mrchrisadams
commented
4 years ago Hey @JamieBeevor I've found another good summary to help work update the calculations, from David Mytton, formerly of Server Density, a server analytics platform.
It goes into a decent amount of detail that would be useful to check against the IEA analysis:
https://davidmytton.blog/how-much-energy-do-data-centers-use/
The bits we'd need to update to make a model that takes into account the lower emissions figures in the analysis from Carbon brief, which I mostly agree with TBH.
You can see the code bits here: https://github.com/thegreenwebfoundation/co2.js/blob/master/src/co2.js#L5-L28
And the code powering the figures are here:
// Use the 1byte model for now from the Shift Project, and assume a US grid mix of around 519 g co2 for the time being.
const CO2_PER_KWH_IN_DC_GREY = 519;
const CO2_PER_KWH_IN_DC_GREEN = 33;
// the better way would be do to this with a weighted average of types of RE generation
// from solar, to wind, to biomass, and hydro and so on, based on how much they
// are used.
// For now, let's use a quoted figure from Ecotricity, which quotes OFGEM, the UK regulator
// and it's better than most quoted figures which pretend there is *no* footprint
// for RE.
// More here:
// https://www.ecotricity.co.uk/layout/set/popup/layout/set/print/for-your-home/britain-s-greenest-energy-company/lifecycle-carbon-emissions
// https://twitter.com/mrchrisadams/status/1227972969756086284
// https://en.wikipedia.org/wiki/Life-cycle_greenhouse-gas_emissions_of_energy_sources
// 33.4, but that's for the UK in 2014/15. Shouldn't be too off though. Probably lower now.
const CO2_PER_KWH_NETWORK_GREY = 495;
// these are the figures pulled from the 1byte model, there is a
// third figure to represent carbon emission from *making* the device used to access a site/app but we lieave it out as it relies on
// us knowing how long it's being used to read content
const KWH_PER_BYTE_IN_DC = 0.00000000072;
const KWH_PER_BYTE_FOR_NETWORK = 0.00000000152;
Updating these numbers based on the peer reviewed data
If the folks at the Shift project can have the 1byte model, I figure we can make one too, which until we have a proper name, I'm gonna suggest as calling the Green Byte Model, to indicate that it is intended to be largely 'compatible' for the same kinds of calculations.
In the model, they use these:
const KWH_PER_BYTE_IN_DC = 0.00000000072;
const KWH_PER_BYTE_FOR_NETWORK = 0.00000000152;I find them a bit hard to read, so lets try going from bytes to gigabytes (i.e. _KWH_PER_BYTE_INDC 1024 1024 * 1024) to make these easier to read:
const KWH_PER_GB_IN_DC = 0.77
const KWH_PER_GB_IN_DC = 1.63How should we change the numbers? This chart below gives a direct comparison for the numbers (click through to see them)
https://cbhighcharts2020.s3.eu-west-2.amazonaws.com/video-streaming/chart-1.html
I think as a starting point, we could used these updated numbers for a Green Byte Model, and make it a thing we could swap for calculations.
An aside about leaving out device emissions and asking for a sanity check.
I left out the CO2 emissions from devices itself, as for this work on making this work in sitespeed, I felt it was better to be explicit about the boundary, and not even try to incorporate it.
For the purposes of the sitespeed use, I figured it would be hard to reliably measure a drop in total life cycle in carbon from devices, just from optimising a website or having it use green energy, but if this is used elsewhere it might make sense to include.
So I'm tagging you to ask a few questions.
- Do you know a more sensible figure for renewables than OFGEM's one from a years back?
- The figure for the IEA power is globally just under 500 g for power. Got an idea how we might end up with a more representative figure than the IEA's figure, to account for datacentres?
I understand the point from the IEA analyst about the biggest tech firms using increasing percentages of power, but historically, my understanding is that datacentres have been cited where there are loads of fossil fuels, and providers that aren't in the spotlight tend not to be in such a hurry to buy PPAs and the rest, and in many cases it's straight up difficult.
I'm thinking of Frankfurt, Amsterdam and well, most of North Virginia, which typically have historically have authorities who are more fossil-fuel friendly than fans of low carbon power - see Greenpeace's Virginia report).
Anyway, let me know, and I'll add some of this. I might tag the original analyst, George Kamiya as he might have some reckons on the energy mix for datacentres.
JamieBeevor
allcontributors[bot]
dvelasquez
martin-laurent
ceddlyburge
fershad
I've been looking for a model that gives numbers closer to my own experiences looking online for the energy from ICT and therefore carbon from ICT than the 1byte model, but nothing has been as easy to use, and I've prioritised having some discussion, compared to having the numbers that are more consistent with the work from Koomey et al.
A new piece of analysis on the carbon brief has some at last.
I've linked directly to the part of the article that lists the sources.
We should make a table listing these in a way that's comparable to the 1byte model, then allow the use of both, to make comparisons possible.