Submitting parallel jobs to OctansCluster via LSF

MPICH2 is the MPI implementation on OctansCluster. MPICH2 is integrated with the LSF scheduler via LSF HPC. Therefore, all MPICH2 jobs need to be properly submitted to the cluster via LSF. (user.pdf: MPICH2 user manual)

All MPICH2 to jobs are submitted to the HPC_linux queue by default. Users are free to request up to 112 procesors per job. If you need more than 112 processors, contact ACS via HelpDesk or directly to get help. If you simply request more than 112CPUs, your job will remain pending.

MPICH2 on OctansCluster was built using Intel compilers. Before you attempt to compile your code, make sure you have the following directories in your PATH variable:

/usr/local/intel/cc/bin

/usr/local/intel/fce/bin

Use mpicc -o to compile your source code (/usr/local/mpich2-intel/bin/mpicc).

/usr/local/mpich2-intel/bin/mpicc -o myjob.mpich2 myjob.c

Submit this job to Platform LSF HPC:

Submit this job from console command line:
bsub <bsub options> -n ### -a mpich2 mpirun.lsf <mpiexec options> job <job options>
Note that -np options of mpiexec will be ignored.
For example:
bsub -a mpich2 -n 10 -o /home/jklc/mpiout.txt  mpirun.lsf /home/jklc/mpich2inteltest/time_bcast_nochk  
In this example, a /home/jklc/mpich2inteltest/time_bcast_nochk MPICH2 job is being submitted requesting 10 CPUs, writting output to /home/jklc/mpiout.txt.

Submit this job using a script bsub < myjobscript.sh

where myjobscript.sh looks like:

#!/bin/sh

    #BSUB -n 8

    #BSUB -a mpich2

    mpirun.lsf ./home/jklc/mpich2inteltest/time_bcast_nochk

-- JaroKlc - 30 Jun 2008