[slurm-users] Multinode MPI job

[Mahmood Naderan]

 >If your SLURM version is at least 18.08 then you should be able to do it
with an heterogeneous job. See
https://slurm.schedmd.com/>heterogeneous_jobs.html
<https://slurm.schedmd.com/heterogeneous_jobs.html>


>From the example in that page, I have written this

#!/bin/bash
#SBATCH --job-name=myQE
#SBATCH --output=big-job
#SBATCH --mem-per-cpu=10g --ntasks=8
#SBATCH packjob
#SBATCH --mem-per-cpu=10g --ntasks=2
#SBATCH --partition=QUARTZ
#SBATCH --account=z5
mpirun pw.x -i mos2.rlx.in


So, I expect that first node runs with 8 cores and 80GB of memory and the
second node runs with 2 cores and 20GB or memory.

This is what I see in the output


$ squeue
             JOBID PARTITION     NAME     USER ST       TIME  NODES
NODELIST(REASON)
             723+0   CLUSTER     myQE   mahmood     R       2:37      1
compute-0-1
             723+1    QUARTZ     myQE   mahmood     R       2:37      1
compute-0-2
$ rocks run host compute-0-1 "ps aux | grep pw.x"
mahmood     25667  0.0  0.0 289544 13224 ?        Sl   11:32   0:00 mpirun
pw.x -i mos2.rlx.in
mahmood     25672 99.4  2.0 3784972 2316076 ?     Rl   11:32   2:49 pw.x -i
mos2.rlx.in
mahmood     25673 99.4  2.0 3783544 2314008 ?     Rl   11:32   2:50 pw.x -i
mos2.rlx.in
mahmood     25674 99.4  2.0 3785532 2314196 ?     Rl   11:32   2:50 pw.x -i
mos2.rlx.in
mahmood     25675 99.2  2.0 3787648 2316048 ?     Rl   11:32   2:49 pw.x -i
mos2.rlx.in
mahmood     25676 99.4  2.0 3786600 2313916 ?     Rl   11:32   2:50 pw.x -i
mos2.rlx.in
mahmood     25677 99.4  2.0 3786344 2314056 ?     Rl   11:32   2:50 pw.x -i
mos2.rlx.in
mahmood     25678 99.4  2.0 3782632 2313892 ?     Rl   11:32   2:50 pw.x -i
mos2.rlx.in
mahmood     25679 99.4  2.0 3784112 2313856 ?     Rl   11:32   2:50 pw.x -i
mos2.rlx.in
mahmood     25889  1.0  0.0 113132  1588 ?        Ss   11:35   0:00 bash -c
ps aux | grep pw.x
mahmood     25925  0.0  0.0 112664   960 ?        S    11:35   0:00 grep
pw.x
$ rocks run host compute-0-2 "ps aux | grep pw.x"
mahmood     28296  0.0  0.0 113132  1588 ?        Ss   11:35   0:00 bash -c
ps aux | grep pw.x
mahmood     28325  0.0  0.0 112664   960 ?        S    11:35   0:00 grep
pw.x



So, compute-0-2 has no pw.x process.
Also, the partition names are weird. We have these entries:

$ cat /etc/slurm/parts
PartitionName=WHEEL RootOnly=yes Priority=1000 Nodes=ALL
PartitionName=RUBY AllowAccounts=y4,y8 Nodes=compute-0-[1-4]
PartitionName=EMERALD AllowAccounts=z2,z33,z7 Nodes=compute-0-[0-4],rocks7
PartitionName=QEMU AllowAccounts=q20_8 Nodes=compute-0-[1-4],rocks7
PartitionName=QUARTZ AllowAccounts=z5 Nodes=compute-0-[1-2],compute-0-4



Any thought?

Regards,
Mahmood
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20190327/ab8024f7/attachment-0001.html>