MPI on Triton¶
The basic parallel programming model for PC-clusters is message passing. MPI (Message Passing Interface) is a library specification for message-passing, proposed, nowadays, as a standard. OpenMPI and MVAPICH are two different (but doing the same) MPI implemetaions, both are installed on triton. MVAPICH is the recommended one. MPI is suitable for both distributed memory computers (thus through interconnect between nodes) and shared memory architectures (computing within one node).
Check out the latest version of MVAPICH / OpenMPI with
$ module avail
By setting up your environment with
module, you will set all the
$ module load mvapich2/1.8a2-gcc-4.4.6
Load OpenMPI (for software compiled on recently updated SL6.2):
$ module load openmpi/1.4.5-gcc-4.4.6
Load OpenMPI (for software compiled on SL6.1):
$ module load compat-openmpi-x86_64
Compiling after that is quite straight forward:
$ mpif90 your_code.f90 -o your_code $ mpicc your_code.c -o your code
Parallel programs are to be run as batch jobs, even test runs. See Executing jobs / Batch system for examples of running in parallel.