The Parallel Implementing Of MASNUM Wave Model Based On MPI

With the development of ocean science, especially the numerical forecastingmodel, many ocean models such as POM、HYCOM and MASNUM need hugecomputation. Meanwhile, it places higher demand for computation speed andprecision. To meet the practical requirement of numerical forecasting models, parallelcomputation becomes one of the important ways to solve problems. On this basis,thebasic approach to solve the problems with parallel computation is summarized asfollows: Firstly,we analyze the influence of the parallel computation on numericalforecasting model, and ascertain the parallel part of the model. And then we presentseveral parallel optimization strategies, and formulate parallelism scheme. Finally, weorganize the parallel computing of the model, and simplify the procedures andenhance the speed, so that we can make it move faster and save the resources.MASNUM wave numerical model adopted in this paper aims to establish numericalforecast model of ocean wave and achieve its simulation and forecast by means ofstudying its propagation law. However, due to the expansion of sea areas wherecomputed and the enhancement of the spatial resolution, the original wave serialmodel can not meet the needs of rapid calculation. This article is intended to place aparallel study on MASNUM wave numerical model on the basis of MPI parallelprogramming environment.This paper refers to the basic theory of parallel processing technique, makingparallel computer, the design and implementation of the parallel algorithm andparallel programming understood in depth. And the parallel programming designtechnique based on MPI was learned to know more about its advantages, like its goodportability, its powerful function, its high efficiency, and so on, laying the groundworkfor realizing the parallel model. On this basis, there introduced MASNUM model andits serial program structure, and made statistics and analyses for each functionalruntime, thereby doing parallel analysis to this model, and ascertaining the parallel part of this model and its proportion in total running hours. The next thing wasbuilding parallel environment, installing and configurating MPICH while compilingsource code and running. Finally, it extracted the appropriate parallel algorithm,including numerical parallel algorithm and parallel algorithm based on Jacobiiteration, and formulated parallelism scheme. In the high performance computingplatformthen, combining MPI with Fortran, to achieve the parallelism and analysisexperimental results combining with speedup ratio and Amdahl law. Analysis resultsshowed that, with the increase of the amount of processor, compared to the originalserial model, parallel model is with higher running speed and superior parallelism, toa certain extent meeting the demand for its operational forecast.