Adaptive Threshold Determination for Temporal Sampling during Smart In-Situ Visualization

In-Situ Visualization [1] has gradually been recognized as a feasible solution to minimize the ever-increasing I/O cost of large-scale simulations as the simulations move to the extreme scale. In this approach, visualization processing is performed while the simulation result is still in the main memory, thus drastically reducing the amount of data output to persistent storage for later offline visual analysis. This approach has matured for more than a decade, and there are currently several open-source tools and libraries, such as KVS, Catalyst, LibSim, Ascent, and SENSEI, as well as commercial solutions, such as Kombyne. Although they have different features and specific strengths, they have in common the support for image-based [2] or video-based [3] in-situ visualization methods to minimize the notorious lack of interactivity in the in-situ visualization. In these methods, a set of rendering images from different viewpoints (camera positions) are generated at every visualization timestep to enable interactive viewpoint changing during subsequent offline visual analysis.