SBIR PHASE II: A Global Multi-Grid General Minimal Residual(GMRES) Scheme for an Adaptive Cartesian/Prism Flow Solver on Distributed Memory Machines

点击翻译按钮获取中文摘要

This Small Business Innovation Research Phase II project will develop a Parallel Computing System (PCS) for an adaptive Cartesian/prism generalized minimal residual (GMRES)/multi-grid flow solver. The system will include the following modules: Performance Monitor, Domain Decompositioner, GMRES/MultiGrid Flow Solver, Dynamic Load Balancer, and Parallel Flow Visualizer. The Performance Monitor can display the convergence of the flow simulation and loads on each processor and if necessary launch the Dynamic Load Balancer. The Domain Decompositioner and Parallel Flow Solver will be generalized and extended to three-dimensions in Phase II. For a cluster of workstations with constantly changing computational loads on each computer, a Dynamic Load Balancer will be developed to balance the loads on all computers in user defined time intervals. The company's commercial flow visualizer, CFD-VIEW, will be adapted for distributed memory machines to post-process the massive simulation data produced by the parallel flow solver. A Message Passing Interface (MPI) will be utilized to provide data communications between different modules and flow solvers. The PCS will run on both parallel computers and a cluster of workstations. The overall system will integrate grid generation/domain decomposition, flow simulation, load balancing and flow visualization into a single user-friendly parallel computing environment to achieve automation, accuracy and efficiency for flow simulation. Through this technology, significant cost reduction in CFD simulation will be possible. A successful Phase II should yield a product which will be very competitive in the current CFD software market because of its automation, efficiency and parallel computation capability.