On Fri, 2016-06-10 at 07:45:08 -0700, [email protected] wrote:
>
> See scheduling tutorial starting on page 27
> http://slurm.schedmd.com/SUG14/sched_tutorial.pdf
Thanks for the pointer. Information seems to be scattered micely, as there's
no indication of the preempt_* parameters close to the description of
Preempt{Mode,Type} in the slurm.conf man page, and it's easily overlooked
twelve pages later. (With 63 pages for 15.08, this man page might be in
need of an alphabetical index.)
What I cannot find is how to set a preference (young vs old jobs, small vs
large, ...), or a detailed description of reordering beyond node counts -
we expect most of our preemptees to be of the same size but different
"run age", and reordering (as described in the slides) would be of little help.
> Smaller jobs tend to get preempted larger jobs.
At least one word seems to be missing, and as I said before, size isn't
everything.
Is there a way to control which signal(s) get(s) sent, in which order,
and how much time apart?
Thank you,
Steffen
>
> On 2016-06-10 07:52, Paul Edmon wrote:
> >I'm interested in this as well. It would be good to have a few knobs
> >to fine tune which jobs preferentially get requeued. I know it adds
> >overhead to the scheduling but the benefit to us outweighs that.
> >Essentially it would be good to both preempt jobs that are small and
> >short before the big ones or have been running for a long time.
> >
> >-Paul Edmon-
> >
> >On 06/10/2016 02:36 AM, Steffen Grunewald wrote:
> >>Good morning everyone,
> >>
> >>is there a way to control the order in which jobs get preempted?
> >>That is, for a queue with PreemptMode=REQUEUE, it would make sense to
> >>preempt jobs first that haven't run for long; for CANCEL with some
> >>GraceTime, perhaps addressing the longest-running jobs first would
> >>be better - can Slurm's strategy be hinted twoards the "right" way?
