Automatic adaptation of SOA systems supported by machine learning

Abstract

Recent advances in the development of information systems have led to increased complexity and cost in terms of the required maintenance and management. On the other hand, systems built in accordance with modern architectural paradigms, such as Service Oriented Architecture (SOA), posses features enabling extensive adaptation, not present in traditional systems. Automatic adaptation mechanisms can be used to facilitate system management. The goal of this work is to show that automatic adaptation can be effectively implemented in SOA systems using machine learning algorithms. The presented concept relies on a combination of clustering and reinforcement learning algorithms. The paper discusses assumptions which are necessary to apply machine learning algorithms to automatic adaptation of SOA systems, and presents a machine learning-based management framework prototype. Possible benefits and disadvantages of the presented approach are discussed and the approach itself is validated with a representative case study.