Experiments and performance evaluation of Event Driven Mashups

In this paper we present the results of a set of experiments aimed at measuring the performance of Event Driven Mashup orchestrators. We show four different orchestrator implementations, one based on BPEL, one based on a plain Web Service, one based on a JSLEE container and one based on a fully distributed scheme. We show that most of the orchestration activity is spent in Web Service processing, that the adoption of the BPEL platform brings the orchestrator throughput down to an unacceptable level, that the usage of the JSLEE container is not convenient if service component interaction is implemented over a Web Service infrastructure, and that the fully distributed solution exhibits lower latency, as expected, but same throughput, with respect to the plain Web Service implementation.