Re: [E-impact] Metrics for E-Impact and Sustainability -- Summary and Next Steps

[Romain Jacob <[e-mail]>]

Hi Dom,

I think you make an important point: we must differentiate between 
proportionality in _traffic_  and _network capacity._

I've been arguing for research to improve energy proportionality within 
the network (L2 and L3). But what I have in mind is not "reducing the 
traffic volume to save energy" (as pointed out many times, there are 
only little gains here). Rather, I'm thinking about "dynamic capacity 
modulation" in ways that yield energy savings.

What I mean by "capacity modulation" is, e.g., turning off ports when 
you don't need them. Another example is adjusting the capacity of the 
transceivers (e.g., by changing the modulation scheme, use fewer 
frequencies/wavelengths, playing with the laser power, etc.). It's 
really unclear how much savings this can yield, but I think it's worth 
considering. I've done some initial research in that direction; the 
first results will be presented at HotCarbon in a couple of weeks (I'm 
working on the final paper at the moment).

[ And thanks Carlos for the constructive warp-up! ]

Best,
-- 
Romain

On 15/06/2023 09:02, Dom Robinson wrote:
> This cisco blog talks about ‘G’
>
> Looks to me like Gbps. If so its a measure of capacity provisioning 
> not traffic use. It would only be comparable to GB at 100% 
> utilisation. At 100% utilisation there is no ‘proportionality’ anyway 
> so its a distraction to infer that that report correlates traffic to 
> energy.
>
> This collective attempt to find traffic / energy proportionality ‘at 
> all costs’ seems lazy. Why do we NEED to wed our energy efficiency to 
> designs revolving around data optimisation?
>
> Is it because economic systems are prolific that measure data transfer 
> for billing and its then easy / lazy to conjure some CO2 correlation? 
> Certainly i believe this is part of the appeal. Its a quick (apparent) 
> win.
>
> I now an convinced that ‘NOT sending data does NOT save any energy’ in 
> transmission networks. The energy quanta scales with provisioning 
> capacity. No one anywhere has convinced me otherwise.
>
> I am and remain wide open to being convinced.
>
> But no one has.
>
> A further part of the problem is conflating layers. Layer 4 has high 
> energy proportionality but in IETF terms that is the ‘ends’ of the 
> end-to-end principle and out of scope.
>
> Layer 3 (IETF home land) is abstract and exists conceptually but is 
> really a point of interop between layer 2 and 4.
>
> Layer 2 is variable. Some radio network have a proportionality with on 
> / off but in practice the requirement for 24/7 signal path 
> availability generally means the on time dominates.
>
> And layer 1 is all about CO2e which is a)certainly capacity 
> provisioning related and not traffic-elastic and b) something of a 
> different conversation from an IETF ‘protocol development’ perspective.
>
> So until someone highlights a good report showing any energy 
> proportionality with traffic I remain focussed on how we are 
> provisioning capacity to layer3 rather than on ‘reducing traffic’.
>
> However I also think that there is MUCH that is energy proportional in 
> Layer 4: Not encoding. Not storing. Not background pre-processing.. 
> all these ‘nots’ DO reduce energy requirements and measurably.
>
> However they are not proportional to traffic signalled over the 
> internet in Layer3.
>
>
>
> -- 
> Dom Robinson
> www.id3as.co.uk <http://www.id3as.co.uk/>
> www.greeningofstreaming.org <http://www.greeningofstreaming.org/>
> uk.linkedin.com/in/domrobinson <http://uk.linkedin.com/in/domrobinson>
> Meet >> https://calendly.com/id3as
>
>> On 15 Jun 2023, at 02:12, Rudolf van der Berg 
>> <[e-mail]> wrote:
>>
>> 
>> That i shouldn't respond this late on a mobile phone. Too many errors 
>> and mixed temperature and traffic in one sentence
>>
>> But first proposal; please get us some real data on energy use of 
>> equipment and show what effect several parameters have on energy use
>>
>> Op do 15 jun. 2023 02:58 schreef Carlos Pignataro (cpignata) 
>> <[e-mail]>:
>>
>>     So the useful proposal is?
>>
>>     More later as well :-)
>>
>>     Carlos Pignataro (He/Him)
>>
>>>     On Jun 14, 2023, at 8:54 PM, Rudolf van der Berg
>>>     <[e-mail]> wrote:
