IEEE Circuits and Systems Magazine - Q1 2023 - 67

nodes that a certain video streaming service exploits,
as was proposed in [33]. Network nodes include the
main network components such as Internet exchange
points (IXPs), local area network (LAN) routers, access
networks, switches, repeaters, home routers, mobile
base stations (4G/5G), etc. A thorough analysis of the
types of network nodes and their power characteristics
is beyond the scope of this work but can be found in the
literature [1], [33], [80].
Similar to end-user devices, the node-wise energy
consumption again depends on the hardware of the
node, the configuration, the software, the software configuration,
and the type of network (wired, wireless). Estimations
and analyses on the energy consumption of
such nodes can be found in the literature [1], [33], [80],
[81], but are not further discussed here.
For energy consumption modeling, we adopt an approach
from a recent study [1], which proposed a simplified
view based on transmission requests. In this
approach, the network nodes are not considered separately,
but the overall energy consumption of all nodes
concerned by a single streaming request is used as a
basis. As this approach can elegantly be mapped to the
modeling of end-user devices and provider-side servers
discussed above, we adopt it for modeling the energy
consumption of the network as
EE Ess s,, ,, ,=+ NW CDN,
NW
NW UT
(15)
where Es,NW,UT is the energy consumption of all network
nodes caused by end-user requests and Es,NW,CDN is the
energy consumption of all network nodes caused by
copying of videos to surrogate servers in a CDN.
The energy consumption caused by end-user requests
is then modeled based on [1] as follows:
E
s NW UT =
,,
∑ ∑

dD∈ s rRsd∈ ,,



UT
()⋅
sdrs dr rr
NW UT
,, ,, ,, ,, ,,
NW UT rate
pp bt
0+⋅ , (16)




where we sum over all streaming requests r in the set
Rs,UT,d from all devices d in the set Ds, where the same
sets as already defined in Eqs. (3) and (4) are reused,
respectively. The energy of a single request is then
given as the product of the duration of the request
tr and the transmission power, which is the sum of a
constant offset power ps,NW,d,r,0 and a variable power
ps,NW,d,r,rate in


W
bps

.
The constant offset power corresponds
to the idle power of network devices and
the variable power depends on the bitrate br of the
requested video [1].
FIRST QUARTER 2023
ps,NW,VP,σ,r,rate
Table 4.
Exemplary power parameters for data transmission.
All values are taken from [1].
ps,NW,UT,d,r,0
ps,NW,UT,d,r,rate
Fixed BB access
1.5 W
0.03 W
Mbps
Optical Link
0.1 W
Mbps
Mobile BB Access
(4G)
0.2 W
0.03 W
Mbps
Atlantic cable
0.05 W
Mbps
IEEE CIRCUITS AND SYSTEMS MAGAZINE
67
Likewise, the energy consumption of data transmission
caused by copying of videos in CDNs can be modeled
as [1] and [17]
E ,,s NW CDN =
∑ ∑

σ∈Σs rRs VP


 ()⋅
pp bt
sr sr rr ,
NW VPσσ0+⋅
,, ,, ,, ,, ,,rate
NW VP
∈ ,,σ copy
,




(17)
where we sum over all copy requests r in the set
Rs,VP,σ,copy and all servers σ in the set ∑s, which are the
same sets as defined in Eqs. (6) and (9), respectively.
Power values for modeling the network power consumption
are listed in Table 4. The top two rows report
values for two types of network connections for end-user
devices, namely a fixed broadband (BB) access network
and mobile BB access network. The bottom row
reports power values for high capacity optical links and
an Atlantic optical submarine cable for long-distance
data transmission. All values are taken from [1], where
it is reported that the distance of data transmission can
be neglected.
E. Model Validation
The model developed in this article is difficult to validate
in general. To still obtain acceptable estimates, we
constructed the proposed energy consumption model
based on existing models for the subsystems, which
were reported in the literature and that have already
been validated. Nevertheless, considering the rapid
technological progress, the reported model parameters
could already be outdated. Still, the estimated energy
values for the overall energy consumption of video services,
which are derived in the next section, provide
first insights into the importance of this high-level modeling
approach.
IEEE Circuits and Systems Magazine - Q1 2023

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