Optimal Adaptation in Autonomic Web Processes with Inter-Service Dependencies

We present methods for optimally adapting Web processes to exogenous events while preserving inter-service dependencies. For example, in a supply chain process, orders placed by the manufacturer may get delayed in arriving. In response to this event, the manufacturer has the choice of either waiting out the delay or changing the supplier. Additionally, there may be compatibility constraints between the different orders, thereby introducing the problem of coordination between them if the manufacturer chooses to change the suppliers. We present our methods within the framework of autonomic Web processes. This framework seeks to add properties of self-configuration, adaptation, and self-optimization to the traditional processes resulting in more dynamic and agile Web processes. We adopt the paradigm that an abstract Web process flow is pre-specified, and service managers are tasked with interacting with the actual Web services. We present two approaches for adapting the Web processes with dependencies. In our first approach, we take a global view of the process, and formulate a multi-agent Markov decision process (MDP) model for controlling the service managers’ actions. We show that this approach is globally optimal; however, it does not scale well to multiple service managers in the process. In our second, decentralized approach, each service manager performs its own decision making using a MDP model and coordinates with others through an external coordination mechanism. While this approach scales well to multiple managers since each manager need not model the others’ states, actions or costs, it’s not optimal. We provide a worst case bound for the loss in optimality for this approach. We empirically evaluate our methods using the supply chain problem, and report on their performance.