The intrinsic complexity of self-organising MASs (multi-agent systems) suggests the use of formal methods at early stages of the design process in order to predict global system evolutions. In particular, we evaluate the use of simulations of high-level system models to analyse properties of a design, which can anticipate the detection of wrong design choices and the tuning of system parameters, so as to rapidly converge to given overall requirements and performance factors.

We take intrusion detection (ID) as a case, and devise an architecture inspired by principles from human immune systems. This is based on the TuCSoN infrastructure, which provides agents with an environment of artifacts—most notably coordination artifacts and agent coordination contexts. We then use stochastic π-calculus for specifying and running quantitative, large-scale simulations, which allow us to verify the basic applicability of our ID and obtain a preliminary set of its main working parameters.