Exploring Visual Comparison of Multivariate Runtime Statistics

Abstract

To understand program behavior or find performance bottlenecks in their software, developers need tools to efficiently compare runtime statistics collected across multiple executions. As there is a variety of useful metrics, a good visualization needs to be able to handle multivariate data and highlight the most important differences between multiple versions. We identify three scenarios for the comparison of execution-relevant changes, and explore possible visualizations of the gathered multivariate runtime statistics.