Performance analysis of a parallel finite element solution to the direct numerical simulation of fluid turbulence on Linux PC clusters

摘要

The parallel performance of a direct numerical simulation (DNS) computer code was tested on three different sets of Linux PC clusters including Intel® Pentium® (P) III 800 MHz, P III 1 GHz, and Xeon™ 2.8 GHz. The DNS code is used to simulate incompressible fluid turbulence and is developed by the Galerkin finite element method (FEM) with equal-order interpolating polynomials. It employs domain decomposition for data structure together with data parallelism for matrix assemblies and operations. This paper discusses the scalability of the code and contrasts the code performance on the Linux PC clusters.