Post-Processing Pipeline Optimization for Interactive Exploration of Multi-Block Turbine Propulsion Simulation Datasets

With the increasing power of current supercomputers, flow field simulations become larger and larger. Those datasets are too large to fit into the main memory of a visualization workstation and too large to be processed in a reasonable time on a stand-alone system. Distributed post-processing can eliminate dominant bottlenecks. With parallel computer systems, the post-processing can show a considerable speed-up. In this paper, we address issues to optimize several phases of the pipeline-based post-processing for interactive exploration of time-dependent, multi-block flow field simulation datasets. Firstly, we present the distributed post-processing framework, its approach to manage multi-block data structures, and how nested parallelization can improve the overall runtime. Scalability, balancing, and efficiency of algorithms optimized for unsteady multi-block datasets are presented. However, complex feature extractions are still time-consuming. For interactive exploration approaches in virtual environments, further strategies like streaming of intermediate data from post-processing backend to visualization frontend are needed. If the extracted visualization objects are too complex to be rendered with interactive frame rates, view-dependent multi-resolution techniques can help to present essential details without losing real-time rendering. To improve the interactivity, several integrated strategies are evaluated. Run-time measurements prove the efficiency of the approaches. An outlook for future steps concludes this paper