基于MPI的地下水数值模拟并行算法研究

Study on the Parallel Algorithm for Numerical Simulation of Groundwater Based on the MPI

地下水数值模拟是研究分析各种地下水问题的重要手段。针对传统地下水数值模拟串行算法在处理海量数据模型时计算时间长、运行效率低的特点,在分布式存储系统上利用消息传递标准MPI,设计了一种采用Cholesky预条件子,粗粒度、低通信开销的PCG并行算法。通过在具有4个处理核心的Linux集群环境中的测试可知,该程序运行所获得的加速比会随着处理核心数的增加而不断增大,并在4个处理核心数的时候获得最大加速比2.21,说明所设计的PCG并行算法具有较好的加速效果和可扩展性。

Numerical simulation of groundwater is an important method used to study and analyze various groundwater problems.Aiming at solving the problems,such as long processing time and low operating performance of the modeling,which would occur when using traditional serial algorithm to process the mass data for numerical simulation of groundwater,a coarse-grained parallel PCG algorithm with low communication overhead which adopted Cholesky preconditioning technique was designed using Message Passing Interface（MPI） in a distributed storage system in this study.According to the test on the Linux cluster environment with four processing cores,the speedup obtained from the programs would be increased as the increase of processing core.The maximum speedup 2.21,would be obtained when four processing cores were used,which indicates that the designed parallel PCG algorithm has a good acceleration effect and expandability.