Dear all, we currently see a change of a default behavior of a job step. On our old cluster (Slurm 20.11.9) a job step take all the resources of my allocation. rotscher@tauruslogin5:~> salloc --partition=interactive --nodes=1 --ntasks=1 --cpus-per-task=24 --hint=nomultithread salloc: Pending job allocation 37851810 salloc: job 37851810 queued and waiting for resources salloc: job 37851810 has been allocated resources salloc: Granted job allocation 37851810 salloc: Waiting for resource configuration salloc: Nodes taurusi6605 are ready for job bash-4.2$ srun numactl -show policy: default preferred node: current physcpubind: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 cpubind: 0 1 nodebind: 0 1 membind: 0 1
If run the same command on our new cluster the job step take only 1 core instead of all without any further paramter. [rotscher@login1 ~]$ salloc --nodes=1 --ntasks=1 --cpus-per-task=24 --hint=nomultithread salloc: Pending job allocation 9197 salloc: job 9197 queued and waiting for resources salloc: job 9197 has been allocated resources salloc: Granted job allocation 9197 salloc: Waiting for resource configuration salloc: Nodes n1601 are ready for job [rotscher@login1 ~]$ srun numactl -show policy: default preferred node: current physcpubind: 0 cpubind: 0 nodebind: 0 membind: 0 1 2 3 4 5 6 7 If I add the parameter „-c 24“ to the job step it also take the hole resources, but the step should take it per default. [rotscher@login1 ~]$ srun -c 24 numactl -show policy: default preferred node: current physcpubind: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 cpubind: 0 1 nodebind: 0 1 membind: 0 1 2 3 4 5 6 7 I searched the slurm.conf documentation, the mailing list and also the changelog, but found no reference to a matching parameter. Do anyone of you know the behavior and how to change it? Best wishes, Danny
smime.p7s
Description: S/MIME cryptographic signature
