On simulations in mas development: deriving stochastic models from agent implementations to examine self-organizing dynamics

The decomposition of a software system into a set of autonomous entities (agents) is a promising modeling approach to today's distributed applications. Agent interplay in these sets of agents, so-called Multi-agent Systems (MAS), enables the rise of complex, i. e. self-organizing and emergent system dynamics. Engineers revise agent implementations and face the challenge to comprehend, validate and finally predict the resulting system-level behaviors. In this paper we discuss the usage of simulation models to assist engineers in examining MAS dynamics. Particularly, we show how stochastic transition systems can be inferred from agent implementations. The resulting models allow engineers to explore the stochastic processes that may underly agent designs, facilitating both system design and validation as well as system optimization. We present tool support to assist model derivation from MAS implementations, following the established BDI agent architecture and exemplify its usage in a case study.