Reputation: 151
I'm curious if there is a way to setup Slurm partitions so that the scheduler will query against available partitions and assign to a partition depending on amount of requested resources, eg. say I have a few queues:
small_cpu = node1,2,3 <- assigned when requesting <= 10 cores
large_cpu = node4,5,6 <- assigned when requesting > 10 cores
big_mem = node7,8,9 <- assign when requesting > 100GB memory
fail/backfill = nodes10,11,12 <- assigned when the other queues are full
I'm able to do this in PBS/Moab, but I'm having difficulty with the Slurm documentation trying to figure it out. Any insight is appreciated.
M
Upvotes: 1
Views: 130
Reputation: 59340
The way to do that exact thing in Slurm is with a job submit plugin that allows you to incorporate any logic in the routing of the jobs to partition (and many more).
See an example here.
Upvotes: 1