Abstract
The maturity of the Computational Fluid Dynamics methods and the increasing computational power of today’s computers have allowed the automotive industry to incorporate the CFD technology in several stages of the design process. As the application of the CFD technology is moving from the component level analysis to the system level, the complexity and the size of the models increase continuously. Successful simulation requires synergy between CAD, grid generation, and solvers.
The requirement for shorter design cycle has put severe limitations on the turnaround time of the numerical simulations. The time required for a) mesh generation (around bodies of complex geometry, such as the geometry of a complete car), and b) obtaining numerical solutions (for flows with complex physics) has traditionally been the pacing item in CFD applications. Unstructured grid generation techniques and parallel algorithms have been instrumental in making such calculations affordable. Availability of these algorithms in commercial packages has proliferated in the last few years and parallel performance has become a very important factor in the selection of such methods for production work.
Although extensive research has been devoted in determining the optimum parallel paradigm, in practice the best parallel performance can be obtained only when the algorithms and paradigms take into consideration the architectural design of the target computer that they are intended for. The present paper addresses the issues related to the porting and optimization of a commercial code (Fluent/UNS) on a cache-coherent (cc) Non Uniform Memory Architecture (NUMA). Issues related to the Message Passing system and the memory to processor affinity are investigated using both a sample CFD code and Fluent/UNS. The scalability of the code when applied to the problem of the front-end cooling simulation of a development prototype family sedan are presented and discussed. Since speed and accuracy are the ultimate goals for using CAE in the design process a discussion concerning the model preparation time, grid generation process, and solution time will be presented. Comparison with available experimental data will be presented and discussed.
