Profile-Directed Optimization of Event-Based Programs
Department of Computer Science
University of Arizona
Tucson, AZ 85721, U.S.A.
Matti A. Hiltunen Richard D. Schlichting
AT&T Labs - Research
Florham Park
NJ 07932, U.S.A.
Event-based systems provide a simple way to create flexible, extensible, and customizable system architectures and give a ``user-driven'' feel to the system. However, the indirect coupling between the raising and handling of events introduces a number of overheads into the system. Such overheads can be surprisingly large, and are especially significant in small mobile systems. This paper describes a framework for profile-guided optimization of event-based systems. Experiments using our approach on two different event-based systems, Cactus and X-windows, indicates that it can achieve significant reductions in event handling overheads and lead to considerable improvements in overall system performance.