Research On Application Of Hybrid Parallel Computing To Marine Water Quality Forecast System

With the space resolution enhancement and all aspects of data improvement, thecalculation of the marine water quality forecast system(hereafter referred to the marinewater quality forecast system as MWQFS) is becoming more and more larger. Therunning efficiency of serial programmed short-term forecasting system is very low. Forlonger-term forecast, the execution time will be much longer. This not only causes awaste of time, even more seriously, may also affect the water quality forecast. Runningthe pure MPI(Message Passing Interface) parallel system on the multi-core cluster willhave a big memory overhead, poor scalability and inadequate utilization of the hardwareof the cluster.Apply hybrid parallel computing technology to Jiaozhou Bay MWQFS, can greatlyimprove the running speed and has a higher operation efficiency. And it has animportant significance for further application of hybrid parallel technology in numericalsimulation of marine.Message Passing Interface is an ideal parallel manner in distributed storageenvironment, while Open Multi-Processing parallel technology, which occupy adominant position in scientific computing, is suitable for shared storage computersystem. In this paper, based on the combination of MPI and OpenMP, a hybrid parallelprogramming model of MPI+OpenMP was established for multi-core cluster, andapplied it into the water quality module of Jiaozhou Bay MWQFS. The main task is asfollows:(1) Made a detailed analysis of the serial program of Jiaozhou Bay MWQFS, andalso its pure non-blocking communication MPI parallel version of water quality module.On this basis, pointed out the deficiency of pure MPI parallel model, then the improveddesign proposal——say MPI+OpenMP hybrid parallel model of the water qualitymodule of Jiaozhou Bay MWQFS was put forward.(2) On the basis of the presentation and the analysis of Jiaozhou Bay MWQFS,realized the parallelization of water quality module by using MPI+OpenMP hybridparallel technology. Compared with the model of single MPI parallel technology, thehybrid model is more suitable for multi-core cluster computer system.(3) Compared the execution result of the hybrid paralleled forecast system with the serial programmed system. It verified that the hybrid parallel model is reliable. Then,the computing efficiency of hybrid model was tested. And compared it with the pureMPI non-blocking communication parallel model, the superiority of the hybrid parallelmodel was proved.The test results show that, MPI+OpenMP hybrid parallel model can significantlyimprove the execution efficiency of the water quality model computation, and hashigher speed-up ratio and scalability compared to pure MPI non-block communicationparallel manner.