A Swarm Reinforcement Learning Method for dynamic reconfiguration with end-to-end constraints in composite web services

Abstract

Service composition is an efficient way to fulfill user requirements in service-oriented architecture by combining several web services to perform a specific task. As a result of the dynamic environment of web services, the emergence of many unexpected events can disrupt or affect the quality of services composing a web service and, hence violate the end-to-end constraints. Dynamically reconfiguring the composite web service is essential to dealing with such issues. However, recent reconfiguration methods failed to meet the end-to-end constraints due to the vast number of web services with the same functionality and different nonfunctional features (QoS). This paper proposes a Swarm Reinforcement Learning approach to replace multiple failed services and maintain the original end-to-end constraints. ACO (Ant Colony Optimization) is used to improve the exchange of information between agents based on the Pheromone-Q values, inspired by real ants’ behavior. Experiments are conducted on real data sets and compared with related work methods to prove the efficiency of the proposed approach in terms of constraint satisfaction.