Hi Dorian, Thanks for the QoS idea (I had thought of a new partition, but rejected it as a bit messy). I think that limiting the number of jobs is a good enough solution, and works very well in practice.
I have actually tried to implement the QoS now, and it seems to work well. New jobs with this QoS are waiting with reason QOSGrpJobsLimit. Best wishes, Marius ________________________________________ From: Dorian Krause <[email protected]> Sent: 02 September 2016 10:44 To: slurm-dev Subject: [slurm-dev] Re: Global (cluster wide) resources Dear Marius, On 09/02/2016 10:16 AM, Pål Marius Bjørnstad wrote: > We have some jobs which do quite intense I/O on a NAS. I would like to limit > the number of such jobs which can run at any time (so that the NAS can > operate more efficiently, and so the CPU resources aren't wasted on jobs just > waiting for the NAS). if you only need to limit the number of jobs and do not require a more fine-grained control over the I/O resource usage a QoS with GrpJobs limit might be sufficient. Best regards, Dorian ------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------ Forschungszentrum Juelich GmbH 52425 Juelich Sitz der Gesellschaft: Juelich Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498 Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender), Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt, Prof. Dr. Sebastian M. Schmidt ------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------
