Colloquium Details
The TAU Parallel Performance Analysis Environment
| Author: | Allen D. Malony University of Oregon |
|---|---|
| Date: | December 04, 2003 |
| Time: | 15:30 |
| Location: | 220 Deschutes |
Abstract
The demands of performance problem solving in parallel and distributed systems, including those of scalability and online monitoring, continues to stress the limits of effective instrumentation, measurement, and analysis tools, ultimately resulting in the evolution of performance technology solutions. Moreover, the management of performance information, generated as a product of a comprehensive experimentation methodology, is an increasingly important component in performance characterization and modeling, as a fundamental basis for performance prediction and evaluation. The objectives of the TAU project during the last several years have been focused on developing a foundation for portable and robust parallel performance measurement. With this foundation established, we are turning our attention on problems of integrated performance analysis to address the new demands of performance assessment, prediction, and control. Our ultimate goal in this regard is to develop an environment to better support the processes of performance problem solving in all forms.
The talk will cover the components of the TAU parallel performance analysis environment that are currently under development. These include new tools for parallel profile analysis, parallel trace analysis, performance databases, online performance monitoring, and empirical performance modeling. We will also discuss future plans to include web-based performance analysis services, model-driven performance diagnosis, performance analysis and quality of service for component software.