Hello all,

I’m wondering how other compute centers handle the logical vs physical cores 
issue within SLURM. 

By default, SLURM reports logical cores when you ask for job status information 
using sacct, qstat, etc. We’d prefer physical cores to be reported to avoid 
confusion for our end users. However, it’s unclear to us how to best accomplish 
this. One route is to simply change the ThreadsPerCore to 1 for all nodes in 
slurm.conf. The slurm daemons complain when we do this but everything seems to 
run as desired. 

Will this have any unintended consequences (especially those impacting 
performance)? Are there any other parameters we should tweak? 

---------------
Will French, Ph.D. - Research Computing Liaison
Advanced Computing Center for Research and Education - Vanderbilt University
www.accre.vanderbilt.edu - 615-343-4134 - Hill Center 201









Reply via email to