CompilingSMP

Using MPI with the optimize and grid commands is described on the MpiParallel page.

Compiling Cloudy for use on Symmetric Multiprocessing machines

This is a very rough outline of how to get a grid running on the new (February 2007) IBM xSeries supercomputer in Lexington.

To compile a wrapper program in cloudy_grid.cpp:

mpiCC -DMPI_ENABLED -c *.cpp

mpiCC -DMPI_ENABLED -o cloudy_grid.exe *.o -lm

The flag DMPI_ENABLED is used by Cloudy to signify that the MP libraries are being invoked. It should also be used in your wrapper routine. To submit the job to the batch handler, use:

srun -b -nX mpi.cs cloudy_grid

where X is the number of processors to use, and "cloudy_grid" is the prefix of the executable and will be assigned to the variable "$1" in the script mpi.cs. That script should look like this:

#!/bin/csh

mpirun $1.exe

exit 0

.

The period on the last line is required.

Use showq and squeue for queue information.

Stderr goes to slurm-XX.out where XX is jobid.

Kill a job with: scancel job_id