Non-intrusive policy optimization for dependable and adaptive service-oriented systems

The Service-Oriented Architecture paradigm facilitates the creation of distributed, composite applications. Services are usually simple to integrate, but often encapsulate complex and dynamic business logic with multiple variations and configurations. The fact that these services typically execute in a dynamic, unpredictable environment additionally complicates manageability and calls for adaptable management strategies. Current system control strategies mostly rely on static approaches, such as predefined policies. In this paper we propose a novel technique to improve management policies for complex service-based systems during runtime. This allows systems to adapt to changing environments, to circumvent unforeseen events and errors, and to resolve incompatibilities of composed services. Our approach requires no knowledge about the internals of services or service platforms, but analyzes log output to realize adaptive policies in a non-intrusive and generic way. Experiments in our testbed show that the approach is highly effective in avoiding incompatibilities and reducing the impact of defects in service implementations.