Parallel Personal Computer Applications of CFD++, a Unified-Grid, Unified Physics Framework

Abstract The CFD++ Computational Fluid Dynamics (CFD) software suite is based on a new unified-grid, unified- physics and unified-computing framework. Implementing this within a parallel-processing environment involves many interesting aspects that are explored in the paper. Contemporary Personal Computer environments include parallel systems that can include 2-CPU machines with the Pentium II CPU and up to 4-CPU machines that use the Pentium-Pro CPU. These can be part of low speed (10Mbit/sec) or higher speed (100Mbit/sec) ethermet networks. The LINUX operating system and the public domain MPI (Message Passing Interface) offer a very convenient operating system environment and parallel-processing library on such computers and computer networks. Domain decomposition tools such as METIS (University of Minnesota) also serve as a complementary tool in being able to take advantage of parallel computing environments. This paper demonstrates the power of this environment to compute interesting small and large-scale CFD problems using the unified-grid, unified-physics approach. Examples to be presented include compressible perfect gas and compressible reacting flows, compressible low speed and incompressible flows, etc. The grid topologies that will be employed in these examples will include structured and unstructured grids, single and multiblock meshes, and various multiblock topologies including patched-aligned, patched-nonaligned and overset meshes. We will also present practical information on the effectiveness of low speed and high-speed networks, as part of standalone networks and as part of larger networks.