An Artificial Immune System Model for Task

– Networked computing resources can be harnessed to provide medium and large–scale computational services. Most systems to harness such resources rely on allocation heuristics and policies to perform the mapping of tasks onto hosts. Such heuristics are typically statically designed and typically only the parameters that govern the heuristic can be adapted over time. This paper describes a novel architecture for task mapping and performance optimization that adapts automatically to new architectures and new programming strategies and algorithms. It will not only perform allocation and performance optimization, but also will learn about new systems and bottlenecks and respond appropriately. An early implementation that uses performance monitoring and an artificial–immune–system–based intelligent system is described, and test results presented, as well as plans for a fully adaptive and intelligent allocation system. Point of Contact: Samuel H. Russ Phone: (601) 325–7775 Fax: (601) 325–7692 Mail: Engineering Research Center P.O.Box 9627 Mississippi State, MS 39762 email: [email protected] Web: http://www.erc.msstate.edu/~russ