PARALLELIZATION OF A MULTILEVEL DOMAIN DECOMPOSITION METHOD

This article deals with the implementation of the finite element method on MIMD (Multiple Instructions Multiple Data) multiprocessor machines with shared or distributed memory. This implementation is based on a multilevel domain decomposition method, using a direct solver for the interface problem. In the first part of this paper, we show the different steps of such an implementation on a shared memory machine, as well as the performance obtained on a CRAY Y-MP. In the second part, the main problems raised by implementation on distributed memory machines are considered. The Performance obtained on a cluster of IBM RS/6000 workstations, connected via an Ethernet network, using the PVM (Parallel Virtual Machine) software is also presented.