Optimize job poll interval for Slurm computer for production

ispg-group / aiidalab-ispg

ATMOSPEC: ab initio UV/vis spectroscopy for everyone

MIT License

6 stars 4 forks source link

Optimize job poll interval for Slurm computer for production #86

Open danielhollas opened 2 years ago

danielhollas commented 2 years ago

https://github.com/aiidateam/aiida-core/issues/5638#issue-1369918831

danielhollas commented 2 years ago

Hmm, interestingly it seems in AiiDA environment the default min job interval is always 10 seconds?

verdi shell
 load_computer("slurm").get_minimum_job_poll_interval()

danielhollas commented 2 years ago

https://aiida.readthedocs.io/projects/aiida-core/en/latest/topics/performance.html?highlight=benchmark%20workflow#benchmarks

https://github.com/aiidateam/aiida-core/pull/5724

danielhollas commented 1 year ago

I noticed that the daemon worker takes 50% CPU when waiting for ORCA job. That seems very excessive. Not sure if it is related to the issue here, or is something else...