1Parallelization of GSL on Clusters of Symmetric Multiprocessors∗

In this paper we discuss the application of an hybrid programming paradigm that combines message-passing (MPI) with shared memory programming (OpenMP). We apply this model to the parallel solution of two basic problems: the sparse matrix-vector product and the dynamic program-ming problem. We compare the results of the hybrid model with the application of a pure MPI model on a cluster of dual Intel Xeon processors. The experimental results show that the behavior of both models depend, among other factors, on the application and on the size of the problems. While with the dynamic programming problem we obtain very good speedups, in the case of the matrix-vector product the algorithms do not take very good profit of the dual processors. 1