Energy Efficient Runtime Framework for Exascale Systems

Abstract

Building an Exascale computer that solves scientific problems by three orders of magnitude faster as the current Petascale systems is harder than just making it huge. Towards the first Exascale computer, energy consumption has been emerged to a crucial factor. Every component will have to change to create an Exascale syestem, which capable of a million trillion of computing per second. To run efficiently on these huge systems and to take advantages of every computational power, software and underlying algorithms should be rewritten. While many computing intensive applications are designed to use Message Passing Interface (MPI) with two-sided communication semantics, a Partitioned Global Address Space (PGAS) is being designed, through providing an abstraction of the global address space, to treat a distributed system as if the memory were shared. The data locality and communication could be optimized through the one sided communication offered by PGAS. In this paper we present an energy aware runtime framework, which is PGAS based and offers MPI as a substrate communication layer.