[slurm-users] Best practice: How much node memory to specify in slurm.conf?

Marcin Stolarek stolarek.marcin at gmail.com
Tue Jan 16 23:44:07 MST 2018

I think that it depends on your kernel and the way the cluster is booted
(for instance initrd size). You can check the memory used by kernel in
dmesg output - search for the line starting with "Memory:". This is fixed.
It may be also good idea to "reserve" some space for cache and buffers -
check htop or /proc/meminfo (Slab) this may depend on your OS (filesystem,
hardware modules) and if you have a limited set of applications - workload.
Size of this part of memory may depend on "node size", number of cores
should be good measurement.


2018-01-17 6:03 GMT+01:00 Greg Wickham <greg.wickham at kaust.edu.sa>:

> We’re using cgroups to limit memory of jobs, but in our slurm.conf the
> total node memory capacity is currently specified.
> Doing this there could be times when physical memory is over subscribed
> (physical allocation per job plus kernel memory requirements) and then
> swapping will occur.
> Is there a recommended “kernel overhead” memory (either % or absolute
> value) that we should deduct from the total physical memory?
> thanks,
>    -greg
> --
> Dr. Greg Wickham
> Advanced Computing Infrastructure Team Lead
> Advanced Computing Core Laboratory
> King Abdullah University of Science and Technology
> Building #1, Office #0124
> greg.wickham at kaust.edu.sa +966 544 700 330
> --
> ________________________________
> This message and its contents including attachments are intended solely
> for the original recipient. If you are not the intended recipient or have
> received this message in error, please notify me immediately and delete
> this message from your computer system. Any unauthorized use or
> distribution is prohibited. Please consider the environment before printing
> this email.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20180117/2df40781/attachment-0001.html>

More information about the slurm-users mailing list