Massive Parallel Computational model for Heterogeneous Exascale Computing System

In next half decade, a drastic change is anticipated in High Performance Computing (HPC) architectures to achieve emerging Exascale computing system which will be thousand-fold increase in current Petascale computing. This heterogeneous (CPUs + GPUs) framework based exascale computing system will be capable to perform one ExaFlops (10(18)) calculation per second. The pioneers in HPC have defined some limitations including energy (< 20 MW), budget (< 100 million $) and time (2020) that leads the number of challenges for targeted technology system. Nevertheless, an early initiative is important that have been taken by different development communities and vendors. Looking forward the tremendous challenges on the road of Exascale, massive parallelism is one of them which requires a new Parallel Programming (PP) model to provide the performance level required for exascale computing. In this paper we have proposed a new hybrid PP model that provides the coarse-grain and fine-grain parallelism at both intra-node and inter-node levels through heterogeneous architecture. The proposed model could be considered a leading model to fulfill performance demand for exascale computing system.