In many senses, today's complex software systems arevery similar to a live organism. This motivates us tostudy software by investigating its behavior. Whiletraditional modular-based analysis identifies staticdata and control flow, the actual behavior isdetermined not just by individual program modules butmore so by their run-time interactions driven byprogram inputs and influenced by the executionenvironment. Program behavior analysis augmentstraditional compiler analysis by analyzing theprogram traces. It builds statistical models toidentify large-scale, aggregate patterns and exposeopportunities for program optimization. This bookdescribes three behavior models and their use inimproving program performance. The new models andtechniques extend the programmer's toolbox with newprimitives for measuring and modifying compositeprogram behavior, a task that has become necessaryfor the production of efficient and reliable softwareon modern computer systems. The behavior models areespecially useful for professionals working onprogram performance optimization. They are alsoenlightening for the programmers who care about theperformance of their software system.