Hi Greg, The qsub command performs resource (number of nodes) allocation. It does not assign processes to nodes. The latter is done by mpirun (mpiexec) command. Once you start an interactive session, do the following export MODULEPATH=/soft/restricted/CNDA/modulefiles module load oneapi module load mpi/aurora_mpich/drop34 mpirun --help Check -n and -ppn options. The first one declares the total number of MPI processes to be started, and the second option tells how many of those will be running on each node. This explanation presumes that your application supports MPI parallel programming model. This is what we are using for distribute parallelism. If you use a different parallel programming model, please let us know. Thank you, Victor -- Victor Anisimov, PhD Application Performance Engineer Argonne Leadership Computing Facility 9700 S. Cass Ave, Bldg 240, Room 1121 Lemont, IL 60439
On Jan 7, 2021, at 8:46 AM, Kosinovsky, Greg via Ecp-aurora-sdk-discuss <[email protected]> wrote:
I need to run a program on Iris (or another Aurora machine) where there are 3 processes on 3 nodes (1 process per node). I am using qsub in interactive mode – and I am new to qsub. I have experimented with using proccount flag, so I either run “qsub -q iris -t 0:30:00 -n 3 --proccount 3 -I” or “qsub -q iris -t 0:30:00 -n 3 --proccount 1 -I” and then start my program with 3 processes within the session. In either case, all 3 processes are launched on a single node – so proccount is clearly not the right flag to regulate this. IS there a qsub flag that would allow me to distribute the processes -- Ecp-aurora-sdk-discuss mailing list [email protected] https://lists.jlse.anl.gov/mailman/listinfo/ecp-aurora-sdk-discuss