Interaktive massiv parallele Visualisierung großer Datenmengen aus Strömungssimulationen

Nowadays the computer driven simulation of physical fluid processes has become an important topic. The fields of application are widely diversified, reaching from the area of product development up to safety analysis and optimization. Whereas the simulation in the past was able to benefit from the perpetual parallelization of the available hardware, the respective visualization of fluid datasets still has noticeable deficits. The result of this imbalance ends up in the situation that a fluid simulation is able to produce much more data per time interval that the corresponding visualization is able to handle. To accomplish this fundamental problem of the visualization of fluid datasets, this dissertation builds up several strategic solutions. In this context the principle of parallelization is transferred to the computation of the visualization. Therefore a concept is developed, which is able to massive-parallely process and display several concurrent visualization methods, no matter if they are of technical-scientific or realistic nature. The parameters of the herein used visualization methods are interactively steerable, their application on dedicated parts of the data space controllable. Derived from an abstract model of a visualization method and its parallel representation as an object in the system, an important point of the developed approach consists of the encapsulation of individual visualization methods in so called "probes": Virtual units, whose position, transparency and size can be freely controlled by the user. Individual modules of the system are decoupled through abstract interfaces. This enables an asynchronous and event driven computation of the visualization algorithms, respectively the geometries produced by them, like particles, volume-rendering or streamlines. Separated from the specialized data formats of the fluid simulation, a progressive data format optimized for the visualization is developed. Using it, a compression of the fluid data can be realized, together with an optimization of the data access for the visualization. Through the massively-parallel system and the progressive data format, the scalability of the visualization is achieved. This allows the system to run on different hardware platforms, from laptops up to multi-core CPU/GPU systems - always adapted to the available computational performance. Through that it is possible to display and investigate large fluid datasets even on relatively slow machines and to overcome the imbalance in the performance between fluid simulation and the corresponding visualization