Tom's picture
Tom's Home Page
I study distributed data analysis at scale by uncovering design patterns for parallel analysis problems and developing solutions for those patterns in the context of very large distributed HPC systems. This research is implemented in software infrastructures for rapidly parallelizing, scaling, and coupling in situ analysis algorithms.

More formally, Tom Peterka is a computer scientist at Argonne National Laboratory, fellow at the Computation Institute of the University of Chicago, adjunct assistant professor at the University of Illinois at Chicago, and fellow at the Northwestern Argonne Institute for Science and Engineering. His research interests are in large-scale parallelism for in situ analysis of scientific data. His work has led to three best paper awards and publications in ACM SIGGRAPH, IEEE VR, IEEE TVCG, and ACM/IEEE SC, among others. Peterka received his Ph.D. in computer science from the University of Illinois at Chicago, and he currently works actively in several DOE- and NSF-funded projects.

Data analysis and visualization can efficiently extract knowledge from scientific data. As computational science approaches exascale, however, managing the scale and complexity of the visualization process can be daunting. For example, the DOE/ASCR Workshop on Visual Analysis and Data Exploration at Extreme Scale (Salt Lake City, 2007) concluded that, "datasets being produced by experiments and simulations are rapidly outstripping our ability to explore and understand them," and the International Exascale Software Project draft road map (Dongarra et al. 2009) agreed that, "analysis and visualization will be limiting factors in gaining insight from exascale data."

There is a critical need to assist scientists with intelligent algorithms that save the most important data and extract the knowledge contained therein. To address these needs, our team is investigating the execution of analysis and visualization tasks in parallel directly on leadership machines, at very large scale. The topics below represent some of the research that our group tackles. The interaction with complex, multivariate, time-varying datasets drives our research, which is implemented in open-source software.

parallel analysis
design patterns

Block Parallelism

parallel volume rendering of a core-collapse supernova

Parallel Volume Rendering

trace of image compositing communication pattern

Parallel Image Compositing

parallel particle tracing of combustion

Parallel Particle Tracing

dynamic parallax autostereo display

Immersive Visualization

scalable data movement tools