Computational fluid dynamics on parallel processors

Greater computational power is needed for solving Computational Fluid Dynamics (CFD) problems of interest in engineering design. Parallel architecture computers offer the promise of providing orders of magnitude greater computational power. In this paper we quantify that promise by considering an explicit CFD method and analyze the potential parallelism for three different parallel computer architectures. The use of an explicit method gives us a best case analysis from the point of view of parallelism, and allows us to uncover potential problems in exploiting significant parallelism. The analysis is validated against experiments on three representative parallel computers. The results allow us to predict the performance of different parallel architectures. In particular, our results show that distributed memory parallel processors offer greater potential speedup. We discuss the import of our model for the development of parallel CFD algorithms and parallel computers. We also discuss our experiences in converting our model code to run on the three different parallel computers.