GreenDIGIT - discussion on environmental impact metrics

Europe/Amsterdam
Catalin Condurache (EGI.eu), Gergely Sipos (EGI.eu)

Meeting Summary for Discussions on environmental impact metrics


Minutes initially generated by Zoom AI Companion, then reviewed by Catalin Condurache (EGI F.)

Quick recap

  • The team discussed the environmental impact metrics for the GreenDIGIT project, focusing on energy consumption.
  • They also discussed the availability of hourly carbon intensity data for electricity in France and other European countries, the need for feedback from brokers to the accounting system, and the development of a system to measure and rank the carbon intensity of data centers. 
  • The conversation ended with discussions on the integration of different services, the importance of accounting for resource usage in a homogeneous way across different services, and the next steps for the project.

Next steps

  • Catalin to schedule w/c 17 March a meeting with GOCDB, APEL and Accounting Portal teams to discuss a GreenDIGIT proposal:
    • extend GOCDB system with static information
    • integrate environmental metrics into EGI accounting records.
  • To define a single metric, GreenSiteRank, which can be based on different inputs for different sites but allows comparison across sites. People to contribute: Andrei, Jiri…
  • Jaime/CSIC to develop an alternative to Electricity Maps (https://app.electricitymaps.com/map/72h/hourly) for carbon intensity metric within Task 6.5. 

Summary

GreenDIGIT Environmental Impact Metrics Discussion

The team discussed the environmental impact metrics for the GreenDIGIT project. The meeting was a follow-up to a previous discussion on February 4th, where they identified some metrics within the EGI Federation and discussed additional metrics that could arise in other projects. Catalin presented a slide deck and reminded the attendees of the previous meeting's objectives and challenges. The main focus was on energy consumption, with water usage as a secondary priority. The goal was to determine which metrics to capture to assess the project's environmental impact.


Energy Data Infrastructure Proposal

Catalin provided a reminder of discussions from February 4th and discussed the proposed infrastructure for collecting and aggregating energy data, with focus on the operation phase of RIs.

He proposed diagrams showing the metrics for HTC sites and for Cloud sites, including energy use per CPU and hardware usage (to compute the carbon footprint), and specific interaction with APEL and cASO tools to extract data and feed to the metric system.

Andrei raised concerns about the confusion between the metric system and the accounting service, suggesting that the metric system should focus on monitoring how resources are used, while the accounting system should handle static site properties. Alessandro Paolini clarified that the metric system would collect data on site efficiency and power usage, which would be used by the accounting system to generate summary reports.

Andrei discussed the need for feedback from brokers to the accounting system (in the diagrams presented), particularly for DIRAC. Alessandro Paolini agreed, noting that Dirac's role as a broker should be considered. 

Catalin suggested that the APEL client should allow use of other benchmarks (along with HEPSCORE) - this to be confirmed with APEL team. Similarly for Cloud area, to be checked with cASO team

 

Hourly Carbon Intensity Data Availability

In the meeting, Jerome and Edouard discussed the availability of hourly carbon intensity data for electricity in France and other European countries. Andrei asked about the frequency of changes in these metrics. Jiri reffered to a previous discussion during the GreenDIGIT Workshop in January about the best option for their usage and the need for hourly frequency of current carbon intensity data. Jaime commented on the carbon intensity metric and the development of an alternative to electricity maps as part of Task 6.5

 

Carbon Intensity Ranking for Data Centers

Jiri proposed a system to measure and rank the carbon intensity of data centers, which could be used for scheduling decisions. He suggested using existing specifications to calculate the power efficiency of each node and combining this with dynamic site attributes like current carbon intensity and local photovoltaic production to create a "GreenSiteRank" metric. Andrei agreed that this information could be used for scheduling decisions and expressed interest in making available this single metric - GreenSiteRank.
 

Carbon Intensity API Access Discussion

Catalin and Andrei discussed the need to split work properly, particularly regarding the carbon intensity bit that requires access to a special API for national-level providers of this information. Jiri suggested using dynamic attributes in the scheduler for this purpose. Andrei proposed that the site should read the data and report through a benchmarked value. They also discussed the use of a pilot job to collect various kinds of information, including the carbon intensity. Jiri expressed confusion about the term "accounting" in the context of the schema, suggesting it might be more appropriate for the right part of the schema/diagram.
 

Input from RIs

On behalf of SoBigData, Roberto suggested the use of a big data observability framework for storage and reliability on the data. The team discussed the possibility of integrating different services, including storage, into a defined set of methods. It was agreed that storage could be evaluated similarly to computing services, but with different metrics. Andrei suggested that storage could be seen as a special type of computer with different metrics, and that a single green rank could be used to benchmark the cost of storage. Roberto agreed that the metrics for storage should be described with the policies implemented for it, as it has some policies that trigger computation or duplication of data for reliability. The team also discussed the importance of accounting for resource usage in a homogeneous way across different services.
 

Project Next Steps and Metrics Integration

In the meeting, Catalin discussed the next steps for the project, including extending the GOCDB system with static information and extending the EGI Accounting records with environmental metrics. The feasibility and a cost analysis for this development were also considered. Catalin highlighted the importance of integrating the activity of sites like CESNET and CSIC into the metric system. The timeline for the deliverable was not clearly defined, but it was mentioned that a proof of concept and demo would be needed. Yuri emphasised the need to start looking at objectives and KPIs for verification issues. The conversation ended with Catalin expressing his intention to save the recording and provide minutes for the meeting.

 

There are minutes attached to this event. Show them.
    • 2:00 PM 2:10 PM
      Introduction 10m
    • 2:10 PM 3:30 PM
      Metrics discussion 1h 20m
      • EGI Federation - HTC, Cloud
      • Other RIs
      • https://docs.google.com/presentation/d/1JpcUdjcVPDe9m0AHdBYCOu7XVa5y6Onn/edit#slide=id.g337620f7447_0_268
    • 3:30 PM 3:50 PM
      Next steps, actions 20m