Flexible Provisioning of Service Workflows

Service-oriented computing is a promising paradigm for highly distributed and complex computer systems. In such systems, services are offered by provider agents over a computer network and automatically discovered and provisioned by consumer agents that need particular resources or behaviours for their workflows. However, in open systems where there are significant degrees of uncertainty and dynamism and where the agents are self-interested, the provisioning of these services needs to be performed in a more flexible way than has hitherto been considered. To this end, we devise a number of heuristics that vary provisioning according to the predicted performance of provider agents. We then empirically benchmark our algorithms and show that they lead to a 350% improvement in average utility, while successfully completing 5-6 times as many workflows as current approaches