IEEE Circuits and Systems Magazine - Q1 2023 - 71

The overall energy consumption of a single online video service
can be optimized by selecting adequate encoder settings.
This simple example also shows that the proposed
model allows for more sophisticated optimization. Many
encoders provide multiple so-called presets, which allow
choosing a tradeoff between the encoding energy
consumption and the bitrate of a video [18], [76], [77].
With these presets, the system engineer can optimize
the overall energy consumption depending on the expected
number of requests. Furthermore, if the number
of requests cannot be predicted accurately, timeadaptive
solutions could be developed, where in the
beginning, low-energy encoding is chosen and another
energy-intensive encoding step is performed if end-user
requests increase significantly.
In addition, we model the impact of providing multiple
streams for a single video, i.e., we study the impact
of Vsr
,, ,,VP trans
σ
on the overall energy consumption. As
system engineers, we can provide a set of bit streams
encoded from a single video, where each bit stream
is coded using different encoder settings. This might
be required to satisfy the needs of different end users
and their devices. The resulting set of bit streams can,
e.g., differ by the bitrate, the codec, or the resolution.
Let's assume that we encode sixteen bit streams that
were created by encoding at two different bitrates, four
different codecs, and two different resolutions. We
further assume that these bit streams are stored on
surrogate servers for availability to end users. We use
the same parametrization as above using high-energy
encoding and we assume to operate 10 surrogate servers.
We find that the overall energy consumption when
providing a set of bit streams with respect to the provision
of a single bit stream increases by a relatively
small percentage of 0.07% (assuming one million requests).
However, this percentage changes drastically
when the number of requests is lower. For example, if
we only assume 1,000 requests, the overall energy increases
almost sixfold when providing 16 streams instead
of a single stream. This example underlines the
importance of smart encoding decisions depending on
the number of user requests.
As a final example, if we scale the number of surrogate
servers Σs , the energy consumption needed for
maintenance of the CDN increases linearly. However, as
the energy to store a stream and to distribute a stream
to multiple servers is relatively small, the impact on the
overall energy consumption is rather small (below 5%
even for a low number of requests).
FIRST QUARTER 2023
C. Directions for Future Research
In the previous subsection, we have seen that the overall
energy consumption of a single online video service
can be optimized by choosing adequate encoder
settings. In general, this idea can be applied to other
tasks and systems involved in online video. To this end,
the collected dependence of the energy consumption
of online video systems on the main parameters used
for modeling has to be identified. An overview of such
dependencies is provided in Table 7. The table shows
which parameters influence the energy consumption of
the considered end-user devices, the networks, and the
server tasks on the provider side.
Further research targeting the energy efficiency of
online video services is summarized in Table 8. The table
shows that for end-user devices, networks, and servers,
a considerable amount of research was performed
to increase their energy efficiency. However, some topics
are still to be addressed, which are discussed in the
following.
First, we can find that many studies do not represent
the state of the art anymore because the investigated devices
and systems are outdated and often not used anymore.
Examples for such studies are [17], [43], [80], and
[91]. For future research, updates for these studies with
measurements on state-of-the-art devices and systems
would be highly interesting for the scientific community.
Concerning the end users, more research on different
types of end-user devices needs to be done. For example,
to the best of the authors' knowledge, there are
no detailed studies on the power consumption of TVs
performing online video streaming. Similar to studies
for smartphones or laptops [16], [42], [43], such studies
could produce more detailed information and opportunities
to further increase the energy efficiency of TVs on
the algorithmic level.
Considering server tasks on the provider side, first,
the energy efficiency of encoding could be analyzed in
more detail. Sophisticated energy models similar to
models available for the decoding energy [49] could be
developed and applied to reduce the encoder's energy
consumption. Second, from a high-level point of view, the
energy consumption of servers including transmission,
decoding, and storage could be addressed to increase
the overall energy efficiency of an online video provider.
Finally, a promising direction of research would be
the optimization of an entire online video service as
IEEE CIRCUITS AND SYSTEMS MAGAZINE
71
IEEE Circuits and Systems Magazine - Q1 2023

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