MPI4Py

Before you start

This article explains how to use MPI4Py on the clusters in such a way that it is possible to use different clusters and/or compilers and MPI libraries.

Loading the environement

Select between the GCC or Intel compiler:

module load gcc python openmpi py-mpi4py

or

module load intel python intel-oneapi-mpi py-mpi4py

Using MPI4Py

Launching MPI jobs

As with traditional MPI jobs you need to use srun to correctly launch the jobs:

mympicode.py

#!/usr/bin/env python3

from mpi4py import MPI
comm = MPI.COMM_WORLD
print("Hello! I m rank %d from %d running in total..." % (comm.rank, comm.size))

srun -N 2 -n 74 -q parallel python3 mympicode.py

• -N 2 is the number of nodes
• -n 74 is the number of tasks
• -q parallel is the QOS (see full documentation to see all the possibilities)

Example output:

Hello! I m rank 21 from 74 running in total...
Hello! I m rank 25 from 74 running in total...
Hello! I m rank 29 from 74 running in total...
Hello! I m rank 50 from 74 running in total...
Hello! I m rank 30 from 74 running in total...
Hello! I m rank 38 from 74 running in total...
Hello! I m rank 15 from 74 running in total...
Hello! I m rank 49 from 74 running in total...
Hello! I m rank 6 from 74 running in total...

Jobs

Failure to use srun will result in only one rank being launched.

Interactif

In the case of OpenMPI you can run sequential jobs in interactive

[user@jed ~]$ python
Python 3.10.4 (main, Nov 30 2022, 00:33:48) [GCC 11.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from mpi4py import MPI
>>> comm = MPI.COMM_WORLD
>>> print("Hello! I'm rank %d from %d running in total..." % (comm.rank, comm.size))
Hello! I'm rank 0 from 1 running in total...

In general you should run python with srun to for this to work

[user@jed ~]$ srun -N1 -n1 python