next up previous
Next: Empirical Results --- Up: Parallel Choleski Factorization of Previous: The Parallel Algorithm

Empirical Results

A stated goal of this block-diagonal-bordered Choleski solver is to simplify the task organization of the parallel Choleski algorithm and have interprocessor communications significantly reduced and regular. The performance of this block-diagonal-bordered Choleski solver is dependent on the ability to order the real power systems sparse matrices into the appropriate form with both uniformly distributed data in the diagonal blocks and a minimum number of equations on the lower border. In this section of the paper, we first examine the performance of the Choleski solver for generated test data that has perfect load balance, in order to understand the performance potential of the block-diagonal-bordered Choleski solver. We then report on the performance of the node-tearing nodal analysis and the performance of the block-diagonal-bordered sparse Choleski solver. In section 8.2, we illustrate the ordering capabilities of the node-tearing nodal analysis by presenting both pseudo-images of selected sparse load-flow matrices and data collected on the load imbalance as a function of the number of processors. But the real performance test of the node-tearing algorithm will occur when the performance of the block-diagonal-bordered sparse Choleski solver is examined for real power system load-flow matrices in section 8.3.





David P. Koester
Sun Oct 22 15:40:25 EDT 1995