Affiliation:
1. Universität Karlsruhe (T.H.), Karlsruhe, Germany
Abstract
A parallelized finite-volume algorithm for the prediction of two- and three-dimensional turbulent flow problems is presented. For the parallelization, the domain decomposition method is applied. Three different procedures for coupling the subdomains and two iterative methods for solving the system of algebraic equations are outlined. As typical applications, the two-dimensional flow in a model combustor, the three-dimensional flow in an experimental combustor and a jet-mixing flow are predicted using the parallel algorithms on two different parallel computing systems (Parsytec GCel, SUN-workstation cluster). A workstation cluster is suitable for a small number of processors (1–16) only. In contrast to this, a transputer cluster of 448 processors can be used efficiently. The comparision of the CPU-times of the parallel and the vectorized version of the CFD code, running on a SNI S600/20 vector computer, shows, that current technology vector processors give the best performance. Since the performance of the individual parallel processor can be increased without difficulty, parallel processing has the potential to solve large problems which cannot be solved by current vector computers.
Publisher
American Society of Mechanical Engineers
Cited by
3 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献