Good morning,
My jobs on yggdrasil are pending for a few day. And I noticed that sometimes the same user are running about 30 jobs and using almost all the available shared gpu from the squeue list.
I also found this
https://hpc-community.unige.ch/t/max-number-of-running-and-pending-jobs-per-user/239
It said the limitation of running and pending jobs per user is 10k. Could you mind my asking that is this limitation reasonable compared to the total number of the available shared gpu?
It’s just an idea. But how about setting a limit on the maximum percentage of the resource proportion? Like one user can only use no more than 60% of the total computational cpu/gpu. So that the computational resources can be cycled more healthily.
Thank you for your time. I’m looking for your insights.
Best,
Jingze