Evaluation of a Scalable In-Situ Visualization System Approach in a Parallelized Computational Fluid Dynamics Application

Current parallel supercomputers provide sufficient performance to simulate unsteady three-dimensional fluid dynamics in high resolution. However, the visualization of the huge amounts of result data cannot be handled by traditional methods, where post-processing modules are usually coupled to the raw data source, either by files or by data flow. To avoid significant bottlenecks of the storage and communication resources, efficient techniques for data extraction and preprocessing at the source have been realized in the parallel, network-distributed chain of our Distributed Simulation and Virtual Reality Environment(DSVR). Here the 3D data extraction is implemented as a parallel library (libDVRP) and can be done in-situ during the numerical simulations, which avoids the storage of raw data for visualization at all.