Hello, I am making myself familiar with Slurm and evaluating if we can use it instead of our existing scheduler. I understand that this is a dev list but I could not find a better place to ask a few questions. So apologies for that and please let me know if there is a more appropriate place.
I've installed slurm-15.08.0-0pre4 on a single SL 6.4 box. On the existing cluster we are using cgroups making sure jobs get X GB per core so that asking for more cores brings you more memory and if you need more memory you need to ask for more cores. And that's what I'm trying to achieve with Slurm. On the one hand cgroups work with sbatch. Running sbatch --mem-per-cpu=1000 --ntasks=1 script.sh creates appropriate directories in /cgroup/memory/slurm/ /cgroup/cpuset/slurm/ /cgroup/freezer/slurm/ and if a jobs needs more memory than the limit it gets killed. A message to the user would be nice why the job was killed, but presumably there is still no template for that and everyone has to create a custom epilog. My real question is why running salloc --mem-per-cpu=1000 --ntasks=1 bash does not create cgroups and therefore gets you an unlimited interactive session? Any pointers to more details about setting up Slurm with cgroups support and the internals than available on the official pages would be highly appreciated. For example, how do cgroups get recycled? Also I'd like to set memory defaults so that all jobs are limited by default. Thank you Dr Igor Kozin | Senior Software Analyst The Institute of Cancer Research | 15 Cotswold Road | Belmont | Sutton | Surrey | SM2 5NG The Institute of Cancer Research: Royal Cancer Hospital, a charitable Company Limited by Guarantee, Registered in England under Company No. 534147 with its Registered Office at 123 Old Brompton Road, London SW7 3RP. This e-mail message is confidential and for use by the addressee only. If the message is received by anyone other than the addressee, please return the message to the sender by replying to it and then delete the message from your computer and network.
