On 24/08/2024 00:03, Matt Jolly wrote:
I use slices on HPC to limit users from monopolising interactive nodes, and our batch jobs kill anything that exceeds requested memory (hope that wasn't 100h into a job!). I do this on login with user slices but the concept is the same.
I used something a bit like that on Pr1mos. I had a batch queue that ran on maximum priority, if users wanted something done quick. But it had a wall-clock limit of 30 secs. I don't think it got used much, but it was nice to know it was there so even if the machine was being hammered you could slip small jobs in quickly.
Cheers, Wol
