A system for predicting the run-time behavior of Web services

In a service oriented architecture requestors should be able to use the services that best fit their needs. In particular, for Web services it should be possible to fully exploit the advantages of dynamic binding. Up to now, no proposed solution allows the requesting agent to dynamically select the most "convenient" service at invoke time. The reason is that currently the requesting agents do not compare the runtime behavior of different services. In this paper, we propose a system that provides and exploits predictions about the behavior of Web services, expressed in terms of availability, reliability and completion time. We also describe a first prototype (eUDDIr) of the specification. EUDDIr relies on a scalable agents-based monitoring architecture that collects data on Web services runtime activity. The computed predictions are used by requesting agents to implement an effective dynamic service selection. Our proposal is well suited whenever requestors do not wish to explicitly deal with QoS aspects, or in the case that provider agents have no convenience in building up the infrastructure for guaranteed QoS, at the same time aiming to provide services of good quality to their customers. Furthermore, the adoption of eUDDIr effectively improves the service requestors "satisfaction" when they are involved in a Web services composition process. © 2005 IEEE