>>>
>>>     Thanks!! This is awesome. It is also way past my bed time, so I
>>>     will respond later. I do see I wrote traffic lights, where I
>>>     meant street lighting! That said it is about their energy use,
>>>     which certainly for most street lights is not proportional to
>>>     traffic and likely not much either for traffic lights, though
>>>     maybe with some demand response systems it might affect the
>>>     distribution of red, green and orange and their number.
>>>
>>>     When it comes to cat videos, given that they form much of what
>>>     some family members communicat(e) some people lack an
>>>     understanding of utility functions, cats trump IAB any day of
>>>     the week. It is not up to us to clarify bits. It is my ISPs job
>>>     to push them and respect my privacy and net neutrality and
>>>     certainly not to assume what is valuable. If your net worth
>>>     beats Grumpy Cat or Taylor Swift's cat than maybe what you do
>>>     may be more valuable than cat videos. So no the point of the
>>>     Internet is not to push bits. It is to interconnect. To be
>>>     connected. The amount of bits and how they are pushed is not
>>>     relevant for the value of being connected. a PSTN line was
>>>     powered 24/7 for good reason. So is my parents FTTH or my FTTH.
>>>
>>>     So yes relating these to desired effect is to relate them to
>>>     being connected. What the bits are and do and your opinion of
>>>     the context of the bits is irrelevant (and should you actually
>>>     peek in the packets, a criminal offense) So if you look for a
>>>     metric relative to something, try per user or port.
>>>
>>>     With regards to there being absolutely no correlation between
>>>     bita transferred and energy consumption, you caught me, I was
>>>     using a style figur, known as a hyperbole. I had hoped that
>>>     someone, maybe at a router manufacturer (Juniper, Nokia, Arista
>>>     etc.) would actually show real data on how power use of a router
>>>     varies under load in real life situations. I used to use this
>>>     Cisco report that said energy use under load vs idle was at most
>>>     20W, which was 1% for A 6509 with 150 ports, but it is old and
>>>     it may just not be the case anymore.
>>>     Cisco 6509 energy.pdf
>>>     <https://drive.google.com/file/d/1isfRXxVnXdZycyMx2-slubPbMutsqBYc/view?usp=drivesdk>
>>>
>>>     But pls point me to resources that have better data. The blog
>>>     you point to doesn't provide data. I can't see what relative
>>>     effects are and how this is affected by traffic shifted over a
>>>     day, peak levels and duration of use,  traffic load, number of
>>>     users, outside temperature, humidity or geographic spread.
>>>
>>>     With regards to logical fallacy. Of geographic spread causes
>>>     such differentiation that the statistics aren't comparable
>>>     between operators, than would it not be realistic for us to
>>>     factor that in into the metrics?  Maybe the IAB can advise
>>>     people to all live in a country as densily populated as the
>>>     Netherlands. However, even if geography determines energy use, I
>>>     would still like to have the relatie impact of traffic levels in
>>>     this equation too! I also take negative correlations, such as
>>>     the one where KPN lowered energy use by 50 over the last decade,
>>>     but increased traffic 20 times!!  I am also willing to admit
>>>     defeat and incompetence If you show that yes overall energy flow
>>>     decreased but all of that was from electricity use of peripheral
>>>     systems and not the actual network it self.
>>>
>>>     How comparability is a logical fallacy is beyond me and defies
>>>     every tenet of methods and techniwques of academic social
>>>     science research I had in my freshman year. It is exactly by
>>>     comparing whether the same traffic as well as different levels
>>>     of traffic on the same network or other networks under differing
>>>     conditions actually changes the results. eg; when traffic is the
>>>     same in same nerworke and difference in temperature is 5
>>>     degrees, if energy use changes, we can attribute the changes to
>>>     traffic.if traffic grows 20% and energy use is same at the same
