论文部分内容阅读
对于CFD等应用领域的问题 ,通常人们采用进一步细分网格 ,以获得更为精确的结果 ,但这样会增大问题规模 ,这就要求应用程序必须具有良好的可扩展性。本文提出了三种可扩展性度量方法 ,即时间可扩展性、效率可扩展性和收敛可扩展性 ,分别对并行应用程序在并行机上的执行时间、效率和收敛速率在扩展过程中的变化进行度量。使用这些方法对三个CFD应用实例进行了可扩展性分析。结果表明 ,这三种新方法克服了以往方法难以衡量由算法实现差别导致的并行系统可扩展性差异的不足 ,并具有易于使用的优点。
For applications in areas such as CFD, it is common for people to subdivide the mesh further to get more accurate results, but this increases the scale of the problem, which requires the application to have good scalability. In this paper, we propose three scalability measures: time scalability, efficiency scalability and convergence scalability. The execution time, efficiency and convergence rate of parallel applications in the parallel machine are respectively changed in the process of expansion measure. Use these methods for scalability analysis of three CFD application instances. The results show that the three new methods overcome the shortcomings of the previous methods, which are difficult to measure the scalability differences of parallel systems caused by the difference of algorithms, and have the advantages of easy to use.