A toolkit for advanced performance analysis

Abstract

Performance visualization is increasingly being considered a powerful approach for the analysis and debugging of parallel programs. Several performance analysis tools have already been developed for the analysis of program behavior. These tools have their specific advantages as well as limitations. A toolkit approach is presented that benefits from the advantages of conventional analysis tools, and also strives to overcome their limitations by giving a user the choice of a tool or analysis method that is most suitable in a particular situation. Various generic and commercially available data analysis tools contribute to the extensibility of the toolkit. One of the fundamental objectives of the toolkit is to practically implement multiple-domain analysis methods. Presently, the toolkit consists of Pablo, ParaGraph, AVS, and Matlab.<>