The ever-increasing amounts of data generated by scientific simulations, coupled with system I/O constraints are fueling a need for in-situ analysis techniques, i.e., performing the analysis concurrently with the simulation. Of particular interest are approaches that produce reduced data representations while maintaining the ability to redefine, extract, and study features in a post-process to obtain scientific insights. One such approach is using topological constructs called segmented merge trees, which record changes in the topology of super-level sets of a scalar function. They encapsulate a wide range of threshold-based features, which can be extracted for analysis and visualization; however, current techniques for their computation are not scalable enough for in-situ analysis. This thesis presents a novel distributed algorithm that, for the first time, allows large scale in-situ computation of segmented merge trees. Existing merge tree computation techniques are restricted to simplicial complexes and 3-d rectilinear grids, instead, we present the theoretical foundations for computing merge trees on CW-complexes, which represent a broader class of meshes.

Based on this theoretical foundation, we present two variants of in-situ feature extraction techniques using segmented merge trees. The first approach is a fast, low communication cost technique that generates an exact solution but has limited scalability. The second is a scalable, local approximation that, nevertheless, is guaranteed to correctly extract all features up to a predefined size. We demonstrate both variants using some of the largest combustion simulations available on leadership class supercomputers. Our approach allows feature-based analysis to be performed in-situ at significantly higher frequency than currently possible and with negligible impact on the overall simulation runtime. We provide a detail performance and scalability analysis of this technique.

Furthermore, as scientific applications target exascale, challenges related to data and energy are becoming dominating concerns and it is crucial to understand the power impact of these in-situ analysis techniques. To this end, this thesis explores the various performance vs. power trade-offs of the presented in-situ technique and studies its behavior when various in-situ computation strategies are employed and extrapolates the power behavior to peta- scale systems to investigate different design choices through projections.