Measuring and Predicting Computer Software Performance: Tools and Approaches
Measurement-based performance evaluation is essential in computer science and industry alike, yet despite its widespread adoption, there remain several unresolved problems in this area. This presentation addresses problems in two areas: (1) reliably measuring and analyzing software performance, and (2) predicting software performance. The first set of problems is addressed with a series of measurement and analysis tools ranging from a distributed, heterogeneous, performance-measurement infrastructure to robust statistical methods. The second set of problems is addressed with two tools, one to analyze and predict software performance as it evolves during development, and another to analyze and predict software performance on machines where it has not been executed.