A wide variety of systems, including web farms, virtual machines, multi-tasking OSes, GRID computing systems, and sensor networks improve their accessibility, availability, resilience and fairness by ``sharing'' resources across the consumers they support. However, research that explores how to share resources generally derives point solutions, where different resource/consumer configurations require separately-designed sharing mechanisms. For instance, a scheduler often has implemented separately a single policy (e.g., FCFS, PS, Foreground Background, Shortest Remaining Processor Time), optimized for a particular load setting and cannot easily be switched to another policy when the situation changes.
This project is developing and analyzing Adaptive Sharing Mechanisms (ASMs) in which the mechanism used to share resources adapts dynamically to both the set of available resources and the current needs of the consumers, such that the system is truly autonomic. The study has been initiated with a modularization of the ASM into separate components, defined by the timescale of operation. A study of the various components using both cutting edge novel control theoretic and scheduling analyses is being performed. The design and analysis of ASMs will provide a theoretical grounding for fully autonomic systems, and the project will lead to efficient utilization of resources in diverse systems like server farms, GRID computing and Sensor Networks.
