Hybrid MPI-OpenMP Paradigm on SMP Clusters: MPEG-2 Encoder and N-Body Simulation (1211.2292v1)

Abstract: Clusters of SMP nodes provide support for a wide diversity of parallel programming paradigms. Combining both shared memory and message passing parallelizations within the same application, the hybrid MPI-OpenMP paradigm is an emerging trend for parallel programming to fully exploit distributed shared-memory architecture. In this paper, we improve the performance of MPEG-2 encoder and n-body simulation by employing the hybrid MPI-OpenMP programming paradigm on SMP clusters. The hierarchical image data structure of the MPEG bit-stream is eminently suitable for the hybrid model to achieve multiple levels of parallelism: MPI for parallelism at the group of pictures level across SMP nodes and OpenMP for parallelism within pictures at the slice level within each SMP node. Similarly, the work load of the force calculation which accounts for upwards of 90% of the cycles in typical computations in the n-body simulation is shared among OpenMP threads after ORB domain decomposition among MPI processes. Besides, loop scheduling of OpenMP threads is adopted with appropriate chunk size to provide better load balance of work, leading to enhanced performance. With the n-body simulation, experimental results demonstrate that the hybrid MPI-OpenMP program outperforms the corresponding pure MPI program by average factors of 1.52 on a 4-way cluster and 1.21 on a 2-way cluster. Likewise, the hybrid model offers a performance improvement of 18% compared to the MPI model for the MPEG-2 encoder.