High Performance Linux Clusters with OSCAR, Rocks, OpenMosix, and MPI

Chapter 13. Getting Started with MPI

This chapter takes you through the creation of a simple program that uses the MPI libraries. It begins with a few brief comments about using MPI. Next, it looks at a program that can be run on a single processor without MPI, i.e., a serial solution to the problem. This is followed by an explanation of how the program can be rewritten using MPI to create a parallel program that divides the task among the machines in a cluster. Finally, some simple ways the solution can be extended are examined. By the time you finish this chapter, you’ll know the basics of using MPI.

Three versions of the initial solution to this problem are included in this chapter. The first version, using C, is presented in detail. This is followed by briefer presentations showing how the code can be rewritten, first using FORTRAN, and then using C++. While the rest of this book sticks to C, these last two versions should give you the basic idea of what’s involved if you would rather use FORTRAN or C++. In general, it is very straightforward to switch between C and FORTRAN. It is a little more difficult to translate code into C++, particularly if you want to make heavy use of objects in your code. You can safely skip either or both the FORTRAN and C++ solutions if you won’t be using these languages.

MPI

The major difficulty in parallel programming is subdividing problems so that different parts can be executed simultaneously on different machines. MPI is a library of routines ...