[E-impact] kWh/GB should be banned and the IETF should warn against its use

[Rudolf van der Berg <[e-mail]>]

Dom Robinson looped me in and I now joined the mailing list.

I was surprised to see that kWh/GB and equivalents as measures of
electricity use of networks and sometimes wider; datacenters and consumer
electronics, are seriously mentioned by some people on this list. kWh/GB is
one of the most misleading and scientifically incorrect ways of describing
the energy use of networks. Even a cursory glance of the academic
literature shows that estimates range so far and wide that there must be
something wrong with it. A slightly deeper analysis would show there is no
basis in physics, electrical engineering, optical and wireless networking,
protocol design or computer science for relating electricity to an
arbitrary selection of  groupings of 8 billion bits, modulated, transmitted
and received over a signal, while generally omitting large numbers of other
modulations of bits on the same signal.

Now I have only a degree in Public Administration, so I may not know
perfectly where and how energy use in networks is affected, but I do deal
with the fall out when others get it wrong, because it stimulates bad ideas
for regulation that would break the internet instead of save energy. So
this is a plea for some good data and metrics.

There are several problems with the metric:

1. Physics: The energy consumption of networking equipment is to a large
extent based on how far the signal needs to travel, spectral bandwidth and
the medium that it is transmitted over. When transmitted through copper
generally more energy is needed to achieve a distance and the signal fades
quicker than over fibre. In wireless networks the frequency and coverage
area affect energy use. Modulations can affect energy use, but to a much
lesser extent than distance, area, spetral use, medium etc.
2. Electrical/optical engineering: Emitting a signal through the use of
electro/optical waves is one thing, the detection of signals appears to be
a greater problem. This was known in the first telegraph systems and
demonstrated when increasing power blew up the first transatlantic
telegraph. In order to limit noise and all kinds of artifacts and to best
detect signals a lower power bearer is preferred. Less power, less noise,
more signal, the exact opposie of kWh/GB
3. networking: the essence of networking is to communicate. In order to be
able to communicate, devices need to be active to receive. Receiving
requires energy even when no human or machine interpretable data is
received. At the same time for many types of networking there is a need
either to have occasional signals exchanged to validate the state of the
network, to update locations of mobile equipment etc. This uses energy
regardless of whether it is identified as traffic. The otherwise really bad
article on Base Station Energy use in Finland by prof. Manner and
associates has some graphs that show really well how base stations use
energy even when there is (hardly) any traffic. The difference between
little use and some use of the equipment is 10-30%, though it is doubtful
this is GB related, it could well be because there are users active and
they need to be managed and slots need to be chosen for them. At least
energy usage doesn't follow GB very well, which seems to me a strong
indication that there is not a good correlation, because the resolution
appears to be per 5 minutes or per hour and I would expect that at such
large resolutions the correlation to be more smooth.
https://ieeexplore.ieee.org/mediastore_new/IEEE/content/media/6287639/10005208/10005276/huttu11-3234192-large.gif
and
https://ieeexplore.ieee.org/mediastore_new/IEEE/content/media/6287639/10005208/10005276/huttu12-3234192-large.gif
4. Protocol and standard design: Layer 1 and 2 standards determine
distance, modulation etc and as a result how and when transmitters and
receivers are active or that they are active all the time. Quite alot of
bits and bytes are headers and related to network control. They somehow
don't feature in many calculations. Neither do we see mention of encryption
and decryption of signals in the analysis of energy use. It would appear
such characteristics might influence energy use. Routing, roaming etc all
appear to have some relevance to how network equipment uses energy
5. Omissions of signals; In reports on energy use of consumer broadband
networks the transmission of IP-TV or DVB-C bits, whether they use
multicast or unicast appear to be omitted from the calculations. This is
quite odd because those modulations make use of the same equipment and
signals. It would also appear that including them would make the kWh/GB
metric more favorable for those operating the networks. Similarly B2B
traffic, roaming traffic, MVNOs etc are at times excluded from the tally of
GB.
6. Computer science: The effort that is needed to process data in digital
systems appears to be very much related to the intentions of the sender and
receiver of those data. The intentions determine the type of network used,
the distance covered, protocols and technologies used.The same file
received from a CDN 20km away is different than from a server 11,000km
away. In addition further processing is often included in calculations, but
not well differentiated. A webpage can by plain HTML, but the same plain
HTML can also cause all kinds of calculations based on the browser and the
intentions of the users.

Apart from theoretical the empirical data also shows the same results. In
the presentation I gave to the students at this weeks RIPE event i showed
some examples of stable or declining energy use in networks, where the
amount of data transmitted did rise over the same period. This would
suggest that in live networks there is no correlation between GB however
defined and kWh. Furthermore so far attempts at using kWh/GB in any kind of
extrapolation or assessment have resulted in gross imperfections. I
included some of them in the presentation, but the way prof. Manner
calculated the energy use of mobile data in Finland and came up with a
figure double of fixed and mobile networks in Finland use together and
triple that of what Finnish mobile networks report should be a clear
indication that no scientific value can be found in the use of kWh/GB.

Lunden, Bergmark and Malmoding give some early indications on how to assess
energy use at the scale of a network and country.
https://www.mdpi.com/2071-1050/14/5/2637?trk=public_post_comment-text
Unfortunately that isn't the level where the IETF plays a significant role.
However the IETF might play a role in clearly signalling that kWh/GB is a
sure sign that the assessment of the ecological impact of the Internet will
be incorrect in the study that uses it and that the study can most likely
be disregarded.