CRII: III: Scalable and Interactive Dependency Visualization to Accelerate Parallel Program Analysis

University Of Arizona, Tucson AZ

Investigators

Abstract

Computer simulations allow scientists and engineers to explore and analyze processes that would be infeasible to observe experimentally such as the physics leading to fusion, possible future climate scenarios, and the aerodynamics of new vehicle designs. Computing larger, more accurate, and more realistic simulations requires large clusters of computing devices working in parallel. Using these parallel computing resources efficiently is difficult. By analyzing the behavior of the simulation programs, scientists and programmers can achieve significant efficiency improvements enabling better simulations and more useful results. This program analysis often involves visualizing the graph of dependencies between different parts of the program and different parallel computing resources. However, there may be hundreds of thousands of parallel devices and thousands of computer instructions involved. Current visualizations are difficult to interpret at this size. This project develops visualization approaches that overcome this limitation, giving programmers the support to analyze and optimize large-scale simulations and ultimately accomplish more science. At the same time, this project furthers the understanding of how to design scalable visualizations which may then be applied to other large and complex data. The objective of this research is the creation of visual encodings and interaction techniques supporting the exploration and analysis of computing dependencies at scale. This project will study how visual design factors affect the interpretability of visual depictions as well as how parallel computing experts interact with the data during exploration. As dependency graphs are generated programmatically, they often exhibit a high degree of regularity. This project will test the efficacy of different depictions of this regularity through controlled user studies. In determining the interactivity needs of program analysts, this project will conduct a design study resulting in an interactive visual approach for exploring program control and data flow. The expected results of this project are a set of validated design guidelines for representing and interacting with program dependency data and an interactive visual tool for analyzing program dependency graphs. All products resulting from this project will be linked from the project website (http://hdc-arizona.github.io/projects/dependencies).