High-Level Reactive Support for Scalable Distributed Systems

To make large-scale distributed systems practical for a broader user base, we are investigating a high-level framework to support and aid the development of reactive distributed applications. The reactive object framework offers large-scale distributed applications the ability to automatically adapt to the state of the distributed system as well as the behavior of the application. The system automatically gathers, condenses, and provides access to performance statistics needed by the distributed application for adaptation. To hide adaptation from the programmer, the system provides default adaptation policies and mechanisms that can be combined to create reactive objects that automatically respond to changes in the system to improve performance. By separating policies from algorithms and data structures, the system encourages reuse of parallelized components. Ultimately, our goal is to create a library of common scientific reactive objects that can be directly incorporated into parallel applications.