Abstract: Hierarchical parallel programming model of multicore clusters has been a hot research in high performance computing. For hardware, the SMP cluster can be divided into two layers architecture between nodes and internal nodes. This paper describes the implementation of hierarchical parallel programming, according to the N-body problem algorithm, it conducts research on Hybrid parallel programming model. This paper proposes an optimization block synchronization of  MPI/OpenMP fine-grained N-body problem algorithm. Based on Dawning TC5000A cluster, execution time and speedup of this algorithm and traditional N-body parallel algorithm are compared, several summaries of the specific discussion are obtained.

Key words: hybrid programming model, block synchronization?algorithm, N-body problem, performance optimization

摘要： 多核集群的层次化并行编程模型一直是高性能计算的研究热点。以SMP集群为例，从硬件上可分为节点间和节点内的两层架构。阐述了层次化并行编程的实现技术，针对N体问题算法进行了基于Hybrid并行编程模型的并行化研究。提出了一种块同步MPI/OpenMP细粒度N体问题的优化算法。基于曙光TC5000A集群，将该算法与传统的N体并行算法进行了执行时间与加速比的比较，得出了几句总结性具体论述。

关键词: 混合编程模型, 块同步算法, N-体问题, 性能优化