Controlling Modelling Artifacts

Abstract

When analysing the performance of a complex system, we typically build abstract models that are small enough to analyse, but still capture the relevant details of the system. But it is difficult to know whether the model accurately describes the real system, or if its behaviour is due to modelling artifacts that were inadvertently introduced. In this paper, we propose a novel methodology to reason about modelling artifacts, given a detailed model and a high-level (more abstract) model of the same system. By a series of automated abstraction steps, we lift the detailed model to the same state space as the high-level model, so that they can be directly compared. There are two key ideas in our approach -- a temporal abstraction, where we only look at the state of the system at certain observable points in time, and a spatial abstraction, where we project onto a smaller state space that summarises the possible configurations of the system (for example, by counting the number of components in a certain state). We motivate our methodology with a case study of the LMAC protocol for wireless sensor networks. In particular, we investigate the accuracy of a recently proposed high-level model of LMAC, and identify some modelling artifacts in the model. Since we can apply our abstractions on-the-fly, while exploring the state space of the detailed model, we can analyse larger networks than are possible with existing techniques.