Blocking for parallel sparse linear system solvers
- Additional Document Info
- View All
The parallel solution of sparse systems of linear equations is discussed. In such systems, parallelism and communication patterns are dependent on the nature of sparsity in the input matrix system. An algorithm, Block Solve, in which processors access blocks of rows from shared memory, is described. Experiments were carried out on the eight-processor Alliant FX/8 to determine the effectiveness of various blocking strategies in reducing execution times. An average block size of between four and eight minimized execution times. The Alliant FX/8 was used to emulate the execution of Block Solve on a shared-memory multiprocessor with private memories.
author list (cited authors)
Abraham, S. G., & Davis, T. A.