Mesoscopic Stochastic Models for Validating Self-Organizing Multi-Agent Systems

The construction of self-organizing Multi-Agent Systems (MAS) is still short of systematic validation methods. These are crucial for acceptance of self-organization in mainstream software engineering. Validation of such systems requires the use of formal descriptions of the underlying coordinating process during the engineering process. The here proposed approach is based on stochastic validation methods applied to the microscopic states under an periodicity assumption. Thus, we propose in this paper Mesoscopic Stochastic Models as a novel methodological artefact for validating self-organizing multi-agent systems. To demonstrate the power of this method we apply it to a self-healing resource-flow system, in which a decentralized coordination process is used to restore the system's functionality after a failure. Information about the error is propagated through an overlay until the system is able to restore its original functionality.