Behavior Models for Computer Programs

In many senses, todays complex software systems are very similar to a live organism. This motivates us to study software by investigating its behavior. While traditional modular-based analysis identifies static data and control flow, the actual behavior is determined not just by individual program modules but more so by their run-time interactions driven by program inputs and influenced by the execution environment. Program behavior analysis augments traditional compiler analysis by analyzing the program traces. It builds statistical models to identify large-scale, aggregate patterns and expose opportunities for program optimization. This book describes three behavior models and their use in improving program performance. The new models and techniques extend the programmers toolbox with new primitives for measuring and modifying composite program behavior, a task that has become necessary for the production of efficient and reliable software on modern computer systems. The behavior models are especially useful for professionals working on program performance optimization. They are also enlightening for the programmers who care about the performance of their software system.