>>>     temp as well as the higher temp, then traffic may not be
>>>     relevant. Or we may see minimal changes in energy use etc. You
>>>     know, like in the Cisco article where energy use per capacity is
>>>     not stable across routers of varying sizes, which suggests
>>>     capacity is only one, but not very dominant factor...
>>>
>>>     More to come later
>>>
>>>     Op do 15 jun. 2023 00:23 schreef Carlos Pignataro (cpignata)
>>>     <cpignata=[e-mail]>:
>>>
>>>         Hi, E-impact,
>>>
>>>         After re-reading and parsing through the thread
>>>         <https://mailarchive.ietf.org/arch/msg/e-impact/t9lzF7j2DQWi4EDp7Pe2JKlcQdg/>,
>>>         I wanted to offer a summary set of highlights and
>>>         observations, hopefully with a bias towards usefulness to
>>>         the IETF-driven outcomes, removing underlying assumptions,
>>>         and way finding common grounds and next steps.
>>>
>>>
>>>         _Desired effect, resources, and comparability: _
>>>
>>>>         On May 15, 2023, at 5:35 PM, Rudolf van der Berg
>>>>         <[e-mail]> wrote:
>>>>         Ok, I will be pedantic... efficient is defined as achieving
>>>>         a desired effect while not wasting resources.
>>>
>>>         First, if a key desired effect of the Internet (roughly
>>>         paraphrasing §2 of RFC 3935) is achieving various types of
>>>         communications between parties connected to it, bits and
>>>         traffic are a vehicle towards that goal. Avoiding here the
>>>         judgement or qualification of usefulness just yet — Michael
>>>         Richardson rightly pointed out that Spam and cat videos are
>>>         not of the same usefulness value. So, packets/bits pushed is
>>>         an important outcome.
>>>
>>>
>>>         On May 15, 2023, at 5:36 PM, Dom Robinson <[e-mail]>
>>>         wrote:
>>>>         “ kWh consumption by operator is somewhat meaningless as well”
>>>>
>>>>         I have to disagree with that, since the absolute amount of
>>>>         energy that they use, regardless of the number of
>>>>         subscribers or distribution, is it direct measure in the
>>>>         drain on energy resources.
>>>
>>>         Also, electrical energy is a key resource — and for now also
>>>         not considering emission factors, etc., as Lars had
>>>         suggested the ‘all is bad’ approach for simplicity.
>>>
>>>         From these two points, I would consider it not unreasonable
>>>         to ratio resources per unit of desired effect — even
>>>         starting at the network level.
>>>
>>>         Taking this as a (potentially unpolished) start, there seem
>>>         to be at least a couple of conflicting points, primarily or
>>>         summed up as
>>>
>>>>         On May 15, 2023, at 3:26 PM, Rudolf van der Berg
>>>>         <[e-mail]> wrote:
>>>>
>>>>         the simple reality is that energy use and data traffic have
>>>>         no correlation in fixed and mobile broadband networks. It
>>>>         is like measuring the efficiency of cars based on wiper
>>>>         fluid use. Or traffic lights energy use based on the number
>>>>         of cars that drive by.
>>>
>>>         The first sentence here expresses an absolute that does not
>>>         match my experience. While clearly the relationship is not
>>>         linear with zero intercept, the blog
>>>         <https://blogs.cisco.com/sp/assessing-the-real-world-environmental-impact-of-routers> referenced
>>>         before explains how, within a networking device, router,
>>>         switch, etc., different modules have different
>>>         characteristics of power consumption.
>>>
>>>           * Some require constant power independent of traffic flow
>>>             — but can be turned off in the absence of traffic
>>>           * Some are simply idle power and independent of traffic load
>>>           * Some are dependent of packet load (pps)
>>>           * Others correlate to and are a function of bit load and
>>>             bandwidth (bps)
>>>
>>>
>>>         Now, this tells me there is *some* correlation — might not
>>>         be linear with zero intercept, but there is some. Even if
>>>         idle power is a high percentage, there’s a variable
>>>         depending on data traffic.
>>>
>>>         The danger of making ~ “no correlation” bold absolute
