On hierarchical parallel environment with multicore processors, mapping of subdomains to CPU/cores were optimized considering both the communication speed of different communication paths and the communication pattern of a parallel application based on the domain decomposition method. We evaluated proposed method on massively paralleled Intel Xeon PC cluster and confirmed that it could reduce communication time and achieve higher parallel performance than without mapping in several benchmark tests.
!!!All Science Journal Classification (ASJC) codes
- コンピュータ サイエンス（全般）