MPI Parallel Computing
MPI parallel computing is available on Linux HPC Clusters Atlas4, Atlas5, Atlas6 and Atlas7.
MPI C/C++ and MPI Fortran compiler are available on the cluster. The following are the sample instructions for compiling and running MPI program on the clusters.
Compile and build the program using MPI compiler.
| MPI C: | mpicc -o cprog.exe cprog.c |
| MPI C++: | mpiCC -o cppprog.exe cppprog.cpp |
| MPI F77: | mpif77 -o f77prog.exe f77prog.f |
| MPI F90: | mpif90 -o f90prog.exe f90prog.f90 |
Submit and run the MPI program.
Using HPC clusters, MPI program needs to submit and run in an LSF batch parallel computing queue. The batch parallel queue on each cluster is as below.
| Atlas4 | atlas4_parallel |
| Atlas5 | atlas5_parallel |
| Atlas6 | atlas6_parallel, short_parallel |
| Atlas7 | atlas7_parallel, short_parallel |
You can submit the job using the following command:
bmpijob ./cprog.exe
This is equivalent to submit the job as:-
bmpijob -n 4 -o stdout.o ./cprog.exe
You can also use LSF’s bsub command directly, and submit a MPI job as:-
| Atlas4> | bsub -a mvapich -q atlas4_parallel -n 4 -o std-output.txt mpirun.lsf ./cprog.exe |
| Atlas5> | bsub -a mvapich -q atlas5_parallel -n 8 -o std-output.txt mpirun.lsf ./cprog.exe |
| Atlas6> | bsub -a mvapich -q atlas6_parallel -n 6 -o std-output.txt mpirun.lsf ./cprog.exe |
| Atlas7> | bsub -a mvapich -q atlas7_parallel -n 12 -o std-output.txt mpirun.lsf ./cprog.exe |
where:
| bmpijob | the MPI job submission command, |
| -n 4 | indicates running job on 4 processor cores |
| -o stdout.o | the file to store job information and standard output message |
| ./cprog.exe | to run MPI program “cprog.exe“ |
To submit/run the program with more than 4 CPU Cores, you can use “-n” flag as:
bmpijob -n 32 ./cpgrog.exe
The maximum number of CPU cores allowed per MPI job is 32 at the moment. Once the job is completed, you can check the results at the same working directory.