>>>         statements, is that a simple counter-example and
>>>         #houseofcards...
>>>
>>>         [as an aside, knowing nothing about traffic light planning,
>>>         but I can find sensibility in the relationship between
>>>         traffic lights🚦and number of cars 🚗 driving by, and would
>>>         not ridicule such measuring of the efficiency of a traffic
>>>         light system — more 🚗🚙🚕🏎️ for a single 🚥 versus 🚦🚦🚦🚦]
>>>
>>>>         Op ma 15 mei 2023 17:25 schreef Rob Wilton (rwilton)
>>>>         <[e-mail]>:
>>>>
>>>>             I struggle to understand how “traffic and energy use
>>>>             are in no way and have never been related”.
>>>>
>>>         I struggled as well…
>>>
>>>>         On May 15, 2023, at 3:26 PM, Rudolf van der Berg
>>>>         <[e-mail]> wrote:
>>>>
>>>>         As I explained there is no scientific basis for the metric.
>>>>         If you look at the work by professor Manner, you will see
>>>>         that using the metric to make prognosis about the Finnish
>>>>         mobile networks electricity use leads to estimates 3 times
>>>>         higher than the official figures.
>>>
>>>         The issue of comparability seems to be the fallacy on the logic.
>>>
>>>         As it was mentioned already by many, same traffic on a more
>>>         geo spread-out network, versus a dense metro network versus
>>>         Ecuador versus Canada temperatures versus…
>>>
>>>
>>>>         On May 26, 2023, at 3:52 AM, Dom Robinson <[e-mail]>
>>>>         wrote:
>>>>
>>>>         I agree that taking an holistic view is paramount.
>>>         +
>>>>         On Jun 9, 2023, at 10:12 AM, Romain Jacob <[e-mail]>
>>>>         wrote:
>>>>
>>>>         We discussed several times on this mailing list that we
>>>>         have a rather poor understanding of the energy consumed by
>>>>         switches and routers depending on their load and
>>>>         configuration. One way to address this problem is to
>>>>         perform "black box" measurement of the power drawn by such
>>>>         device and fit a power model for it.
>>>>
>>>>
>>>
>>>         One thing I’d expect we might agree is that the problem
>>>         space is highly complex, brutally interdependent, and deeply
>>>         nuanced.
>>>         In this, RFC 1925, truth #5 came to mind.
>>>
>>>         Instead of decomposing a routing system, taking a ‘black
>>>         box’ approach shows some relationship of traffic <-> energy.
>>>         The scope of comparison might be narrow, and as someone
>>>         suggested could even be limited to a given deployment as
>>>         f(time).
>>>
>>>         We could define the most realistic variations, including:
>>>
>>>           * kWh/GB Optics excluded
>>>           * kWh/GB including storage to transport, etc.
>>>
>>>
>>>         Or:
>>>>         On Jun 9, 2023, at 10:12 AM, Romain Jacob <[e-mail]>
>>>>         wrote:
>>>>
>>>>               o energy efficiency, measured as "maximum power
>>>>                 consumed divided by the maximum capacity"
>>>>               o energy proportionality index, capturing how far the
>>>>                 device is from being perfectly power-proportional
>>>>                 to the load: (max_power - idle_power)/ max_power
>>>>
>>>         But:
>>>
>>>>         On May 15, 2023, at 6:07 PM, Brian E Carpenter
>>>>         <[e-mail]> wrote:
>>>>
>>>>         All very interesting but how does this help the IETF
>>>>         improve its protocol designs?
>>>
>>>         Suggesting and anchoring on a small number of relevant
>>>         metrics, as per above (as a start), instead of attempting to
>>>         “ban” others.
>>>
>>>         And also benchmarking power measurement.
>>>
>>>         Thanks!
>>>
>>>         Carlos Pignataro (He/Him)
>>>
>>>         -- 
>>>         E-impact mailing list
>>>         [e-mail]
>>>         https://www.ietf.org/mailman/listinfo/e-impact
>>>
>>
>> -- 
>> E-impact mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/e-impact
>
-- 
Romain JACOB
Postdoctoral Researcher
ETH Zurich
Networked Systems Group (NSG)
Lead: Prof. Laurent Vanbever
www.romainjacob.net <https://www.romainjacob.net/>
@RJacobPartner <https://twitter.com/RJacobPartner>
@[e-mail] <https://discuss.systems/@jacobr>
Gloriastrasse 35, ETZ G81
8092 Zurich
[phone]