Parallel computing has become a key technology to efficiently tackle complex scientific and engineering problems. The ability of parallelism of an algorithm provides a useful rationale to recourse to it. Forgotten in years, the Jacobi algorithm has been identified with a high parallelism affinity, which is used today as a preconditioner for multigrid methods. This paper describes the message passing implementation of Karniadakis and Kirby as tested in a 9-node cluster with distributed memory. We outline details about tasks distribution and their mapping to the processes accompanied with a thorough description of communications running in a parallel environment.
iterative methods, systems of linear equations, C++ MPI, parallel programming