In Situ Summarization for Existing Petascale and Future Exascale HPC Infrastructures

We propose extending our existing, scalable in situ visualization and analysis infrastructure -- based on VTK, VTK-m, and SENSEI -- to include summarization and compression algorithms. In Phase I, we will investigate summarization and compression algorithms, comparing wavelet-based techniques cited in the call to statistics-on-features, sorting-based compression, and feature-based extracts that provide post-hoc flexibility with low storage requirements. One example of feature-based extracts are renderings of simulation data where each pixel stores -- instead of a color -- (1) a scalar field value, (2) a depth into the scene from the camera's focal point, and (3) the surface normal of geometry at that pixel. These values can be used post-hoc to allow flexible exploration: they can be composed with other renderings of different features from the same viewpoint, relighted to clarify surface shape, and recolored to capture both small-scale and large-scale variations in scalar values. Beyond an investigation into promising new techniques, the Phase I work will also include an evaluation of algorithm scalability using simple surrogate algorithms targeting 2 many-core architectures: Intel MIC and nVidia Tesla.