In-Situ Visualisierung und Streaming von Plasmasimulationsdaten

Computer simulations are important scientific instruments these days. Visualizations help to understand and interpret these data. Since with the classic post processing approach too much data is produced per time step on recent peta scale and future exa scale systems this thesis introduces the in situ visualization library ISAAC, which is able to visualize distributed data right after creation without the need to store or transmit them. Using templates and template meta programming an abstract and reusable interface is described, which still enables simulation specific optimizations. The visualization over ray tracing works on the original data of the simulation, even if it runs on computation accelerators like Nvidia GPUs or Intel Xeon Phis. In these cases ISAAC is using the accelerators, too. Furthermore a way of sending and receiving meta data from and to simulations is described, especially to be able to steer them.

Besides this C++ library for visualization of simulations, a generic central server for creating video streams is motivated and described and also a simple HTML reference client for showing and steering simulations is implemented. For the evaluation of the solution, ISAAC is used to visualize the distributed gpu-based plasma simulation PIConGPU form the HZDR. Furthermore the speed of rendering dependent on different parameters is measured and discussed.