mpirun

mpirun is a program which is required to run other mpi based programs.

Description

“mpirun” is a shell script that attempts to hide the differences in
starting jobs for various devices from the user. Mpirun attempts to
determine what kind of machine it is running on and start the required
number of jobs on that machine. On workstation clusters, if you are
not using Chameleon, you must supply a file that lists the different
machines that mpirun can use to run remote jobs or specify this file
every time you run mpirun with the -machine file option. The default
file is in util/machines/machines.<arch>.

mpirun typically works like this

 mpirun -np <number of processes> <program name and arguments>

If mpirun cannot determine what kind of machine you are on, and it
is supported by the mpi implementation, you can the -machine
and -arch options to tell it what kind of machine you are running
on. The current valid values for machine are

          chameleon (including chameleon/pvm, chameleon/p4, etc...)<br />          meiko     (the meiko device on the meiko)<br />          paragon   (the ch_nx device on a paragon not running NQS)<br />          p4        (the ch_p4 device on a workstation cluster)<br />          ibmspx    (ch_eui for IBM SP2)<br />          anlspx    (ch_eui for ANLs SPx)<br />          ksr       (ch_p4 for KSR 1 and 2)<br />          sgi_mp    (ch_shmem for SGI multiprocessors)<br />          cray_t3d  (t3d for Cray T3D)<br />          smp       (ch_shmem for SMPs)<br />          execer    (a custom script for starting ch_p4 programs<br />                     without using a procgroup file. This script<br />                     currently does not work well with interactive<br />                     jobs)<br />

You should only have to specify mr_arch if mpirun does not recognize
your machine, the default value is wrong, or you are using the p4 or
execer devices. The full list of options is

Here is a copy of the official mannual for mpirun

Synopsis:       mpirun [options] <app>
                mpirun [options] <where> <program> [<prog args>]

Description:    Start an MPI application in LAM/MPI.

Notes:
                [options]       Zero or more of the options listed below
                <app>           LAM/MPI appschema
                <where>         List of LAM nodes and/or CPUs (examples
                                below)
                <program>       Must be a LAM/MPI program that either
                                invokes MPI_INIT or has exactly one of
                                its children invoke MPI_INIT
                <prog args>     Optional list of command line arguments
                                to <program>

Options:
                -c <num>        Run <num> copies of <program> (same as -np)
                -client <rank>  <host>:<port>
                                Run IMPI job; connect to the IMPI server <host>
                                at port <port> as IMPI client number <rank>
                -D              Change current working directory of new
                                processes to the directory where the
                                executable resides
                -f              Do not open stdio descriptors
                -ger            Turn on GER mode
                -h              Print this help message
                -l              Force line-buffered output
                -lamd           Use LAM daemon (LAMD) mode (opposite of -c2c)
                -nger           Turn off GER mode
                -np <num>       Run <num> copies of <program> (same as -c)
                -nx             Don’t export LAM_MPI_* environment variables
                -O              Universe is homogeneous
                -pty / -npty    Use/don’t use pseudo terminals when stdout is
                                a tty
                -s <nodeid>     Load <program> from node <nodeid>
                -sigs / -nsigs  Catch/don’t catch signals in MPI application
                -ssi <n> <arg>  Set environment variable LAM_MPI_SSI_<n>=<arg>
                -toff           Enable tracing with generation initially off
                -ton, -t        Enable tracing with generation initially on
                -tv        Launch processes under TotalView Debugger
        -v              Be verbose
                -w / -nw        Wait/don’t wait for application to complete
                -wd <dir>       Change current working directory of new
                                processes to <dir>
                -x <envlist>    Export environment vars in <envlist>

Nodes:          n<list>, e.g., n0-3,5
CPUS:           c<list>, e.g., c0-3,5
Extras:         h (local node), o (origin node), N (all nodes), C (all CPUs)

Examples:       mpirun n0-7 prog1
                Executes “prog1″ on nodes 0 through 7.

                mpirun -lamd -x FOO=bar,DISPLAY N prog2
                Executes “prog2″ on all nodes using the LAMD RPI. 
                In the environment of each process, set FOO to the value
                “bar”, and set DISPLAY to the current value.

                mpirun n0 N prog3
                Run “prog3″ on node 0, *and* all nodes.  This executes *2*
                copies on n0.

                mpirun C prog4 arg1 arg2
                Run “prog4″ on each available CPU with command line
                arguments of “arg1″ and “arg2″.  If each node has a
                CPU count of 1, the “C” is equivalent to “N”.  If at
                least one node has a CPU count greater than 1, LAM
                will run neighboring ranks of MPI_COMM_WORLD on that
                node.  For example, if node 0 has a CPU count of 4 and
                node 1 has a CPU count of 2, “prog4″ will have
                MPI_COMM_WORLD ranks 0 through 3 on n0, and ranks 4
                and 5 on n1.

                mpirun c0 C prog5
                Similar to the “prog3″ example above, this runs “prog5″
                on CPU 0 *and* on each available CPU.  This executes
                *2* copies on the node where CPU 0 is (i.e., n0).
                This is probably not a useful use of the “C” notation;
                it is only shown here for an example.

Defaults:       -c2c -w -pty -nger -nsigs
—————————————————————————–