According to SLURM QOS document, one can specify a Quality of Service (QOS) for each job submitted to SLURM. So one can use different QOSs in order to send jobs to different partitions. But there should be a machanism that force the user to use only the related QOS while sending jobs to the partitions. I mean that, if I have two partitions (and qos ) like PartitionHigh (QOShigh) and PartitionLow (QOSlow), the user should be able to send jobs to PartitionHigh olnly using the QOSHigh. Using QOSHigh on the partition PartitionLow somehow should be forbidden. I couldn't find such a mechanism on SLURM documents.

Furthermore, is there a mechanism to limit gres usage like maxGPU per group,user or job which may be defined in QOS definition (or account/user/partition definition)?


On 02/25/2013 07:22 PM, Danny Auble wrote:
Using one partition with multiple QOS is probably the simplest way to potentially satisfy your needs. See
http://www.schedmd.com/slurmdocs/qos.html

On a side note http://www.mail-archive.com/[email protected] points to emails from the old LLNL list only.

To find a full archive you should use

http://dir.gmane.org/gmane.comp.distributed.slurm.devel
or
http://groups.google.com/group/slurm-devel

as listed in the docs, http://www.schedmd.com/slurmdocs/mail.html

On 02/25/2013 03:05 AM, Sébastien Kozlowskyj wrote:

Hi all,

I ‘am new user to SLUMR and I want to know : What is the best way to limit the maximum number of CPU used by each user in one partition but not in others ? By Association and/or QOS ?

I have two partitions:

scontrol show partition

PartitionName=Def

   AllocNodes=ALL AllowGroups=ALL Default=YES

   DefaultTime=NONE DisableRootJobs=NO GraceTime=0 Hidden=NO

MaxNodes=1 MaxTime=5-00:00:00 MinNodes=1

   Nodes=dr[01-26]

Priority=10000 RootOnly=NO ReqResv=NO Shared=NO PreemptMode=OFF

State=UP TotalCPUs=416 TotalNodes=26 DefMemPerNode=UNLIMITED MaxMemPerNode=UNLIMITED

PartitionName=Long

   AllocNodes=ALL AllowGroups=ALL Default=NO

   DefaultTime=NONE DisableRootJobs=NO GraceTime=0 Hidden=NO

MaxNodes=1 MaxTime=21-00:00:00 MinNodes=1

   Nodes=dr[01-26]

Priority=5000 RootOnly=NO ReqResv=NO Shared=NO PreemptMode=OFF

State=UP TotalCPUs=416 TotalNodes=26 DefMemPerNode=UNLIMITED MaxMemPerNode=UNLIMITED

I would like to limit the maximum number of CPU used by user on ‘long’ partition, e.g. 32, but not on the ‘def’ partition.

Thanks in advance,

Sébastien,

******************************************************
Sébastien Kozlowskyj
Administrateur Système
Service de Chimie des Matériaux Nouveaux

Centre de Recherche en Electronique et Photonique Moléculaires
Université de Mons

Place du Parc, 20
B-7000 Mons (Belgique)
e-mail: [email protected] <mailto:[email protected]>

Tel.: ++32-(0)65-373866
Fax: ++32-(0)65-373861
******************************************************


Reply via email to