Evaluating multiagent systems: a record/replay approach

Abstract

The adoption of multiagent systems by companies and users requires that it is possible to verify the properties of the multiagent systems. The main candidate of this verification is model checking but as sketched in the paper, it presents drawbacks such as combinatorial explosion or it does not take the environment into account. Model checking seems to be restricted to small systems with few agents. We present a different approach used in testing called the record/replay mechanism. We add as well a post-mortem analysis. This work describes the advantages of this approach and how we have implemented it in the MadKit platform.