Porting generalized eigenvalue software on distributed memory machines using systolic model principles
This paper deals with the issue of porting sequential application on distributed memory machines. Three constraints direct the selection of the parallelization paradigm: little rewriting of the non-parallel source code, portability on several distributed memory machines and the overheads due to data and computation partitioning. The parallelization paradigm found to be most effective for our software on distributed memory machine was to provide the user with a client/server architecture.
KeywordsSystolic Array Performance Number Server Architecture Triangular Part Virtual Parallel Machine
Unable to display preview. Download preview PDF.
- [Cha88]F. Chatelin. Valeurs Propres de Matrices. Masson, 1988.Google Scholar
- [ML88]J.H Moreno and T. Lang. Graph-based partitioning for matrix algorithms for systolic arrays: Application to transitive closure. ICPP, 1:28–31, 1988.Google Scholar
- [Qui83]P. Quinton. The systematic design of systolic arrays. Technical Report 193, IRISA, 1983.Google Scholar