This research project investigates how to extend the scalable model of dynamic multithreading to encompass three key functionalities --- multiprogramming, interprocess communication, and support for I/O --- without requiring painstaking manual planning, tuning, or configuring. The research focuses particularly on using provably good ``feedback-driven'' strategies to accomplish this end. The researchers are developing a multiprogrammed multithreaded system to embody these strategies and provide a research vehicle for discovering additional properties of interacting adaptive jobs. Among the techniques under study is ``history-based feedback,'' which offers a promising foundation for provably effective scheduling and resource allocation strategies. This research hopes to make multiprocessors more efficient and easier to use by the vast majority of computer users, not just by expert computer scientists. Open-source software developed in the course of the research is freely available to anyone on the World Wide Web. Course materials on multiprocessor scheduling are distributed via the MIT OpenCourseWare project.
