Resource Management in Federated Computing Environments (thesis)

Abstract:

This dissertation demonstrates that resource management mechanisms designed to support several degrees of autonomy allow the establishment of effective cooperation among autonomous computing sites. Using two distinct resources, storage and processing, this dissertation introduces two distinct notions for
autonomy: functional autonomy and service autonomy. For functional autonomy, it is shown that it is practical to implement a network file system that provides increased client independence from the file server while maintaining a useful level of functionality. For service autonomy, a load sharing mechanism has been implemented that provides sites' owners with ways to control how much performance degradation local jobs will perceive when remote jobs are serviced. This dissertation claims that, although some very large distributed systems will be formed by the growth of integrated distributed systems, most will be created by joining together a number of separate, already existing, and independent distributed systems. In an environment composed of a multitude of cooperating sub-systems the autonomy of each entity must be respected in order to convince the owners of each of the separate components to join the federation. Consequently, any viable distributed system architecture must support the notion of autonomy if it is to scale at all in the real world.