[slurm-users] sacct: error

[Eric F. Alemany]

Thank you Chris, Marcus, Patrick and Ray.

I guess i am still a bit confused. We will se what happen when we run a job asking for the CPU’s of the cluster.


_____________________________________________________________________________________________________

Eric F.  Alemany
System Administrator for Research

Division of Radiation & Cancer  Biology
Department of Radiation Oncology

Stanford University School of Medicine
Stanford, California 94305

Tel:1-650-498-7969<tel:1-650-498-7969>  No Texting
Fax:1-650-723-7382<tel:1-650-723-7382>



On May 7, 2018, at 4:08 AM, Chris Samuel <chris at csamuel.org<mailto:chris at csamuel.org>> wrote:

On Monday, 7 May 2018 5:41:27 PM AEST Marcus Wagner wrote:

To me it looks like CPUs is the synonym for hardware threads.

Interesting, at ${JOB-1} we experimented with HT on a system back in 2013 and
I didn't do the slurm.conf side at that time, but then you could only request
physical cores and you would be allocated both thread units (unless you lied
to Slurm and listed them as physical cores instead).

Everything I've ever done since has had ThreadsPerCore=1, until today where we
have some KNL nodes which are ThreadsPerCore=4.

I can confirm there that you get 4 hardware thread units (a single core) when
you request  -n 1 -c 1 - here from an interactive job on a KNL node:

[csamuel at gina1 ~]$ cat /sys/fs/cgroup/cpuset/$(cat /proc/$$/cpuset)/
cpuset.cpus
0,68,136,204

So in the sense of what you put in slurm.conf you are indeed right,
CPUs=boards*sockets*cores*threads, but from the point of view of what you
*request* CPUs are just boards*sockets*cores.

Confusing!

All the best,
Chris
--
Chris Samuel  :  http://www.csamuel.org/  :  Melbourne, VIC



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20180507/dbb47dfc/attachment.html>