Initial Experiments in Visualizing Fine-Grained Execution of Parallel Software Through Cycle-Level Simulation
Strong, Rick; Mudigonda, Jayaram; Mogul, Jeffrey; Binkert, Nathan
Keyword(s): Visualization, Parallel programs, Simulation
Abstract: Programmers wishing to obtain the best possible performance from multi-threaded software on parallel hardware must often understand detailed architecture- level interactions. Visualizations based on execution traces are often useful in gaining such understanding. Previous visualization tools have used traces from actual executions. We have experimented instead with visualizations based on traces from cycle-level simulations; while this approach imposes significant performance penalties, it exposes thread behavior, architectural events, and inter-thread interactions in minute detail unavailable via other means. We describe experiments with a prototype of our tool on several simple applications and on kernel code, and sketch the design of a usable tool.
Additional Publication Information: Submitted to First USENIX Workshop on Hot Topics in Parallelism (HotPar), Berkeley, CA, March 30-31, 2009
External Posting Date: December 6, 2008 [Fulltext]. Approved for External Publication
Internal Posting Date: December 6, 2008 [Fulltext]