The International Conference for High Performance Computing, Networking, Storage and Analysis
Visualizing the Behavior of Large Programs.
Authors: Hoa Nguyen (University of Utah), Greg Bronevetsky (Lawrence Livermore National Laboratory)
Abstract: As High Performance Computing simulations grow more complex, debugging and optimizing them becomes vastly more difficult. This makes it necessary to design visualization tools that help developers understand the behavior of their applications. We propose novel visualizations and effective mechanisms to interact with them to support developer efforts to easily and efficiently analyze the behavior of large applications. These visualization techniques address some of the major issues still plaguing program behavior analysis, including reducing human error, improving speed and quality of decision making, and mitigating program size and complexity issues in analyzing behaviors. A case study on HPC programs demonstrates that our methods can help developers efficiently explore the behavior of their applications.