Hi,

Does anyone know what conditions might cause srun to peg a CPU (looks like 
mostly system time)?  Might it have to do with keeping stdout/stderr open?

I have a script that submits batches of jobs using srun (I find this more 
convenient than using sbatch) but when it submits to many jobs at the same time 
(20+) some of the srun instances can start to use 100% cpu and make the head 
node angry.  I’d like to be able to submit as many as I can but prevent this 
from happening.

Thanks,
-Craig

Reply via email to