IEEE Circuits and Systems Magazine - Q1 2023 - 62

and tablet PCs instead of large TV screens or desktop
PCs is highly beneficial for saving power [1].
III. Energy Consumption Modeling
In this section, we construct an energy model by beginning
with the worldwide energy consumption E caused
by online video applications per year. The model covers
the major online video services including applications
such as:
■ on-demand streaming,
■ teleconferencing,
■ remote Desktop,
■ peer-to-peer (P2P),
■ content delivery networks (CDN),
■ surveillance,
■ Internet protocol television (IPTV),
■ social media,
■ messaging, and
■ cloud backup.
Note that we explicitly only consider the energy consumption
of services and devices related to direct and
live online streaming. That means that regarding the
recording of videos, we consider the case of a single
capturing device (i.e., end-user recording for teleconferencing
or social networks). In contrast, we do not consider
complex systems for live TV broadcast or movie
production. Furthermore, we do not take into account
the energy consumed for the production and disposal
of devices or for further related activities such as the
standardization efforts needed for the establishment of
transmission networks and video codecs.
In Section III-A, we define the model at the global level
and assign all devices needed during the use of online
video services to the global online video system, which
consumes the energy E every year. The energy on this
global level is then split up into the energy consumption
caused by the online video services mentioned above,
which corresponds to the viewpoint of a system engineer
of a single service. Afterward, in Sections III-B to
III-D, we develop more detailed energy models by considering
the end-user terminals, the networks, and the
provider's data centers separately. Finally, we discuss
how the model can be validated in Section III-E.
A. Model Definition
As practically, it is difficult to measure and describe
the global energy consumption E caused by all online
video services, we break this energy up into parts that
we can estimate using values reported in the literature.
To this end, we refer to the reports released by the
video service providers, which often provide statistical
data on the use of their services as, e.g., listed in [1].
62
IEEE CIRCUITS AND SYSTEMS MAGAZINE
The system engineer can then consider one service at a
time. Hence, we split up the worldwide energy consumption
as follows:
EEs
sS
=∑ ,
∈
(1)
where S is the complete set of online video services and
s the index of a single service. The energy Es is the energy
consumption caused by the s-th service per year. For
example, this energy represents the total energy consumption
caused by an on-demand streaming service,
a teleconferencing service, or a social media platform.
Then, the system engineer can further decompose
the service-level energy as follows:
Ess ssEE E=+ +,, ,
UT
VP
NW,
(2)
where Es,UT is the energy demand of all end-user terminals
using the s-th video service, Es,VP is the yearly energy
consumed by the service's DCs and servers, and Es,NW
the energy required for the transmission of the videos
via networks.
B. End-User Devices
In this section, we consider the total yearly energy consumption
of all end-user devices such as tablet PCs, TVs,
smartphones, and desktop PCs including their displays.
We sum up over all Ds end-user devices, which make use
of service s, as follows:
EEUT ,,, ,
∈
ss d
dDs
UT = ∑
(3)
where Es,UT,d denotes the energy consumption of the d-th
device for the s-th service. Next, note that each device can
request multiple streaming events, which are summed up
in the set of requests Rs,UT,d. The total energy consumption
of device d can then be calculated by summing up the energy
consumption of all requests r as follows:
EEUT
rRsd∈ ,,
sd = ∑
UT
,,
UT
Finally, each request r is characterized by the direction
of streaming. This means that video streaming can
be performed either by download and playback (Rx),
by capture and upload (Tx), or both can be done at the
same time. To cover these possibilities, we propose to
compute the request-wise energy consumption based
on [1], [16], and [43] as
Ep pp t()+
sdrs dr sdrs dr
,, ,,UT,, ,, ,, ,, ,, ,
UT Rx
UT =+ UT Tx ⋅ r ,
(5)
FIRST QUARTER 2023
sd r .
,, ,
(4)
IEEE Circuits and Systems Magazine - Q1 2023

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