Hi Ulf,

Ulf Markwardt <[email protected]> writes:

> Dear list,
>
> we run our Slurm in a shared mode, our nodes have different sizes of memory
> installed.
>
> How can I make sure that a job can only be scheduled if the remaining cores 
> and
> the remaining memory on each node have a ratio of 400MB/core?
>
> We have managed it in our older systems without a problem, and would like to
> install this functionality to our Slurm system as well.
>
> Thank you,
> Ulf

We have the same set-up.  We decided to just give different weights to
the different node categories.  This way a job will be scheduled on a
lower-memory node rather that a higher-memory one, if both are
available.

This still means of course that you can have jobs using most of a node's
memory, but only a few cores.  However, this is probably unavoidable if
you have an unpredictable mix of jobs.

Just my 2€¢.

Cheers,

Loris

-- 
Dr. Loris Bennett (Mr.)
ZEDAT, Freie Universität Berlin         Email [email protected]

Reply via email to