Sampling Exascale Data in Situ: Challenges and Opportunities
Ayan Biswas, scientist at Los Alamos National Laboratory
Tuesday, June 18, 2019
Brickyard (BYENG) 510, Tempe campus [map]
Supercomputers are becoming increasingly powerful, but their components have not scaled proportionately. Computing power is growing enormously and is enabling finely resolved simulations that can produce never-before-seen features. However, I/O capabilities lag by orders of magnitude which means that only a fraction of the simulation data can be stored for post hoc analysis. Pre-specified plans for saving features and quantities of interest will not work for features that haven’t been seen before. Data-driven intelligent sampling schemes are needed that can detect and save important parts of the simulation while it is running. Here, we discuss some of our recent methods which attempt to reduce the data in situ using intelligent sampling methods. We explore both the particle-based and grid-based datasets to apply our proposed methods and show the efficacy over existing generic methods.
About the speaker
Ayan Biswas is a scientist at Los Alamos National Laboratory (LANL). His background is in computer graphics and data visualization. He received his doctoral degree from The Ohio State University (advisor: Prof. Han-Wei Shen) and then joined LANL as a post-doctorate researcher (mentor: James Ahrens). Later, he was converted to a scientist in the data science at scale team. During his doctorate, his research focus was primarily concentrated on exploration and quantification of uncertainty in scientific visualization for multivariate datasets. In recent years, his research involves exploration of exascale simulations and creating intelligent algorithms for data analysis in such framework. Particularly, his interest lies in exploring novel techniques for in situ data reduction, uncertainty quantification and use of machine learning to minimize information loss on such very large-scale datasets.