P24: A Deployment of HPC Algorithm into Pre/Post-Processing for Industrial CFD on K-Computer
Abstract: Pre- and post-processing is still a major problem in industrial computational fluid dynamics (CFD). With the rapid development of computers, physical solvers are getting faster, while pre- remains slow because it's mainly a serial process. A methodology using MPI+OpenMP hybrid parallelization has been proposed to eliminate the manual work required during pre-processing for correcting the surface imperfections of CAD data. Compared to the rapidly increasing amount of data in recent years, the speed-up of visualization is insufficient. We address this limitation of post- by adapting the in-situ visualization to parallelize the post-processing using libsim (Visit) library. The performance of pre-/post- processing is investigated in this work, and we show that the pre-processing time has been reduced from several days in the conventional framework to order of minutes. The post-processing time has been reduced seconds order per frame, and approximately 30% increase of computational time was observed in vehicle aerodynamics cases.
Award: Best Poster Finalist (BP): no
Two-page extended abstract: pdf