Methoden des parallelen Postprocessing numerischer Strömungssimulationsdaten für die echtzeitfähige Visualisierung und Interaktion in VR-basierten Arbeitsumgebungen

Because of the steadily increasing performance of supercomputers, computational fluid dynamics (CFD) simulations are capable of producing constantly growing amounts of raw data. These data sets are essentially useless without subsequent post-processing. One particularly attractive evaluation approach is the interactive exploration within virtual environments. However, common visualization systems are not able to process large data sets while maintaining real-time interaction and visualization at the same time. Therefore, the obvious idea is to decouple flow feature extraction from visualization.The work presented mainly covers the functionality of the parallel CFD post-processing toolkit Viracocha. The distributed framework architecture relieves the visualization host by moving all time-consuming computation tasks to the parallelization backend. This makes it easier to guarantee real-time interaction within virtual environments. Additionally, the response time a user has to wait before requested post-processing results are visualized can be substantially reduced by optimized extraction algorithms adapted to parallel environments.Two further aspects are discussed in more detail. A considerable bottleneck in parallelization of CFD post-processing is the time needed to load large data sets. Therefore, a first extension of Viracocha aims at the reduction of the loading time, by implementing strategies mainly based on data caching and prefetching. The second aspect concerns an approach called data streaming, which minimizes the time a user has to wait for first results of a requested extraction. In order to achieve this, Viracocha sends back coarse intermediate data to the virtual environment before the final result is available. Some implemented streaming strategies also make use of multi-resolution data structures