Open MPI User's Mailing List Archives

Subject: Re: [OMPI users] "almost there" in getting MPI to run
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2011-12-16 09:06:39

• Next message: Jeff Squyres: "Re: [OMPI users] MPI 2 support in sm btl"
• Previous message: Jeff Squyres: "Re: [OMPI users] MPI_BCAST and fortran subarrays"
• In reply to: Joao Amaral: "[OMPI users] "almost there" in getting MPI to run"
• Next in thread: Joao Amaral: "Re: [OMPI users] "almost there" in getting MPI to run"
• Reply: Joao Amaral: "Re: [OMPI users] "almost there" in getting MPI to run"

On Dec 15, 2011, at 9:09 AM, Joao Amaral wrote:

> Strangely(?), in both my laptop and the cluster, the number of threads from the command
>
> call MPI_Comm_size ( MPI_COMM_WORLD, p, error )

Did you really mean "threads" there?

Open MPI is process-parallel, meaning that calling COMM_SIZE on a communicator will return the number of *processes* in that communicator, not threads.

See the man page for mpirun(1) for how to launch multiple processes in a single MPI job. Quick example:

    mpirun --hostfile your_host_file -np 8 your_mpi_program

where:

your_host_file: a file listing the host names to launch on
8: number of copies of your_mpi_program to launch
your_mpi_program: your compiled executable MPI application

-- 
Jeff Squyres
jsquyres_at_[hidden]
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/

• Next message: Jeff Squyres: "Re: [OMPI users] MPI 2 support in sm btl"
• Previous message: Jeff Squyres: "Re: [OMPI users] MPI_BCAST and fortran subarrays"
• In reply to: Joao Amaral: "[OMPI users] "almost there" in getting MPI to run"
• Next in thread: Joao Amaral: "Re: [OMPI users] "almost there" in getting MPI to run"
• Reply: Joao Amaral: "Re: [OMPI users] "almost there" in getting MPI to run"