To start a parallel job with MPICH2, create a file named mpd.hosts that contains the hostnames of all machines on which the parallel job is to be run, with one machine name per line in the file. You can run the mpdboot utility by logging in to one of the machines on the list, going to the directory with the mpd.hosts file, and typing
mpdboot –n <number of processors>
This will start a process named mpd on each of the machines listed. This background process will listen for send and receive requests. Verify that mpd started properly on all machines by typing
mpdtrace
This command will list all machines that have the mpd process running and are available for use in a parallel job.
You can now start the job with the mpiexec command. You must be logged into one of the nodes in the mpd.hosts list and must be in the directory containing the input files. The most basic way to start CONVERGE in parallel is
mpiexec –n <number of processors> converge
To dump the output to a file and start a four-processor job in the background, use the following command:
mpiexec –n 4 converge < /dev/null > logfile.out &