Automated Analysis of Multithreaded Programs for Performance Modeling

The behavior of multithreaded programs is often difficult to understand and predict. Synchronization operations and limited computational resources combine to produce com-plex non-linear dependencies between a program’s configu-ration parameters and its performance. Performance mod-els are used to understand these dependencies. Such models are complex, and constructing them requires a solid under-standing of the program’s behavior. As a result, building models of complex applications manually is extremely time-consuming and error-prone. In this paper we demonstrate that such models can be built automatically. This paper presents our approach for automatically mod-eling multithreaded programs. Our framework uses a com-bination of static and dynamic analyses of a single repre-sentative run of a system to build a model that can then be explored under a variety of configurations. We show how the models are constructed and show they accurately predict the performance of various multithreaded programs, includ-ing complex industrial applications. Categories and Subject Descriptor