Three-Dimensional (3D) Large Fluid Flow Computations for U.S. Army Applications on KSR-1, CM-200, CM-5, and Cray C-90

The thrust of this research is to investigate the use of emerging massively parallel computer architectures for production type fluid flow applications. Typical flow computations in a ballistic simulation have reached a stage where a grid set involving a couple of million points are needed for adequate understanding of the flow behavior and computing the resulting forces acting on a projectile with a complex geometrical shape. For acceptable turnaround time, the simulation requires a system with large memory and high performance. Because of the hardware scalability and availability of large In-core memory, the emerging massively parallel processors (MPP) architectures such as Connection Machine's CM.5 and Kendall Square Research KSR-1 have become attractive for ballistic computations. Thus, this research is focused on implementation of complex real-world applications (not simple scaled down problems) on MPPs with the goal of sustainable performance. Because performance measurements obtained using the kernel of the full application and/or scaled down application size or 2-D cases could lead to considerably higher performance estimates, these factors are discussed in light of results from full 3.D applications obtained on MPPs. (AN) Prepared in collaboration with Computer Sciences Corp.