Collaborative Research: Progress-Based Resource Management Using Control

Oregon Health & Science University, Portland OR

Investigators

Abstract

The next generation of computing will be dominated by applications which fall between the existing categories "real-time" and "general purpose". These "real-rate" applications have performance contraints, such as throughput, smoothness, and bounded latency that are looser than real-time but stricter than general purpose applications. This project will further the state-of-the-art in resource management for real-rate applications, aiming to provide more predictable, accurate, and dynamic resource allocation. In particular, it will develop mechanisms and policies that can manage resources accurately for variable rate, high performance applications such as video streaming, sensor networks, or routers. It will also develop modeling and analysis techniques based on control theory that support reasoning about the dynamic behavior of applications and computer systems. Activities included in the project include: evaluating control algorithms for suitability for different application classes, inventing implementation techniques for building low overhead highly accurate controllers, developing techniques to analyze complex systems composed of multiple adaptive applications and resource managers, and defining notions such as quality-of-service or correctness that incorporate dynamics.