Block-Striped Partitioning and Neville Elimination

In this paper a message-passing parallel implementation of the solution of linear systems by means of the Neville elimination is described. This type of approach is especially suited to the case of totally positive linear systems, which appears in different application fields. Standard data partitioning techniques, such as block row and block column schemes, are considered on three different topologies: ring, mesh and hypercube. The theoretical performance of the proposed parallel algorithms, in terms of run time, speed-up and efficiency, is derived. Experimental results obtained on an IBM SP2 multicomputer confirm the high performance of the block row parallel algorithm.