DART-MPI: An MPI-based Implementation of a PGAS Runtime System

摘要

A Partitioned Global Address Space (PGAS) approach treats a distributedsystem as if the memory were shared on a global level. Given such a global viewon memory, the user may program applications very much like shared memorysystems. This greatly simplifies the tasks of developing parallel applications,because no explicit communication has to be specified in the program for dataexchange between different computing nodes. In this paper we present DART, aruntime environment, which implements the PGAS paradigm on large-scalehigh-performance computing clusters. A specific feature of our implementationis the use of one-sided communication of the Message Passing Interface (MPI)version 3 (i.e. MPI-3) as the underlying communication substrate. We evaluatedthe performance of the implementation with several low-level kernels in orderto determine overheads and limitations in comparison to the underlying MPI-3.