Marc Baboulin : October 15, 2003

Parallel distributed cholesky factorization for in-core large dense problems.

Abstract

We describe a parallel implementation of an in-core solver for large dense Hermitian linear systems on distributed memory computers. As the standard library ScaLAPACK does not provide packed storage routines for the Cholesky factorization and requires to allocate memory for the whole matrix, we have developed routines that fill this gap, using a left-looking algorithm.

We analyse and compare the performance with ScaLAPACK on recent platforms and we present results in the area of geodesy and electro-magnetics. This is joint work with Luc Giraud and Serge Gratton.