| Summary: | 碩士 === 國立成功大學 === 水利及海洋工程學系碩博士班 === 90 === ABSTRACT
Computation of unsteady flows or turbulent flow for many hydraulic engineering applications, particularly in providing information for real time forecasting, requires high-speed computing. Modern developments in parallel computing using PC clusters have made the high speed engineering computations possible.
In this thesis, we utilize two PCs and MPI to build up a parallel computing cluster. Domain decomposition is used to parallelize a turbulent flow computation code. We compare the parallel code with the sequential code in convergence、accuracy and parallel efficiency. When the computed values of the sequential and the parallel codes are compared, the results show that the overall average difference is less than 1%, but the parallel code have a computing efficiency of over 95%. It is also found that domain decomposition is very useful in turbulent codes that use the SIMPLER scheme, since the pressure is solved iteratively using the Poisson Equation like solver.
It is also found the parallel hardware construction can be an obstruction to engineers who are interested in using parallel computing. The experiences of constructing the hardware and software of present study are written as the appendix for those who need them as a reference.
|
|---|
