[slurm-users] DefMemPerGPU bug?

Bas van der Vlies bas.vandervlies at surfsara.nl
Mon Mar 30 10:48:14 UTC 2020

We have the same issue see:
  * https://bugs.schedmd.com/show_bug.cgi?id=8527
  * temporary fix we switched back to DefMemPerCpu


On 26/03/2020 16:42, Wayne Hendricks wrote:
> When using 20.02/cons_tres and defining DefMemPerGPU, jobs submitted 
> that request GPUs without defining “—mem” will not run more than one job 
> per node. I can see where it is allocating the correct amount of memory 
> for the job per GPUs requested, but no other jobs will run on the node. 
> If a value for “—mem” is defined, other jobs will share the node. Is 
> this the expected behavior? I understand that when jobs do not request 
> memory it is assumed that the job is running on the whole node, but here 
> when we are asking for GPUs there is a default memory set with 
> DefMemPerGPU and it seems this is not being taken into account. Let me 
> know if there is a reason for this behavior or if there is another way 
> to set the default job memory.
> Config:
> SelectType=select/cons_tres
> SelectTypeParameters=CR_Core_Memory,CR_CORE_DEFAULT_DIST_BLOCK
> PartitionName=p100 Nodes=ucs480 OverSubscribe=FORCE:4 DefCpuPerGPU=20 
> DefMemPerGPU=125000 Default=YES MaxTime=INFINITE State=UP
> Node and job state when two jobs submitted with each requesting half the 
> GPUs (no —mem specified):
>     CfgTRES=cpu=80,mem=500000M,billing=80
>     AllocTRES=cpu=40,mem=250000M
> Job state:
> 872      p100  test-s6 wayne.he PD       0:00      1 (Resources)
> 871      p100  test-s5 wayne.he  R       0:03      1 ucs480

Bas van der Vlies
| Operations, Support & Development | SURFsara | Science Park 140 | 1098 XG
| T +31 (0) 20 800 1300 | bas.vandervlies at surfsara.nl | www.surfsara.nl |

More information about the slurm-users mailing list