UK Logo High Performance Computing August 28, 2008
Home News System Status User Policies Account Info Hardware Software Documentation FAQs Search
mpirun - Running MPI on the HP Superdome cluster


   MPI Documentation  
HP Superdome Cluster


Running MPI Applications

This section introduces the methods used to run an HP MPI application. The examples below demonstrate three basic methods. Refer to the Command Line Options section of this document for all the mpirun command line options.

You should use the -j option to display the MPI job ID. The job ID is useful during troubleshooting to check for a hung job using mpijob or terminate a job using mpiclean.

There are four methods you can use to start your application. The first two are useful for short interactive (test) jobs, or in script files for jobs that only run on one node; the others are more general.



Command Line Options

mpirun starts an HP MPI application. Syntax for the command follows:


where

-np #i
Specifies the number of processes to run. (Used when not running under LSF).


-help
Prints usage information for the utility.


-version
Prints the version information.

-j
Prints the HP MPI job ID.

-p
Turns on pretend mode. That is, the system goes through the motions of starting an HP MPI application but does not create processes. This is useful for debugging.

-ck
Behaves like the -p option, but supports two additional checks of your MPI application; it checks if the specified host machines and programs are available, and also checks for access or permission problems.

-v
Turns on verbose mode.

-d
Turns on debug mode.

-w
Wait for all application processes to exit before returning. This is the default behavior.

-W
Do not wait for the application to terminate before returning.

-t spec
Enables runtime trace generation for all processes. spec specifies options used when tracing. The options are the same as those for the environment variable MPI_XMPI. For example, the following is a valid command line:

mpirun -t mytrace:off:nc -np 4 myprog

-i spec
Enables runtime instrumentation profiling for all processes. spec specifies options used when profiling. The options are the same as those for the environment variable MPI_INSTR. For example:

mpirun -i mytrace:nd:nc -np 4 myprog

-h host
Specifies a host on which to start the processes (default is the local host).

-l user
Specifies the username on the target host (default is local username).

-e var[=val]
Sets the environment variable var for the program and gives it the value val if provided. Environment variable substitutions (for example, $FOO) are supported in the val argument.

-sp paths
Sets the target shell PATH environment variable to paths. Search paths are separated by a colon.

program
Specifies the name of the executable file to run.

args
Specifies command-line arguments to the program - a space separated list of arguments.

lsf_options
Specifies bsub options that the load-sharing facility (LSF) applies to the entire job (that is, every host). Refer to the bsub(1) man page for a list of options you can use.

runpam
Script (either the supplied example, available at /usr/local/bin/runpam or a script supplied by the user) which gets a list of hosts to be used from LSF and internally executes mpirun to run the job on the hosts selected by LSF. This is superior to selecting hosts by hand because it allows LSF to better control scheduling and reduce conflicts. If you wish to write your own script for the purpose, the actual name of the script may be whatever you desire. Look at /usr/local/bin/runpam for an example of how to write such a script.

-n #
Specifies the number of processors required to LSF. Required when using runpam.

other_lsf_options
As lsf_options - but note that the -n # option is required and shown seperately.

-commd
Routes all off-host communication through the daemons rather than between processes.

Send comments/questions to: help-hpc@uky.edu
Last modified:
August 11 2003 09:29:22.