On the basis of graph partitioning, a method and heuristic algorithms of mapping the parallel programs onto distributed computer systems with hierarchical structure are proposed. The optimization is achieved due to mapping intensively interacting parallel processes onto processor cores connected by rapid network channels. All hierarchical levels of the communication network is taken into account by the proposed method. Some results of mapping the MPI-programs taken from the benchmark suites SPEC MPI and NAS Parallel Benchmarks onto a geographically-distributed multicluster computer system are discussed. The work was supported by Russian Foundation for Basic Research (projects 11-07-00105, 12-07-31016, 10-07-00157, and 12-07-31016), by Ministry of Education and Science of the Russian Federation in the framework of the Federal Program "Scientific and science-education personnel of innovative Russia" for 2009-2013 (project 2012-1.1-12-000-1005-018), and by Council for grants of the Russian Federation President for the support of leading scientific schools (project SS-2175.2012.9).

Keywords: parallel programs mapping, geographically-distributed computer systems, GRID-systems, parallel computing, MPI

• PDF (in Russian) (3.00MB)



• PDF. zip (in Russian) (2.83MB)