XpressSpace: a programming framework for coupling partitioned global address space simulation codes

摘要

Complex coupled multiphysics simulations are playing increasingly important roles in scientific and engineeringrnapplications such as fusion, combustion, and climate modeling. At the same time, extreme scales,rnincreased levels of concurrency, and the advent of multicores are making programming of high-end parallelrncomputing systems on which these simulations run challenging. Although partitioned global address spacern(PGAS) languages attempt to address the problem by providing a shared memory abstraction for parallelrnprocesses within a single program, the PGAS model does not easily support data coupling across multiplernheterogeneous programs, which is necessary for coupled multiphysics simulations. This paper exploresrnhow multiphysics-coupled simulations can be supported by the PGAS programming model. Specifically,rnin this paper, we present the design and implementation of the XpressSpace programming system, whichrnextends existing PGAS data sharing and data accessmodels with a semantically specialized shared data spacernabstraction to enable data coupling across multiple independent PGAS executables. XpressSpace supports arnglobal-view style programming interface that is consistent with the PGAS memory model, and provides anrnefficient runtime system that can dynamically capture the data decomposition of global-view data-structuresrnsuch as arrays, and enable fast exchange of these distributed data-structures between coupled applications.rnIn this paper, we also evaluate the performance and scalability of a prototype implementation of XpressSpacernby using different coupling patterns extracted from real world multiphysics simulation scenarios, onrnthe Jaguar Cray XT5 system at Oak Ridge National Laboratory.