Article ID: | iaor20083322 |
Country: | Belarus |
Volume: | 3 |
Start Page Number: | 113 |
End Page Number: | 118 |
Publication Date: | Sep 2007 |
Journal: | Mathematical Sciences |
Authors: | Bakhanovich S.V., Sobolevsky P.I. |
A problem of tiling optimization for mapping of algorithms onto distributed memory parallel architectures is investigated. An approach of tiling optimization in LSGP mapping strategy is developed. The approach takes into account technical characteristics of computational environment: quantity of processors, processor performance, latency and bandwidth of communication channels. The method is intended for algorithms with uniform dependences.