Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Parallel partitioned-inverse sparse matrix solutions

Conference ·
OSTI ID:54444
; ;  [1]
  1. Univ. of Wisconsin, Madison, WI (United States)

The partitioned inverse method has been demonstrated to be quite effective for parallel sparse matrix solutions on massively parallel machines. Though experiments on CM-2 have illustrated the advantage of using partitions, Intel iPSC/860 multiprocessor only favors fewer and denser partitions, particularly in the case of extremely sparse matrices. Different decomposition and communication algorithms are investigated here on the iPSC/860 to improve the performance. The decomposition is done in an interleave fashion in two different directions (row-wise and column-wise). The need for synchronization due to the interchange of intermediate solution vectors makes load balancing an important factor in obtaining an optimum performance.

OSTI ID:
54444
Report Number(s):
DOE/ER/25151--1-Vol.1; CONF-930331--Vol.1; CNN: Grant ECS-8907391; Grant ECS09215271
Country of Publication:
United States
Language:
English