D. Scientific visualization and parallel processing
Efficient scalability of the complete process chain - including simulation, data analysis, storage, and visualization - in massively parallel and distributed environments especially depends on limiting sequential bottlenecks caused by unbalanced latencies (e.g. interconnect communication, file I/O, data transport) and insufficiently parallelized parts of the processing pipeline, where complex optimization efforts are required. Interpretation of results of scientific computing has to be supported by explorative 3D visualization, where the previously mentioned aspects have to be taken into account, too. For this purpose, the innovative, distributed framework DSVR, consisting of in-situ parallel data extraction and interactive 3D presentation of time series, based on a streaming concept, was chosen by meteorology, climatology, and astrophysics disciplines to efficiently integrate visualization in their simulation applications, but has to be further developed. Efficient storage, retrieval, and analysis of simulation results or data extracts also poses new challenges related to large--scale scientific data management.