Enhancing BPEL processes with self-tuning behavior

The performance of BPEL processes depends on the composing Web services. Monitoring Web service performance and adapting to changes in service performance are essential for creating self-tuning BPEL processes. BPEL lacks constructs for performance monitoring and offers little support for runtime adaptability. In this paper, we present a monitoring and dynamic binding infrastructure for BPEL that transparently enhances existing BPEL processes with self-tuning behavior. The portable infrastructure relies on an automated transformation of BPEL processes and is compatible with any BPEL engine. Its flexible monitoring mechanism supports both complete and more efficient sampling-based monitoring. The infrastructure allows for pluggable service selection strategies. We present a probabilistic strategy, where the service selection probability depends on the previously monitored service performance. A thorough evaluation confirms that our infrastructure significantly increases BPEL throughput in settings where Web service performance fluctuates.