基于MPI的LAGFD-WAM海浪数值模式并行算法研究

Study on Parallel Algorithm for MPI-based LAGFD-WAM Numerical Wave Model

在LAGFD-WAM海浪串行数值模式基础上,利用MPI信息传递机制实现其并行化。通过对模拟区域合理划分,对数据采取分块加载,实现了各个节点的负载平衡;通过对算法的改进实现了粗粒度计算,大幅度减少了通信量,从而提高了程序的执行效率。对串行计算和并行计算的效率比较表明,本文建立的方法能够得到较高的加速比。对全球海浪模式,加速比和CPU数目大致呈线性关系。对高分辨率的区域海浪模式,在128 CPU条件下加速比可以达到91.9。

Based on the serial program of the LAGFD-WAM numerical wave model, the parallel computation of the numerical wave model was implemented by using message passing interface （MPI）. The computation loads at different computational nodes are balanced by making reasonable partition of simulated region and partitioned loading of data. The coarse granularity computation is made by improving the algorithm so as to greatly reduce the traffic and to raise the programme execution efficiency. It is shown from the comparison between the serial and parallel computation efficiencies that the parallel algorithm can achieve a higher acceleration rate. For a grobal wave model, the acceleration rate shows a largely linear relation with the number of CPU, and for a regional wave model of high resolution, the acceleration rate by using parallel algorithm and 128 CPUs can be as high as 91.9.