Performance Study of PC-cluster Computation by the Fractional Time-step Method

• Main Authors: Kuang-Chuan Ou, 歐廣權
• Other Authors: Keh-Chin Chang
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/58183357557765780325

碩士 === 國立成功大學 === 航空太空工程學系碩博士班 === 94 === In order to overcome the large demand of computer memory and computation time required for the complex flow computations, a homemade PC cluster which is constructed by nine personal computers (PC), available in the electronic hardware stores, together with Gigabit Ethernet serves as the computational tool for flow simulation. This parallel computational platform is operated under the Linux operating system on which the MPI (Messages Passing Interface) libraries are installed. The test problem is a three dimensional (3D) lid-driven cavity flow and is numerically solved with the fractional time-step method. Two domain decomposition methods based on 1D and 2D topologies are investigated at various combination of (from one to eight) computers by comparing their parallel efficiencies. It shows that the parallel efficiency is decreased with the increasing number of PC. The lower the ratio of the data transfer quantity through the network to the computational load of a processor is, the higher the parallel efficiency is in the course of parallel computing. The criteria for determining the number of processors in a computational job is to assure high computational loading for every processor in the PC cluster. With the fractional time-step method employed in this study, the parallel efficiency can be improved by reducing the data transfer quantity through network transmission. For the presently investigated 3D problem, the use of the 2D topology of domain decomposition method leads to better parallel efficiency than the use of the 1D topology of domain decomposition method. One drawback of the homemade PC cluster is stemmed from the combination of different versions of processors. The overall computational efficiency of a PC cluster is primarily dominated by the processor possessing lowest computational efficiency. It is therefore suggested to construct a PC cluster by selecting the processors with same level of computational speed.