Fast networks have made it possible to aggregate distributed CPU, memory, storage, and data to provide the potential for application performance superior to that attainable on any single system. However, achieving performance on these metacomputing systems has proven difficult. Currently, application developers use customized application schedules to achieve performance on a metacomputer. Such schedules are based on a prediction of how the application will execute, given its resource requirements, the performance capabilities of the system resources, and the time-varying contention effects caused by competing applications executing on the metacomputer. This project's goal is to use the application-centric scheduling paradigm that is emerging from the practices of metacomputing applications developers as the basis for scheduling software which promotes application performance on metacomputing systems. To do this, the PIs will design and implement application-centric scheduling agents - called AppLeS (Application Level Scheduler) agents - for individual metacomputing applications. AppLeS agents will schedule the components of an individual parallel application on a non-dedicated system according to the dynamically changing performance each resource can deliver.
