turkmath.org

Solution of sparse linear systems requires two main steps which are critical: (1) Reordering the sparse matrix and (2) direct or iterative algorithms for solving the reordered system of equations. In this talk we will talk about novel algorithms that improve the parallel scalability in both steps. In the first part of this talk, we will present a new multithreaded and recursive variation of the DS factorization based parallel direct sparse solver. We show the improvement compared to a sparse LU factorization based multithreaded direct solver on a shared memory architecture. In the second part, we will present recent results on obtaining the Fiedler vector and the permutation induced by the Fiedler vector effectively on a parallel computing platform. We will show a significant parallel improvement using our algorithm compared to a highly effective sequential counterpart.