HPC Jeep = High Performance Computing: Job Efficiency and Energy Usage, Monitoring & Reporting
What’s the scope of this project?
Traditionally, resources on high performance computing (HPC) services have been granted in units of compute time (core-hours, node-hours, etc.) with users not informed, nor paying much attention to their energy use; and services not generally reporting breakdowns of energy use other than for total electricity charges. Without an understanding of how research areas, projects and users use energy and how efficiently they use resources it is difficult to plan future procurements or service strategies towards a net zero goal.
This project will look at what level of energy and efficiency information can be extracted from current and historical per-job data from ARCHER2 and DiRAC services; and how this data can be analysed and synthesised to provide the information required for funders (who procure and set service strategy and objectives) and researchers (who make decisions about how best to use the resources they are granted) to make informed decisions about how to manage HPC resources to extract the maximum research benefit in the most energy efficient way.
Alastair Basden, Durham University
Andy Turner, University of Edinburgh
Project reference document here: DOI: 10.5281/zenodo.6787599
HPC-JEEP’s first report has been published and the supporting dataset is also available.
Blog post by Andy Turner - Software power draw on ARCHER2 - the benefits of being an RSE