Testing a Network by Inferring Representative State Machines from Network Traces

This paper describes an innovative approach to network testing based on automatically generating and analyzing state machine models of network behavior. The models are generated by the network test tool AGATE (Automatic Generator of Automata for TEsting), which is also described in this paper. The proposed test approach mimics experimental method, requiring repeated cycles of observing the network, modeling the network, making predictions about network behavior, and evaluating predictions. This paper focusses on the modeling step, in which the test tool AGATE automatically generates representative state machines from observed network traces. The generated state machines closely approximate the behavior of components of the network under test. Faults in the system may be immediately apparent from the state machines, but more importantly the state machines can be used for formal analysis. We propose this as a cost-effective alternative to manually defining a state machine before beginning tests.