Parallel Simulation Of3-D Shallow Water Flow With Free Surface

In recent years, more and more human activities such as resourcedevelopments and engineering constructions are carried out in estuariesand coasts, which bring a high speed development to the research in theregion. And the computation of environmental hydrodynamics is alwaysthe fundamental issue and research focus. In order to get a precisesimulation of the process of environmental hydrodynamics, plenty ofresearchers and institutes developed their own numerical simulationsoftwares to simulate the flows in estuaries and coasts. While as theresearch target of the hydrodynamics in the domain is going furtherly, ahigher requirement of the resolution is urgent. On the other hand, as thespatial scales of estuaries and coasts are immensely huge while therequirement of spatial resolution increased greatly, the calculation amountof numerical simulation grows explosively. At the same time, as computerhardwares develop rapidly nowadays and high performance computerswith different architectures like “fat node” and cluster are well developedand commonly used in recent years. To make full use of the hardwareresources, it is the only way to parallelize the traditonal serial codes.This paper compared the two mainly architectures of highperformance computer, namely, distributed memory multiprocessor andshared memory multiprocessor as well as the tools implementing parallelcomputations on these two different architectures which are MPI andOpenMP. Considering the hardware developments based on many-cores, we selected OpenMP as the tool to realize the parallelizing serialsimulation codes OCERM(Ocern Coast Estuary and River Model). Inorder to get a better parallel efficiency, a lot of work was done to optimizethe parallelized codes, including ignoring the unnecessarysynchrinizations and optimizing the order of loops to make full use of thehigh speed L2cache. Based on the parallelized codes, the open channelflow in a rectangle domain was used as validation case to research therelationship between parallel efficiency and factors such as calculationamount, parallel granularity and bond thread technique. Especially, theextensibility of parallelized OCERM which solves the linear equationswith Bi-CGSTAB method was discussed. Finally, the tide flow simulationin the Yangtze estuary and Hangzhou bay was carried out to verify thecorrectness of the parallelized simulation model.