Green IO Conference at apidays Paris 2024 - From Watts to Action, Pierre Segonne & Marcus Garsdal, Electricity Maps

From Watts to Action: A Data-Powered Blueprint for Sustainable Computing
Pierre Segonne, Team Lead at Electricity Maps
Marcus Garsdal, MLOps Engineer at Electricity Maps

Green IO Conference at apidays Paris 2024 - The Future API Stack for Mass Innovation
December 3 - 5, 2024

------

Check out our conferences at https://www.apidays.global/

Do you want to sponsor or talk at one of our conferences?
https://apidays.typeform.com/to/ILJeAaV8

Learn more on APIscene, the global media made by the community for the community:
https://www.apiscene.io

Explore the API ecosystem with the API Landscape:
https://apilandscape.apiscene.io/

Transcript

1. From Watts to Action: A Data-Powered Blueprint for Sustainable Computing Pierre Segonne / Marcus Garsdal electricitymaps.com
2. 2 Your speakers today Representing the grid forecasts team of Electricity Maps Pierre Segonne is the lead for the Grid Forecasts team of Electricity Maps. Marcus Garsdal is in charge of Machine Learning operations within the Grid Forecasts team of Electricity Maps
4. Climate change is caused by fossil fuels and our world is powered by them 4 CO2-emitting fuels representing 85% of our energy must get to 0 by 2050 for 2°C target Nuclear Renewables (hydro, wind and solar)
5. The solution is easy 5 ELECTRIFY ALL THE THINGS! (with clean electricity)
6. We foresee a world where billions of grid-connected systems optimise when and where they consume electricity in order to reduce their cost and carbon
7. The Tech industry must become a positive actor in grid decarbonisation
8. Electricity consumption from data centres is booming 8 17 % 32 % 2022 2026 Source: Electricity 2024 Analysis and Forecast to 2026 IEA. Link
9. 9 2 % 15-20 % 2022 2026 Source: Electricity 2024 Analysis and Forecast to 2026 IEA. Link Source: Analyseforudsætninger til Energinet 2022  Datacentre. Energistyrelsen, 2022. Link Electricity consumption from data centres is booming
10. Globally, we can’t ignore the resulting emissions 10 ` 1 year Source: Data Centers and Transmission Networks Overview. IEA. Link Source: ICT sector electricity consumption and greenhouse gas emissions. Malmodin et al 2023 Data centres emissions (126 Mt CO₂)
11. Globally, we can’t ignore the resulting emissions 11 ` 11 = 1 year Source: Data Centers and Transmission Networks Overview. IEA. Link Source: ICT sector electricity consumption and greenhouse gas emissions. Malmodin et al 2023 Data centres emissions (126 Mt CO₂) 󰎙
12. Globally, we can’t ignore the resulting emissions 12 12 ` = 1 year Source: Data Centers and Transmission Networks Overview. IEA. Link Source: ICT sector electricity consumption and greenhouse gas emissions. Malmodin et al 2023 Data centres emissions (126 Mt CO₂) 󰎙 󰎙
13. Globally, we can’t ignore the resulting emissions 13 13 ` = 1 year Source: Data Centers and Transmission Networks Overview. IEA. Link Source: ICT sector electricity consumption and greenhouse gas emissions. Malmodin et al 2023 Data centres emissions (126 Mt CO₂) 󰎙 󰎙 󰎙
14. Luckily, computing is often flexible 14 Analytics exports ML workloads Background services DB backups
15. And flexibility matters CO₂ emissions: Same location, different times 15 Source: Electricity Maps. Link
16. And flexibility matters CO₂ emissions: different locations, same time 16 Source: Electricity Maps. Link x9
17. Outline Understanding electricity grids A blueprint for sustainable computing Reduce Locate Provide Optimise
18. What happens when I shift the execution time of a single compute job?
