I'd also agree that this is an interesting and important topic.
In talking with some colleagues, it also has the potential to bring
in people to the IETF who have otherwise not yet been motivated to participate.
There are likely a wide range of avenues to explore. On the server and network
device side, I suspect a common theme will be looking at ways to reduce
server utilization and/or to be able to power down network devices.
(eg, see this paper from a few years back: https://arxiv.org/pdf/1109.5641.pdf)
Design choices being made in QUIC or other protocols that result in measurably
higher CPU usage on servers and/or network devices and/or clients has
a direct impact on their energy consumption. There may be conversations to be
had as to whether a few percent performance improvement, or some small
privacy improvement, is always worth the added power utilization.
If U.S. data centers are projected to consume 73 billion kWh in 2020 [1],
then a change in a widespread protocol that increases average CPU resource usage by 0.1%
across the board has a 73 million kWh impact in the U.S. alone.
For people interested in this topic, it was pointed out to me that
the ACM now has an emerging interest group EIG-ENERGY (https://energy.acm.org/)
and its next conference has a paper-submission deadline for February:
Best,
Erik
On Wed, Jul 24, 2019 at 2:54 PM Jari Arkko <jari.arkko@xxxxxxxxx> wrote:
I think it is an interesting topic and very much on topic for us to think about. And a topic where potential savings could be significant. IT is a huge energy consumer (as well as a producer of short-lived gadgets). IT also has the potential to help optimise many non-IT processes, and therefore directly reduce energy consumption and waste.
But of course this is also complex topic and not one where only IETF or only standards help. And a topic where continuous improvements in compute and comms are often offset by more high res videos and ever growing software packages (as well as auto-play and zillions of adverts, placed on my screen after way too much ML analysis).
I’d be eager to work on this, assuming we can come up with proposals that can have an impact.
Jari
