Improving the Performance of Parallel Factorised Sparse Approximate Inverse Preconditioners
Martyn R. Field
Hitachi Dublin Laboratory,
O'Reilly Institute,
Trinity College,
Dublin 2, Ireland

We have shown previously that the Factorised Sparse Approximate Inverse (FSAI) preconditioner proposed by Kolotilina and Yeremin is an excellent parallel preconditioner. The preconditioner gives significant improvements in the performance of the iterative solver and the preconditioner has excellent parallel scalability. However the number of iterations to solve some problems is still large so there is room for improvement. Therefore, in this paper we investigate ways of improving the performance of this preconditioner by using various different sparsity patterns for the preconditioner and by using different renumbering strategies for the unknowns.