This paper describes formal probabilistic models of flooding and gossiping protocols, and explores the influence of different modeling choices and assumptions on the results of performance analysis. We use Prism, a model checker for probabilistic systems, for the formal analysis of protocols and small network topologies, and use in addition Monte-Carlo simulation, implemented in Matlab, to establish if the results and effects found during formal analysis extend to larger networks. This combination of approaches has several advantages. The formal model has well defined synchronisation primitives with clear semantics for modeling synchronous and asynchronous communication between nodes. Model checking of the probabilistic model determines exact probabilities and performance bounds, even if the model is non-deterministic; results that cannot be obtained by simulation. However, Monte-Carlo simulation can then be used in addition to study effects that only emerge in larger networks, such as phase transition.
展开▼
