Towards commercialization of utility-based resource allocation

Abstract

Previous experience with a data center prototype called Unity established that utility functions provide a natural framework for self-optimization in distributed autonomic computing systems [1]. In an effort to bring the promise of utility-based resource allocation to the marketplace, we have infused methods prototyped in Unity into two interacting commercial products: WebSphere Extended Deployment (a middleware application server environment) and Tivoli Intelligent Orchestrator Global Resource Manager (a component of an IBM provisioning management solution). We describe several challenges to commercialization that stemmed from the need to reconcile the fundamentally different types of objectives to which the two products managed, and detail how we addressed those challenges via modifications to existing internal computations and to the type of information exchanged between them. Furthermore, we describe an experiment that demonstrates quantitatively the commercial viability of utility-based resource allocation, and the flexible and responsive adjustment to changes in workload and objectives that it provides. © 2006 IEEE.