MPI and OpenMP paradigms on cluster with multicores and its application on FFT

Abstract

At present the cluster of workstation with multicores is most popular in high performance architecture. However, pure MPI paradigm can't benefit from the computing capability of multicores. Parallel programming can combine distributed memory parallelization with shared memory parallelization. And hybrid MPI+OpenMP may be a superior solution because of its less communication, declined consumption of the memory, and improved load balance. In this paper, we propose an improved parallel FFT algorithm based on hybrid architecture and the results shows the algorithm has good scalability and high efficiency.