IEEE Circuits and Systems Magazine - Q1 2023 - 70

introductory example, we consider a system engineer
working for a provider of freely accessible videos
(tubes). His task is to choose optimal encoder settings
for the lowest overall energy consumption. A new amateur
movie called " The Lord of the Streams " is uploaded
and shall be posted on the server. The video has a duration
of two hours and is recorded with a high spatial and
temporal resolution (4K at 60 fps). The system engineer
now needs to choose a suitable encoder. As we will see,
the main difficulty of a good choice is that in advance,
it is unknown how many end users will choose to watch
the video, i.e., the number of devices Ds requesting
the video is unknown beforehand.
When offering the video through the service, the system
engineer now has to answer the following questions:
■ Which codec and bitrate to choose?
■ How much
energy
(ps,VP,σ,trans,r,enc = ?)?
■ Is it required to provide multiple streams at
multiple bitrates coded in different codecs
(?)?
V,, ,, =
sr
VP trans
σ
■ Is it required to store the video on surrogate servers
and if so, on how many (?)?
Σs =
For an illustration of the impact of these choices, we
analyze the impact of two parameters: The encoding energy
and the number of requests. From Table 3, we can
choose two different encoding energy values. As indicated
in [18], the encoding energy has a significant impact
on the bitrate because the bitrate for high-energy
encoding is reported to be 2.5 times smaller than that
for low-energy encoding. Hence, the two options are
1) psr
VP transenc = 200 mJ
,, ,, ,
σ
VP transenc
,, ,, ,
σ
2) psr = 90 svideo
kJ
svideo
and Br = 80 GByte.
and Br = 32 GByte.
Using the same parametrization as for the on-demand
video service in the last subsection, Fig. 5 shows
the estimated overall energy consumption for this particular
video depending on the number of requests
∑ ∈dD sd
s R ,,
s R ,,
UT .
For a small number of requests (lower than
∑=∈dD sd
UT ,),
1 000 it is apparently beneficial to
choose the encoder with the low encoding energy
consumption (red line) instead of the encoder with
the high energy consumption (blue line), because
the overall yearly energy consumption is much
smaller. However, for a large number of requests
(, ),
∑>∈dD sd 10 000 the red line crosses the blue
s R UT
,,
line such that energy-intensive encoding is preferable
for a reduction of the overall energy consumption.
The reason is that for this high number of requests,
the energy savings due to the smaller number of bits
to be transmitted to millions of users are higher than
the encoding energy consumption.
70
IEEE CIRCUITS AND SYSTEMS MAGAZINE
Table 6.
Estimated overall energy consumption and GHG
emissions for three different encoding configurations.
Encoder
High energy
Low energy
Optimal
Overall Es
52.8 GWh
46.7 GWh
43.2 GWh
GHG emissions
18,480 t CO2E
16,345 t CO2E
15,120 t CO2E
To give an example, assuming one million requests,
31.8 MWh can be saved with the energy-intensive encoder,
which corresponds to GHG emissions of roughly
11.1 tCO2E (at 350 gCOE
kWh
2
). On the other hand, if there is
only a single request for a video, one can save 191 kWh
by choosing the low energy encoder, which corresponds
to GHG emissions of roughly 66.8 kgCO2E.
With this information, we can take this example
one step further. Assume that the system engineer
needs to handle 50,000 videos, where 10 videos are
requested one million times and the remaining videos
just once. We consider three cases: The system
engineer can first encode all videos with the highenergy
encoder, second with the low-energy encoder,
and third, he can choose the energetically optimal
encoder for all videos. The resulting overall energy
consumption values and the corresponding GHG
emissions are summarized in Table 6.
The results show clearly that by optimally configuring
the encoder, potential overall energy savings of almost
10% can be achieved, if a low-energy encoder was
used before. Comparing the optimized overall energy
consumption with the case of high-energy encoding, the
savings even exceed 20%.
FIRST QUARTER 2023
to
spend
on
encoding
Figure 5. Yearly energy consumption (vertical axis) caused
by a single video depending on the number of requests
(horizontal axis). The blue line corresponds to high encoding
energy and low bitrate, the red curve corresponds to low
encoding energy and high bitrate.
IEEE Circuits and Systems Magazine - Q1 2023

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