices such
as laptops and tablet PCs.
Considering the energy consumption on the provider
side (VP), we split up the total energy consumption of
the servers by focusing on the server tasks as shown
in Eq. (7). Accordingly, Fig. 4 illustrates the energy consumption
corresponding to the offset, transmission and
reception of videos, copy of videos in CDNs, decoding
and encoding of videos, and storage of videos on surrogate
servers. Missing bars indicate that the task is not
performed by the corresponding service.
We can see that for on-demand video streaming and
IPTV, the energy required to send the videos (Tx) is the
highest. For the social network, the energy consumption
of encoding is crucial, which is also confirmed in the literature
[18]. For the teleconferencing service, as most
processing is performed at the end-user side, the main
power consumption is related to the offset energy of the
servers.
B. Case Study for Energy Optimization
Next to facilitating overall estimates, as shown above,
the proposed model allows the optimization of the
overall energy efficiency on the video level. As an
Figure 3. Energy consumption from the end user's point of
view depending on the service (color) and end-user device
(horizontal axis). The vertical axis shows the yearly energy
consumption in Wh and is plotted on a logarithmic scale.
FIRST QUARTER 2023
Figure 4. Energy consumption from the video provider's
point of view depending on service (color) and tasks (horizontal
axis). The vertical axis shows the yearly energy consumption
in Wh and is plotted on a logarithmic scale.
IEEE CIRCUITS AND SYSTEMS MAGAZINE
69
IEEE Circuits and Systems Magazine - Q1 2023

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