How do I run a job using LAM/MPI?


To run your LAM/MPI on our debugging cluster use the lamrun script.

Syntax: lamrun nodes executable
Example: lamrun 4 a.out this would start a.out on 4 nodes.

If that's too much automation for you, you can also do it like this:
From your home directory enter:

% lamboot -d hostsfile

This starts the lam daemons on those machines listed in the hostsfile.
The "-d" option prints debug information - it is nice to see what it is doing.
Note: the LAM daemon will stay "resident" even when you logout.

Running your mpi program:

To run 4 instances of your program a.out in client to client mode (fastest):
% mpiexec 4 a.out

To run your program in debugging mode:
# mpidebug 4 a.out

The shell scripts mpiexec & mpidebug will run lamclean after your job is finished, this will clean up any traces of your mpi program.
You can run your mpi job manually by using mpirun (see man mpirun for syntax, if you do so you should execute lamclean after each run yourself.

Killing the lam daemons.

If for some reason you want to stop the lam daemons on all your nodes you use the command
% lamhalt

Note: Only on the interactive nodes (node-101 through node-108) (debugging cluster) will you be able to control the execution of your mpi job.
To run your jobs on the production cluster you will need to submit your job to the scheduler.

 


If you have any questions regarding this topic e-mail to support@stat.ucla.edu

Date Created: 2003-01-30 17:30:04 Date Last Modified: 2003-11-03 16:20:15


UCLA Department of Statistics
Last updated: 13-Jul-2009
Access count is: 40447, since 02-Oct-2003
Maintained by: Web Staff [webstaff@stat.ucla.edu]