作者: Steven Saunders , Lawrence Rauchwerger
关键词:
摘要: ARMI is a communication library that provides framework for expressing fine-grain parallelism and mapping it to particular machine using shared-memory message passing calls. The an advanced implementation of the RMI protocol handles low-level details such as scheduling incoming aggregating outgoing coarsen when necessary. These can be tuned different platforms allow user codes achieve highest performance possible without manual modification. used by STAPL, our generic parallel library, provide portable, transparent layer. We present basic design well mechanisms in current Pthreads/OpenMP, MPI implementations and/or combination thereof. Performance comparisons between explicit use Pthreads or are given on variety machines, including HP V2200, SGI Origin 3800, IBM Regatta-HPC RS6000 SP cluster.