Hi,

I'm trying to run an "embarassingly parallel" job in slurm.  My cluster
machines have 32 cores each (16 physical with hyperthreading).  When I try
and run my command (call it foo) via slurm, each process seems to only use
10% cpu each and run very slowly, while if I run my jobs directly from the
command line they each use 100% cpu and finish much more quickly.  I think
these are the relevant bits of my slurm.conf:

SchedulerType=sched/builtin
SelectType=select/cons_res
SelectTypeParameters=CR_CPU
NodeName=node1 CoresPerSocket=8 ThreadsPerCore=2 Sockets=2 State=UNKNOWN


This is the command I'm using to run my jobs

 for file in $@; do

sbatch --wrap="foo $file"
done

How do I get slurm to run these jobs simultaneously and have each use 100%
of a core?

thanks in advance for any help.

Reply via email to