Abstract
On-demand acquisition of computational power from remote providers is the basic motivation of Grid Computing and Service Oriented Computing in general. Both concepts propose to eliminate inefficiencies in enterprises' IT infrastructure, reduce costs and increase flexibility. However, participants waive control and risk non-availability at runtime. An important phase in the procurement process is the selection of a provider out of a candidate list. This article compares two approaches for service selection. In a centralized service selection concept, a resource broker selects and assigns partners, while a coordinator-free approach uses bilateral negotiations to build pairs. A network simulation analyzes allocation results using technical and economic metrics, and shows the decentralized approach's higher adaptiveness to changing network dynamics and service instance density.