MPI

MPI

MPI (Message Passing Inerface)[1] is a library specification for message passing standardized by committee of vendors, implementors, and users

Compiling and Running MPI jobs

Copy the hello.mpi  file.

#include <mpi.h>

#include <stdio.h>

int main(int argc, char** argv) {

    // Initialize the MPI environment

    MPI_Init(&argc, &argv);

    // Get the number of processes

    int world_size;

    MPI_Comm_size(MPI_COMM_WORLD, &world_size);

    // Get the rank of the process

    int world_rank;

    MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);

    // Get the name of the processor

    char processor_name[MPI_MAX_PROCESSOR_NAME];

    int name_len;

    MPI_Get_processor_name(processor_name, &name_len);

    // Print off a hello world message

    printf("Hello, World! from processor %s, rank %d out of %d processors\n",

           processor_name, world_rank, world_size);

    // Finalize the MPI environment

    MPI_Finalize();

    return 0;

}

Load OpenMPI module:

module load OpenMPI/4.0.3-GCC-9.3.0

Compile the code:

mpicc -o hello hello.c

Execute using 4 MPI tasks (e.g. -n 4). For Markov cluster,  you need to use different partition and class account (e.g. -A csds438 -p markov_cpu).

srun -n 4 hello

output:

Hello, World! from processor classct015, rank 0 out of 4 processors

Hello, World! from processor classct015, rank 1 out of 4 processors

Hello, World! from processor classct015, rank 2 out of 4 processors

Hello, World! from processor classct015, rank 3 out of 4 processors

References:

[1] MPI Standard: http://www.mcs.anl.gov/research/projects/mpi/