MPI Hello World

MPI Hello World

Many parallel jobs are using MPI at the lowest level to manage parallel compute resources.

This is a 'Hello World' program that will test the operation of sending jobs to remote workers.

 /*
 * Sample MPI "hello world" application in C
 */


#include <stdio.h>
#include "mpi.h"

int main(int argc, char* argv[])
{
    int rank, size, len;
    char version[MPI_MAX_LIBRARY_VERSION_STRING];

    MPI_Init(&argc, &argv);
    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
    MPI_Comm_size(MPI_COMM_WORLD, &size);
    printf("Hello, world, I am %d of %d \n", rank, size);
    MPI_Finalize();

    return 0;
}
 

create a folder and save this file as hello_mpi.c

You can compile this program with

mpicc -g hello_mpi.c -o hello_mpi

You now have an executable called hello_mpi

You can run the command with

mpirun ./hello_mpi

With no other arguments, mpirun will run all of the tasks on the local machine, usually one for each CPU core. You should see a hello world from each process, on each cpu core.

Now, to run your mpi program on the cluster, you will need to create a hostfile. First, lets create a simple hostfile that just runs four processes on the local machine

localhost slots=4 max_slots=8
 

Put that line in a file called localhost . Now run your program with that hostfile, using

mpirun --hostfile localhost ./hello_mpi

Hello, world, I am 0 of 4 
Hello, world, I am 2 of 4 
Hello, world, I am 1 of 4 
Hello, world, I am 3 of 4 
 

Now create a hostfile called cluster_hosts with the following entries:

pnode01 slots=4 max_slots=8
pnode02 slots=4 max_slots=8
pnode03 slots=4 max_slots=8
pnode04 slots=4 max_slots=8
 

and keep adding lines until you get to pnode40 .

The next step assumes you have set up your ssh keys as described in With 40 nodes listed in your cluster_hosts file, run your program again with

mpirun --hostfile cluster_hosts ./hello_mpi