Good evening, I'm trying (with no much success so far) to use slurm on a single machine and achieve the following:
- config multiple "partitions" (queues) with different (enforced) resource limits (cpu, mem and time) on each - set access controls on them in order to permit only selected users (or "accounts") to submit jobs As a bonus I'd like to limit (using cgroups, perhaps) the resources for interactive users in order to force them to use the queueing system. I have tried to follow the instructions in http://rolk.github.io/2015/04/20/slurm-cluster that covers pretty much of what I'm trying to do but it lacks the details on setting up the queue limits and the interaction of cgroups and slurm (would it be necessary to run srun or slurmd setuid to user "slurm" so the process submited would not be constrained by cgroups?) But, the real question is, for now: is slurm suited for this scenario or it'd be better to use some other tool? Please advice. Machine details: # slurmd -C NodeName=m1 CPUs=192 Boards=6 SocketsPerBoard=12 CoresPerSocket=8 ThreadsPerCore=2 RealMemory=1499581 TmpDisk=141063 OS: Debian 9.5 Slurm version: 16.05.9 (from the repositories) Thanks for your time Marcelo.
