2

I've been trying to move some existing processes to a revamped linux cluster that now runs on slurm. I thought I have it done, but my problem now is trying to get multiple cores to run.

Here is my submission script.

   #!/bin/bash
   #
   #SBATCH --job-name=test_mpi
   #SBATCH --output=res_mpi.txt
   #
   #SBATCH -n 4
   #SBATCH --time=10:00
   srun mkdir -p /tmp/tedhyu/new
  srun cp Ru13.in /tmp/tedhyu/new/lcao.in
  srun cp ~tedhyu/atom_pbe/* /tmp/tedhyu/new
  srun cd /tmp/tedhyu/new
  srun -N 1  -n 4 --chdir=/tmp/tedhyu/new  mpiexec ~tedhyu/bin/origin1_centos6.4_mpich2_quest_265c.x

When I "qstat -n" it only shows one core:

Job id Username Queue Name SessID NDS TSK Memory Time Use S Time


11778 tedhyu atom test_mpi -- 1 4 -- 00:10 C 00:00
node3-5/4

Here is the first few lines of my output that shows only 1 core is running:

    srun: error: node3-5: tasks 0-3: Exited with exit code 1
     MPINFO::: Global Communicator        :::
     MPINFO::: Global Context = ****      :::
     MPINFO::: Global Size =       1      :::
     MPINFO::: Global Root =       0      :::
     MPINFO::: Global Rank =       0      :::
     DEV: VDW development version

Global Size should equal 4

If anyone can point me in the right direction... Thanks!!!

1
  • 1
    thanks. Sorry, just starting out on this.
    – ted y
    Jan 26, 2021 at 18:51

1 Answer 1

0

do not use srun in the last line of the script. Just launch the executable there with mpirun or mpiexec.

what srun does : it is launches $SLURM_NTASKS instances ( one per reserved CPU core) of the command which comes after it. You don't want that, you want mpiexec to spawn the tasks to the CPUs. E.g. your last line can be

mpirun -np $SLURM_NTASKS ./myexecutable.exe

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .