DCSC logo
+Open all         -Close all
 To use OPEN-MPI to compile and run run parallel application 
 you'll need to take the following 2 steps:

 1. Compile your application.

    OPEN-MPI makes available 4 wrapper functions: mpif77, mpif90, mpicc, and mpiCC.
    They will invoke an underlying compiler and take care of linking against
    the OPEN-MPI mpi libraries. You simply use mpif77 in place of f77 (or pgf77).
    The underlying compilers are the Gnu compilers (gcc, g++, and g77).

    If you wish to use the Intel Fortran or C/C++ compiler you can use the command
    "mpi-selector-menu" to set your environment variables.
    Select "openmpi_intel_qlc-xxx for the Intel compilers and "openmpi_gcc_qlc-xxx"
    to switch back to the gcc compilers again.

 2. Submit a script to the PBS queue.

    Such a script will have to do the following:

    -- request the number of nodes needed
    -- copy input files and executable to /scratch
    -- run the program using 'mpirun'
    -- copy output back to the user account

    an example script is supplied in pbs_open_mpi_script

    This example script also has solutions for a couple of common problems:
    a. How can the environment variables be set up for all processes.
    b. How can input (or output) files be distributed to (collected from)
       the individual execution nodes.

 More information.

 You can read more about OPEN-MPI at the website: http://www.open-mpi.org/
 The OPEN-MPI web-site has a very comprehensive FAQ.

 Please also consult the manpage on mpirun on the frontend:
 man -M /usr/local/lib/openmpi/man mpirun