[slurm-users] clarification on slurm scheduler and the "nice" parameter

[Lyn Gerner]

Hi Matteo,

Hard to say without seeing your priority config values, but I'm guessing
you want to take a look at
https://slurm.schedmd.com/priority_multifactor.html.

Regards,
Lyn

On Tue, Apr 14, 2020 at 12:02 AM Matteo F <mfasco984 at gmail.com> wrote:

> Hello there,
> I am having problems understanding the slurm scheduler, with regard to the
> "nice" parameter.
>
> I have two types of job: one is low priority and uses 4 CPUs (--nice=20),
> the other one is high priority and uses 24 CPUs (--nice=10).
> When I submit, let's say, 50 low-priority jobs, only 6 are executed - this
> is fine since a job uses 4 CPUs and the node has 24.
> However, when I submit my high priority job that must use 24 CPUs, things
> get strange.
>
> What I was expecting:
> - slurm would have stopped starting low-priority queued jobs (switching
> from PD -> R)
> - waited to have 24 CPUs free (in this case, to have no running jobs)
> - run the high priority job
> - when the job has completed, start the low priority jobs as usual
>
> What I instead observed:
> - slurm keep starting queue job like I didn't specified a nice parameter.
>
>
> (partial) slurm config:
> SwitchType=switch/none
> TaskPlugin=task/none
> FastSchedule=1
> SchedulerType=sched/backfill
> SelectType=select/cons_res
> SelectTypeParameters=CR_Core_Memory
> NodeName=node01 CPUs=24 RealMemory=120000 Sockets=2 CoresPerSocket=6
> ThreadsPerCore=2   State=UNKNOWN
>
> Low priority job:
> #SBATCH --job-name=task4
> #SBATCH --ntasks=4
> #SBATCH --mem=1gb
> #SBATCH --time=10:00:00
> #SBATCH --output=%j.out
> #SBATCH --error=%j.err
> #SBATCH --partition=ogre
> #SBATCH --account=ogre
> #SBATCH --nice=20
>
> High priority job:
> #SBATCH --job-name=task24
> #SBATCH --ntasks=24
> #SBATCH --mem=1gb
> #SBATCH --time=10:00:00
> #SBATCH --output=%j.out
> #SBATCH --error=%j.err
> #SBATCH --partition=ogre
> #SBATCH --account=ogre
> #SBATCH --nice=10
>
> Do you have any idea of what I am missing?
>
> Thanks a lot.
> Matteo
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20200414/8fff72cc/attachment.htm>