Evaluating the Impact of Design Constraints on Expected System Performance

Abstract

Collective adaptive systems are difficult to design due, in part, to the presence of uncertainty in their actions, their communications, and their environment. Statistical methods can be used to account for this uncertainty by modeling its sources as a set of stochastic processes. These systems share characteristics with multi-agent systems in which a critical challenge is to manage a large decision space that exponentially increases with the number of actors. Often, system designers must constrain the system to limit the scope of its decision space to a manageable degree. Unnecessarily constraining a system can have unintended consequences on the system's performance. Thus, it is important to have techniques that can incorporate sources of uncertainty to evaluate the expected change in performance of a system under prescribed constraints. In this paper, we explore the use of stochastic multi-player games to model the expected change in system performance given two applicable constraints that are posed as design questions. Our experiment uses a model problem investigated in the domain of smart cyber-physical systems that employs a stochastic timed automaton to evaluate which input values better satisfy the system goal. We augment the model problem with two sources of uncertainty to evaluate the effects of distinct design questions against achieving the same goal. Our results demonstrate the potential for using stochastic multi-player games to evaluate the expected benefit or harm of enforcing specified design constraints on a collective adaptive system.