On Feb 14, 2006, at 2:08 PM, Mike Haskell wrote:

odev2@mhaskell:  srun -N4 -n4 --nodelist=odev[0-3,6] `pwd`/hello
1 of 4:  Hello world!
3 of 4:  Hello world!
2 of 4:  Hello world!
0 of 4:  Hello world!

When I run it with OpenMPI (openmpi-1.0.2a4), I get:

odev2@mhaskell:  srun -N4 -n4 --nodelist=odev[0-3,6] `pwd`/hello
0 of 1:  Hello world!
0 of 1:  Hello world!
0 of 1:  Hello world!
0 of 1:  Hello world!

This is actually not a mode that we support in SLURM [yet] -- we currently require the use of mpirun. For example:

        srun -N4 -n4 -A ...
        mpirun -np 4 hello

The "-A" option to srun will give you an interactive shell where you can run mpirun as many times as you'd like. The mpirun will automatically get the hostnames from SLURM -- no need to provide a hostfile or anything like that.

Similarly, you can write a script and submit it to slurm with the -b switch (where only one copy of the script runs, not N copies). You'd put the mpirun command in there, and just like with the interactive mode, it'll get the hosts from SLURM, etc.

(FWIW, we use SLURM on our clusters too -- these two modes have worked quite well for us; let us know if you have problems with them)

Hope that helps!

--
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/


Reply via email to