Service Composition Under Probabilistic Requirements

With the growth of the Internet-of-Things and online Web services, more services with more capabilities are available to us. The ability to generate new, more useful services from existing ones has been the focus of much research for over a decade. The goal is, given a specification of the behavior of the target service, to build a controller, known as an orchestrator, that uses existing services to satisfy the requirements of the target service. The model of services and requirements used in most work is that of a finite state machine. This implies that the specification can either be satisfied or not, with no middle ground. This is a major drawback, since often an exact solution cannot be obtained. In this paper, we suggest a probabilistic model for service composition: we annotate the target service with probabilities describing the likelihood of requesting each action in a state, and rewards for being able execute actions. We show how to solve the resulting problem by solving a certain Markov decision process (MDP) derived from the service and requirement specifications. The solution to this MDP induces an orchestrator that coincides with the exact solution if a composition exists, and otherwise, it provides an approximate solution that maximizes the expected sum of values of user requests that can be serviced.