MPI

Overview

  • MPI is an parallel programming API to execute processes over multiple cores / computers

  • Possible implementations are openmpi or mpich2

  • Remote processes can be e.g. executed via torque, slurm or ssh

mpirun --hostfile my_hostfile -np 4 my_parallel_application

Terminology

  • A communicator defines a group of processes

  • Each process in a group has a unique rank

  • An optional tag can be used to uniquely identify a message

Python source sample

  • pip install mpi4py

  • Run with mpiexec -n 4 ./test.py

#!/usr/bin/python

from mpi4py import MPI

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

print "Hello world from process ", rank, " of ", comm.Get_size(), " processes"

if rank == 1:
  data = [1, 5, 7, 13]
  comm.send(data, dest=0, tag=11)
  print rank, " sending data to process 0"

elif rank == 0:
  data = comm.recv(source=1, tag=11)
  print rank, " got ", data
  comm.bcast("thanks for all the fish!", root=0)