Hydraulic High-performance Parallel Computing Platform To Build

In China, there are some top water resources and hydropower projects, such as the Three Gorges Project, the Xiaolangdi Multi-Purpose Dam Project and the south to north water transfer project. With the enlargement of the scale of projects and the deepening of study on complex mechanical problem, the lack of scientific computing capability becomes the choke point. Even though we simplify the numerical model of engineering problem, one runtime of corresponding serial program is still several hours or several days or even several dozen days. This situation seriously restricts the development of water science. Many scientists' great ideas are unable to be realized because they lack enough computing capability. So it is urgent to set up a high performance parallel computing platform to support researches on water science.This paper investigates some large-scale HPC platform in American, Europe and China, and set up a HPC platform based on Sun Fire 6800 server which is introduced by the "948" project(200414) of Ministry of Water Resources. To know the floating-point performance and parallel performance of the HPC platform based on Sun Fire 6800 server, this paper makes two kinds of test by using HPL and LLCBench. This paper also does some researches on developing MPI programs, and writes a parallel program for matrix multiplication by using Fortran95 and MPI. The result of the parallel program on 4 CPU shows that using parallel method to do matrix multiplication can reduce the runtime and improve computing efficiency greatly. Finally, for a program to simulating static and dynamic loading to concrete, the parallelization is considered from servel aspects in the paper. On Sun Fire 6800 with 4 processors, for a concrete simulation program with 71013 discrete points and 78800 finite elements, the time spent with the parallel program decreases from about 15 days to about 2 hours. The development of the parallel algorithm reduces the simulation greatly, which is very important in studies of concrete materials.Presently, high performance parallel computing is not used a lot in water science, and we hope this paper will offer some reference for other researchers.