Re: Spreading parallel across nodes on HPC system

[Christian Meesters]

On 11/11/22 08:05, Ken Mankoff wrote:

> Hi Rob,
>
> On 2022-11-10 at 21:21 +01, Rob Sargent <robjsargent@gmail.com> wrote:
> > I do this, in slurm bash script, to get the number of jobs I want to
> > run (turns out it's better for me to not load the full hyper-threaded
> > count)
> >
> >    cores=`grep -c processor /proc/cpuinfo`
> >    cores=$(( $cores / 2 ))
> >
> >    parallel --jobs $cores etc :::: <file with list of jobs>
> >
> > or sometimes the same jobs many times with
> >
> >    parallel --jobs $cores etc ::: {1..300}
> >       
> I apologize if I am missing something, but I don't see how this solves distributing to different hosts (nodes), where each host may have a different number of CPUs or cores.
>
>   -k.
>
>
>

Quite simple: The slurmstepd is aware of the cluster's configuration. It is hence distributing accordingly to each srun's instance resource demands. srun, however, is greedy (it is per default the MPI starter). Hence, the '--mem-per-cpu' restriction per call.

As to the question of the environment variable `SLURM_CPUS_PER_TASK`: If your cluster is configured as usual, it is '1' per default. If not, leave it away. For one core application its not necessary. As said: it's a template.

Best regards

Christian