Interactive parallel post-processing of simulation results on unstructured grids [article]

Florian Niebling, Universität Stuttgart, Universität Stuttgart
Numerical simulations and the assessment of their results are constantly gaining importance in product design and optimization workflows in many different fields of engineering. The availability of massively parallel manycore computing resources enables simulations to be executed with accuracies posing very high requirements on the methods for interactive post-processing of the simulation results. A traditional post-processing of such large-scale simulation datasets on single workstations is
more » ... en no longer possible due to the limited resources such as main memory, the low number of compute cores and the available network bandwidth to the simulation cluster. In this work, concepts and solutions are presented that enable interactive post-processing of large-scale datasets generated by fluid dynamic simulations on unstructured grids through the use of parallel manycore environments. A software architecture the parallel post-processing and visualization, as well as specific optimizations of frequently used methods for post-processing are introduced that enable the interactive use of parallel manycore resources. The implementation of the methods and algorithms is based on existing manycore devices in the form of programmable graphics hardware, which are no longer solemnly usable for computer graphics applications, but are getting increasingly interesting for general purpose computing. It will be shown, that methods for visualization of fluid simulation data such as the interactive computation of cut-surfaces or particle traces is made possible even for large-scale unstructured data. Additionally, an algorithm for the dense texture-based visualization of flow fields will be introduced that combines the presented methods for the extraction of cut-surfaces, isosurfaces and particle tracing. This algorithm for line integral convolution enables the interactive post-processing of flow fields on partitioned and distributed unstructured grids. The methods introduced in this thesis are evaluated using several large-scale simu [...]
doi:10.18419/opus-2262 fatcat:xgugfzclpfajhlu5rbsx2m7jvi