Performance Visualization and Analysis of Parallel Discrete Event Simulations with Projections

PPL Paper Number: 05-19
PPL CVS: PDESperfPADS06

Authors:
Chee Wai Lee, Terry L. Wilmarth and Laxmikant V. Kale
Parallel Programming Laboratory, Department of Computer Science, University of Illinois at Urbana-Champaign

PPL Technical Report 05-19, UIUC.


Abstract

In designing software for Parallel Discrete Event Simulation (PDES), detailed performance analysis has proved essential to address the challenges of minimizing overhead, maximizing parallelism and achieving scalability. This paper describes Projections, a performance visualization and analysis tool for parallel applications in Charm++, and its usage in the analysis of PDES applications. We discuss PDES analysis from three points of view: PDES tool design, PDES simulation design and PDES model analysis. We describe enhancements to Projections that make PDES-specific performance visualizations possible. We also develop generic analysis capabilities that can be used by other PDES engines. We demonstrate the utility of these features with Pose applications.


[html] [postscript] [PDF] [bibtex] [text reference]