19. A power differential? 19 1 Building an AWS EC2 carbon emissions dataset, Benjamin Davy, Medium To study the effect on emissions, we first need to look at power I do not execute my job at 10:00 as planned The instance I should have used is idle The data centre consumption is reduced by ~½ of the normal consumption of the instance The instance I should have used is picked up by some other service through spot The data centre consumption is unchanged Best scenario Worst scenario 𝜟P  100 W¹ 𝜟P  0 W
20. A (over)simplification of the grid 20 Source: Inertia and the power grid: a guide without the spin, NREL, 2020 Heavy rotating mass = load Fixed frequency 50 Hz
21. A (over)simplification of the grid 21 Source: Inertia and the power grid: a guide without the spin, NREL, 2020 Frequency will increase Smaller mass = smaller load
22. The Synchronous grid of Continental Europe 22 Source: ENTSOE Transmission system map The equivalent inertia is gigantic¹! 1 667 GW of capacity
23. 23 Answer: At most, the frequency will increase from 50 Hz To 50.000000000…001 Hz
24. What happens when a nuclear reactor trips?
25. When the grid is actually significantly affected 25 On the 17th of November 2024 at 1525 a nuclear reactor in Finland failed and went from full power 1.6 GW) to 0 Source: Electricity Maps + 25%
26. When the grid is actually significantly affected 26 Source: Hybrid Greentech The grid must react immediately! 𝜟frequency = 0.5 Hz
27. To sum up ... 27 𝜟 ~ 100 W 𝜟 ~ 1 GW Nothing changes on the grid. Emissions stay the same Multiple other plants must compensate the imbalance. 𝜟 in Emissions . . . Itʼs complicated! 100 kW < 𝜟 < 100 MW
28. Marginal emissions 28 ? A solution to understand how the grid reacts? The marginal emissions represents the emissions rate of the electricity generator(s) that are responding to changes in load on the local grid at a certain time. Trigger your job Identifies the power plant responding to the additional demand Computes emissions from it
29. Marginal emissions Don’t be fooled - marginal emissions are unfit for optimising consumption 29 Read more here: https://www.electricitymaps.com/marginal-signals and https://www.electricitymaps.com/blog/innovations-marginal-signals A Optimisation based on marginal emissions sounds too good to be true? … It is!
30. 30 A data-powered blueprint for sustainable computing
31. 31 1 - Reduce total energy consumption X Y Z Tune provisioned resources Reassessed our cloud run provisioning Give teams ownership of their energy consumption Implement monitoring tools Surface the ratio energy used / usefulness as KPI X kWh
32. 32 2 - Move services to low-carbon regions North Carolina Montréal Stockholm Frankfurt
33. 33 2 - Move services to low-carbon regions Ex, other providers 𑇐 Azure 𑇐 AWS Result on our own footprint GCP New compute heavy project 𝜟 30%
34. Not all providers offer as actionable tools to reduce emissions A once in a generation chance to fixing corporate emissions reporting¹ 34 3 - Pick the right providers Big tech is pushing actively within the revision of the GHG protocol What they advocate for reveals where they stand Read more here The AWS customer carbon footprint guarantees to all customers that, no matter what, their emissions will be 0 in 2025! 1: NRDC
35. 35 4 - Optimise the execution of your compute Your job Custom scheduler Carbon-aware SDK Help the grid decarbonise
36. 36 Electricity Maps’ Forecasts
37. Electricity Maps’ forecasts Predicting the future state of the whole grid 37 Exchanges All production modes
38. Electricity Maps’ forecasts Using our own forecasts to flowtrace the future grid state 38 Exchanges All production modes Flowtracing Flowtraced data served to customers Forecasts
39. 39 How Electricity Maps’ data is being used Data centres EV charging Consumer app
40. electricitymaps.com Together, let us enable a data-driven decarbonisation of electricity
41. Thank you for your attention. www.electricitymaps.com Pierre Segonne / Marcus Garsdal