[MITgcm-support] MPI problems

[Gus Correa]

Hi Neil

Different MPI implementations (MPICH, OpenMPI, Intel MPI, etc)
have different parameters for the
mpirun (or mpiexec) command.
That is not standardized.
Some may call it "-machinefile", others "-hostfile".

"man mpirun"/"man mpiexec" may tell you more about
the right parameters and syntax for the MPI you're using.

Also, make sure the mpirun/mpiexec you're using is part of
the same MPI implementation that you used to compile the MITgcm.

"which mpirun" and "which mpif77"/"which mpif90"/"which mpicc"
will help you check if they match or not.

If they do not match, you need to tweak with your PATH and
LD_LIBRARY_PATH environment variables to have them match.
Using full path ( say /my/mpi/location/mpiexec ) is an alternative.

I hope this helps,
Gus Correa


On 06/07/2013 11:36 AM, Neil Patel wrote:
> I've successfully compiled and ran mitgcmuv on running on a single processor. With MPI it compiled but when running I have issues. Typing mpirun gets:
>
> terra exp4_test/build>  mpirun -np 4 -machinefile mf ./mitgcmuv
> MPI: invalid option -machinefile
> terra exp4_test/build>  mpirun -np 4 mf ./mitgcmuv
> MPI: terra: 0xb1f000051008f8c: /bin/sh: line 0: exec: mf: not found
> MPI: could not run executable (case #4)
> Killed
> terra exp4_test/build>  mpirun -np 4 ./mitgcmuv
> ABNORMAL END: PROGRAM MAIN
> ABNORMAL END: PROGRAM MAIN
> ABNORMAL END: PROGRAM MAIN
> ABNORMAL END: PROGRAM MAIN
>
> Trying different options. Anyone have any clue what the issue might be? I haven't used MPI before and no one else has on this particular machine, so perhaps I'm not using the right calling sequence.
>
> Thanks,
>
> Neil
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support