Decide on Equinix metro for cluster and identify carbon intensity for this location

[rossf7]

For #3 we are manually creating an Equinix cluster in their Paris metro (their name for region). This was selected because France has a low average carbon intensity.

Here is the full list of metros.
https://deploy.equinix.com/developers/docs/metal/locations/metros/

We should decide if we're happy with this location or wish to use another?

For calculating the SCI score we need to know the carbon intensity at the location where the cluster is hosted.

https://github.com/Green-Software-Foundation/sci/blob/main/Software_Carbon_Intensity/Software_Carbon_Intensity_Specification.md#location-based-marginal-carbon-intensity

Where do we get the carbon intensity value from?
How often should we update the carbon intensity value?

Using a dynamic value may make it harder to compare between SCI scores.

[rossf7]

FYI @nikimanoledaki

Edit: Niki here is the issue from https://github.com/cncf-tags/green-reviews-tooling/pull/2/files#r1363534009

[leonardpahlke]

cc @vielmetti

[vielmetti]

Thanks @leonardpahlke - I will consult with our sustainability folks and look for an overall recommendation. We generally have best availability for sponsored project activity in EU in our AM (Amsterdam) metro.

There look to be some data center level sustainability metrics not captured in the document you share, most notably PUE ("Power Utilization Effectiveness") which tracks the efficiency not only of the energy used by the equipment directly but also the efficiency of the cooling used to run it. We track this closely because cooling improvements can make for a better data center even if you don't change any of the compute elements.

Can you review

• https://sustainability.equinix.com/environment/operational-sustainability/
• https://sustainability.equinix.com/environment/climate-commitments-and-carbon-emissions/

and see which if any of the metrics and answers and questions spelled out there give you any of the answers you need? Note that some of the breakdowns are by region and not by data center.

[JacobValdemar]

For the carbon intensity of the electricity I would recommend to use a static (average for 1 year) location-based carbon intensity. A trustworthy source for this is:

European Environment Agency, Greenhouse gas emission intensity of electricity generation, Data Visualization, Oct. 25, 2023. [Online]. Available: https://www.eea.europa.eu/data-and-maps/daviz/co2-emission-intensity-14

The latest data point was 2022 in which France had a greenhouse gas emission intensity of electricity generation of $68\space gCO_2e/kW h$.

Considering whether to use another location: France has a low carbon intensity and is not representative for most of the world. Ireland would probably be more representative. If you just want to minimize the carbon impact of this cluster, then Sweden probably has the lowest carbon intensity of the regions available.

Also, beware that the lower the carbon intensity is, the more significant the embodied impact will be relative to the operational impact. See, Carbon Efficient Karpenter p. 41 / fig. 12 for an intuition about this.

[rossf7]

I will consult with our sustainability folks and look for an overall recommendation. We generally have best availability for sponsored project activity in EU in our AM (Amsterdam) metro.

@vielmetti It would be great to get a recommendation on which EU metro to use. From the docs you posted, in EMEA you purchase 100% renewable energy and have lower emissions than Americas or APAC. So EU suits us well.

A side note but living in Catalonia which has a drought state of emergency I appreciate your focus on water usage as well as carbon!

Moving from Paris to Amsterdam is a simple code change but will involve rebuilding the cluster. So I'd prefer to do it sooner than later.

@JacobValdemar I agree France is not representative and I think Netherlands would be fine as an alternative to Ireland.

For the carbon intensity of the electricity I would recommend to use a static (average for 1 year) location-based carbon intensity

Yes, I also prefer a static annual figure over a dynamic value such as from WattTime. To allow projects to more easily compare their SCI scores.

Also, beware that the lower the carbon intensity is, the more significant the embodied impact will be relative to the operational impact.

Yes, this is an important aspect and we have #40 for this.

@vielmetti Are you able to provide data for embodied carbon per instance type? Help with this would be much appreciated.

cc @nikimanoledaki

[vielmetti]

@rossf7

Regarding embodied carbon, I have some questions which I will put here but that probably need to be discussed further.

I'm looking at our https://blog.equinix.com/blog/2023/11/06/how-to-talk-about-sustainability-in-the-data-center-industry/ where we discuss "embodied carbon" in the context of whole-building life cycle analysis (WBLCA). Reasonably speaking you want to not just capture the device itself but also the structure it's housed in. I don't know if the comparable analysis for other cloud providers factors that in or just the device itself.

BOAVITZA https://github.com/Boavizta/boaviztapi has readouts for several of the devices we stock, but not all of them.

Also missing from the analysis of #40 is any kind of understanding of networking and how that affects things. At the very minimum you might want to give a proportional usage for top-of-rack switching, but maybe someone has done a lifecycle analysis at the level of "picojoules per bit" to get data from here to there.