next up previous
Next: Algorithm Performance on Up: Comparing Parallel Direct Previous: Parallel LU Solvers

Conclusions

Due to the guaranteed convergence of the Gauss-Seidel algorithm for positive definite or diagonally dominate matrices and the relative performance for the parallel Choleski solver, there is potential for significant speedup by selecting the parallel Gauss-Seidel method for solving those power systems network matrices that would normally require double precision Choleski factorization. For those applications that require LU factorization, more information concerning the iterative solver convergence would be required in addition to the rate of convergence before decisions could be made concerning the selection of direct versus iterative methods.



David P. Koester
Sun Oct 22 17:27:14 EDT 1995