MPI

MPI is discussed in Slides from lectures starting with Lecture 18.

MPI stands for Message Passing Interface and is a standard approach for programming distributed memory machines such as clusters, supercomputers, or heterogeneous networks of computers. It can also be used on a single shared memory computer, although it is often more cumbersome to program in MPI than in OpenMP.

MPI implementations

A number of different implementations are available (open source and vendor supplied for specific machines). See this list, for example.

MPI on the class VM

The VM has open-mpi installed.

The libraries necessary to compile MPI programs are already installed on the class VM.

You should be able to do the following:

$ cd $CLASSHG/codes/mpi
$ mpif90 test1.f90
$ mpiexec -n 4 a.out

and see output like:

Hello from Process number           1  of            4  processes
Hello from Process number           3  of            4  processes
Hello from Process number           0  of            4  processes
Hello from Process number           2  of            4  processes

For hints on installation on a Mac, see Installing MPI on a Mac.

Sample codes

There are a few sample codes in the $CLASSHG/codes/mpi directory.

See Jacobi iteration using MPI.

See also the samples in the list below.

Further reading

• Argonne tutorials
• Tutorial slides by Bill Gropp
• Livermore tutorials
• Open MPI
• The MPI Standard
• Some sample codes
• LAM MPI tutorials
• Google “MPI tutorial” to find more.