Enabling Exascale Through ParalleX Paradigm

Abstract

Continued progress in high performance computing for CFD and other performance critical application domains will require new methods of exploiting advanced technologies through the end of the next decade. By that time Exaflops performance is anticipated, but only if major challenges of scalability, efficiency, power, and programmability can be addressed. Historically dramatic changes have been accompanied with and enabled by new models of computation to provide the governing principles for the co-design and operation of the cooperating system layers from programming models, through system software, to processor architecture. This paper discusses the requirements and opportunities for such a new model in order to achieve Exascale computing and offers an example of an experimental model, ParalleX that is permitting the exploration of a potentially important new paradigm. While in its early stages, ParalleX shows promise as a conceptual framework for exploring these new transformative techniques, thus making possible the next generation CFD codes at Exascale performance.