Bill Broadley <b...@cse.ucdavis.edu> writes: > I've looked at sarray and arrayrun. Both seem a bit awkward.
As the author of arrayrun, I have to admit that using a separate job for submitting the subjobs is a bit awkward. (You don't have to run arrayrun in a job script, though, you could run it interactively (probably a good idea to use screen, then). However, that doesn't solve the problem of having to poll the queue.) But that was the best I could come up with (short of modifying slurm) that would mimic array jobs such as in SGE, where more subjobs get started as more nodes/cpus become available. Using srun within a fixed allocation limits your subjobs to the resources in that allocation. (Now that slurm jobs can grow and shrink, there might be a way to avoid this.) -- Regards, Bjørn-Helge Mevik, dr. scient, Research Computing Services, University of Oslo