[slurm-users] priority: 'job size'-factor scaling parameter

Daan van Rossum d.r.vanrossum at gmx.de
Thu Sep 20 05:34:14 MDT 2018

Dear Slurm users,

Is there a 'Job Size'-factor equivalent of the 'Job Age'-factor's PriorityMaxAge parameter in Slurm?

What I am looking for is a scaling parameter (or threshold parameter) to normalize the job size factor to a value of 1 for '10-core 1-hour' jobs instead of for 10000-core 1-minute jobs.  Does that exist?

This is the issue I'm trying to solve:

We prioritize many-core, short jobs using:
PriorityFavorSmall=NO  #-- A job that requests all the nodes on the machine will get a job size factor of 1.0.
PriorityFlags=SMALL_RELATIVE_TO_TIME  #-- The job size in CPUs is divided by the time limit in minutes.
But users typically use a small fraction of the CPUs for many minutes, leading to job size factors on the order of 1e-6, as evident with 'sprio -n', whereas values on the order of 1 are needed for the weights to work as intended.
(The reason why this can not be tackled by just increasing the size-weight by a factor 10^6 is that this would dominate the Fairshare and Age factors in certain situations.)

Thanks for your help!  D
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20180920/6c7e464c/attachment.sig>

More information about the slurm-users mailing list