Running MPICH program

MPICH is one of the most popular implementations of MPI.

In this tutorial, we aim to introduce that in a CSCI4160 cluster with 3 machines: node-1, node-2, and node-3.

How to compile MPI program

Here we will use mpi_hello_world in https://mpitutorial.com/tutorials/ as an example.

SSH to node-1

You can login node-1 by using your CSE account just like linux9.

Clone the mpitutorial git repository

cd ~ # make sure that your are now in the home directory 
git clone https://github.com/wesleykendall/mpitutorial
cd  ~/mpitutorial/tutorials/mpi-hello-world/code

Compile the mpi_hello_world using mpicc

Instead of using gcc, we use mpicc such that the mpich library is linked.

mpicc -o mpi_hello_world mpi_hello_world.c

Now you should get the executable mpi_hello_world

Run MPI program in CSCI4160 cluster using SLURM

We have setup a small SLURM cluster consists of node-1, node-2 and node-3. Use salloc command to launch your mpi program on that cluster:

salloc -N 3 mpirun ./mpi_hello_world

-N 3: request 3 nodes from the cluster

node-1:~/mpitutorial/tutorials/mpi-hello-world/code$ salloc -N 3 mpirun ./mpi_hello_world
salloc: Granted job allocation 23695
Hello world from processor node-3.cse.cuhk.edu.hk, rank 2 out of 3 processors
Hello world from processor node-1.cse.cuhk.edu.hk, rank 0 out of 3 processors
Hello world from processor node-2.cse.cuhk.edu.hk, rank 1 out of 3 processors
salloc: Relinquishing job allocation 23695

You can see that there are 3 Hello world messages from 3 processors with different hostnames: node-1 node-2 node-3

MPICH Programming

I suggest you go through the tutorials below, which will help you a lot in Asgn1c.

Name	link
MPI Send and Receive	link
Dynamic Receiving with MPI_Probe (can help to query the number of permissible points on another machine)	link
Custom MPI data type (you may use this to ship struct point datatype)	link

In the links above, they launch MPI using a scriptrun.py . Using salloc to launch MPI is also OK.