The problem

Researchers see wall-clock time and core-hours, but many schedulers also record energy per job, and almost nobody looks.

The fix you can use today

• Check whether your scheduler records job energy (for example sacct on Slurm).
• Compare energy per result between code versions and settings.
• Estimate before you run with a calculator for bigger campaigns.

The UK's ARCHER2 supercomputer lets users look up the energy of every job. GreenCode is exploring checks like this for jobs on shared clusters.

Tools that help

• Slurm energy accounting: records the energy each job consumed, so users can look it up after the run
• Green Algorithms calculator: estimates the carbon footprint of a computation from its runtime, hardware and location
• nf-co2footprint: a Nextflow plugin that estimates the energy and CO2 of each task in a pipeline run

How GreenCode helps

GreenCode helps tackle this by measuring research code's energy before and after a change, pointing to the code behind each hotspot, and checking that the optimised code still gives the same results. More on this.

Tags

#hpc#measurement#research-software

Sources

• Research software and HPC energy: ARCHER2 records energy per job
• ARCHER2 user documentation: energy use and emissions
• SchedMD (Slurm), sacct manual: ConsumedEnergy field

All GreenCode Tech Tips