宽带RAM模型在对称多处理器集群上的并行设计

Parallel computation of RAM model on SMP cluster

摘要: 为了提高声场模型的计算效率以满足当前水声研究对声场计算速度的要求,针对对称多处理器集群系统多节点并且节点内存在多个处理器的特点,利用共享存储模型OpenMP和消息传递编程模型MPI(Message Passing Inter-face)对声学计算模型RAM(Range-dependent Acoustic Model)进行并行编程,构建了并行计算平台,实现了RAM模型在对称多处理器集群系统上节点间和节点内两级并行,并通过实验对该平台的性能进行了测试.实验结果表明,RAM模型适用于并行计算,该并行计算方法具有很高的并行效率,可以大幅度提高声场计算速度.

Abstract: In order to accelerate the sound field predication,the Range-dependent Acoustic Model(RAM) is parallelized based on OpenMP and MPI in this paper.The implementation platform is built on a symmetrical multiprocessing cluster,where multiple nodes and CPUs are included.The parallel computation is done both between nodes and within nodes.The experimental results indicate that the RAM can work well in parallel,and the proposed method has high parallel efficiency and can improve the sound field predication greatly.