IEEE Circuits and Systems Magazine - Q1 2023 - 75

There is a high potential and need for further research targeting
energy efficiency in online video, which is essential for sustainable
deployment of online video technology in the future.
energy consumption of online video services in Section
IV-A. For all services, we assume that there are 100
million users which employ Ds =100 million devices.
Each user makes use of the service one hour per day
on average (cf., Section IV-A). For the on-demand video
service, we assume that all videos have a duration
of two hours, i.e., tr = 7200 s. For the targeted average
use of an hour a day, this results in one request every
second day such that the number of requests per device
is the size of the set of requests Rsd
UT = 365 Fur,,
2
.
thermore, we assume that all requests are performed
from TV sets such that we take the power value for
TVs from Table 2. We assume that the service provides
Rs,, ,VP trans
σ
end users. These were encoded from Rs,, ,VP Rxσ
sr
VP trans
σ
=1000 different videos to be requested by
=1000
input videos, which means that here, we adopt the simplified
assumption that only a single version of each input
video is provided ().
V,, ,, =1 Furthermore,
half of these videos are stored on Σs −=1 999 surrogate
servers of the CDN because they are frequently
requested [17] such that Vs,, ,
VP store = 500 For the
σ
.
network connection, as TV sets are used, we take
transmission energy consumption values for fixed BB
access networks and we assume that complex software
transcoding is employed.
For IPTV, we make similar assumptions. The main differences
to on-demand services are that we assume a
higher bitrate ()br =10 Mbit
because due to real-time res
strictions,
encoding can only be done in real time such
that also the corresponding lower encoding power consumption
from Table 3 is employed. Furthermore, we assume
that IPTV is constantly broadcasting content such
that encoding of videos is performed 24 hours on 365
days of the year. Furthermore, we assume that the video
is not stored.
For the social network service, we assume that all
videos are recorded by end users. We assume that each
uploaded video has a duration of 5 minutes and that
each video is watched by 10 friends. When requesting
one hour of video in total per day (which means
that twelve videos are requested), the number of requests
for downloading videos per device and per
year is then Rsd
down =⋅ =
UT
,,
that all requests are performed from smartphones. As
each video is watched ten times, the corresponding
FIRST QUARTER 2023
365 12 4380 where we assume
,
number of uploaded videos is one-tenth of this number
given by Rsd
up =⋅ =
,,
UT /,365 12 10 438 such that on
average, each user uploads more than one video per
day. Furthermore, all uploaded videos are transcoded
to a standard codec and container format such that
RD Rss sd
VP trans =⋅ =⋅
up
,, ,
σ
UT
,,
videos are only requested a few times, they are only
stored on a single server (no copies are stored on surrogate
servers). Still, a network of Σs =1000 servers
ensures that all videos are stored close to the end users.
Encoding and decoding are done using software
implementations resulting in energy-intensive processing
(cf., Table 3) at a bitrate of br = 5 Mbit
. To take into
s
account that social network providers will probably
not use the most sophisticated and energy-intensive
encoding process (),
lower value of 1 kJ
90 kJ
svideo
svideo
mobile network.
For teleconferencing, we assume that end users use
different devices to participate in conferences. For simplicity,
we assume that the share of devices is evenly
distributed among smartphones, tablet PCs, Laptops,
and desktop PCs. Each end user is logged in one hour
per day in a single conference such that Rsd
UT = 365
,,
and tr = 3600 s. Furthermore, as high visual qualities
are not as crucial for teleconferencing as for on-demand
streaming, we assume that the resolution of the videos
is significantly lower than HD, which results in a reduced
bitrate of 2 Mbit
.
s
Finally, we consider the energy consumption of the
transmission network using values from Table 4. We assume
that the social network is accessed using a mobile
BB access link and the other services use fixed BB access.
The copying of videos via the CDN is taken into
account using the values from the fixed BB access link.
Acknowledgment
The authors would like to thank Wolfgang Heyn for
his profound literature research, where he collected
methods and data to describe the global energy consumption
of online video services. This work was supported
in part by the Deutsche Forschungsgemeinschaft
(DFG, German Research Foundation), under
Project 447638564.
IEEE CIRCUITS AND SYSTEMS MAGAZINE
75
we choose a significantly
. The network is chosen to be a
43 8109 However, as all
..
IEEE Circuits and Systems Magazine - Q1 2023

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