Givens and Householder Reductions for Linear Least Squares on aCluster of Workstations

We report on the properties of implementations of fast-Givens rotation andHouseholder reflector based parallel algorithms for the solution of linearleast squares problems on a cluster of workstations. It is shown that theGivens rotations enable communication hiding and take greater advantage ofparallelism than Householder reflectors, provided the matrices are sufficientlylarge.