Description:

Modern profiling tools leverage visualization to make complex program behavior and performance information readable. Whereas they provide support for representing single executions, solutions to compare and summarize multiple runs are still lacking. Such solutions can potentially reveal misleading uncertainties in performance measurement, help identify performance regressions, or support a practical analysis of algorithmic complexity. In this project group, we will first assess the state of the art, both studying available tools as well as latest results discussed in the academic community. Based on the outcomes, we will then design and develop our own prototype that advances the state of the art regarding the visual comparison of executions and the visualization of performance variability. Challenges will be the efficient profiling of multiple program executions, the scalable encoding of the required information in a visual representation, and the integration of the approach into an interactive, Web-based framework. 

Detailed group presentation: Tuesday, June 13, 10:15 am, SGW-009 

Figure: The Timeline View of the Chrome Developer Tools with expanded call stacks summarizes a single client-side execution of a Web page.