A Parallel Software for a Saltwater Intrusion Problem

In this paper we present a parallel implementation of a 2D numerical model for the solution of a transient density driven flow in porous media. The physical processes involved are multiscale, therefore computation time are usually long, thus a special effort has been made to speed-up computations by using parallel architectures. As most of the CPU time is spent in solving large sparse linear systems, it is important to choose an efficient linear solver which can deal with symmetric and non-symmetric sparse matrices. Accordingly, the parallel direct linear solver MUMPS is used for solving both transport and flow. However, scalability is not completely achieved. Therefore, parallelism is generalised to all computations. Matrices and data are distributed thanks to the partitioning of the METIS package instead of centralising on one processor. Actually, each processor treats its own spatial sub-domain and transfers data to the other ones when necessary. The resulting software is tested on a standard benchmark : the Henry test case. We use a homogeneous parallel cluster of PCs interconnected by a fast network. Investigation on the network and on the MUMPS options are carried out in order to obtain good performances.