A linear solver, the QMR, for a very large linear system is implemented for network-based computing. Network- based computing with cluster of workstations or inexpensive personal computers recently has become an important and very successful technique. Network-based computing enables fast computation and resolves shortage of storage with inexpensive computers on the desks. Cluster of inexpensive computers offers them aggregated computing power and sufficient storage to challenge large-scale problems. Assume that the total storage available in computers is just sufficient for solving the problem. This is often the case in real-world applications since the linear system is too large. Thus, any preconditioner which requires more storage than that available is excluded. Classical Jacobi, Gauss-Seidel, and §OR and its variant, SSOR, is considered. Performance of preconditioners is analyzed.
[1]
Roland W. Freund,et al.
An Implementation of the QMR Method Based on Coupled Two-Term Recurrences
,
1994,
SIAM J. Sci. Comput..
[2]
Hyoung Joong Kim,et al.
Cost-effective parallel processing for remote sensing applications
,
1996,
IGARSS '96. 1996 International Geoscience and Remote Sensing Symposium.
[3]
Fabio Reale,et al.
Parallel computing on Unix workstation arrays
,
1994
.
[4]
Gene H. Golub,et al.
Scientific computing: an introduction with parallel computing
,
1993
.
[5]
Z. Andjelic,et al.
Parallel computation of electric fields in a heterogeneous workstation cluster
,
1995,
HPCN Europe.
[6]
S. Eisenstat.
Efficient Implementation of a Class of Preconditioned Conjugate Gradient Methods
,
1981
.