Probabilistic analysis of CAN with faults

As CAN (Controller Area Network) is increasingly used in safety-critical applications, there is a need for accurate predictions of failure probability. In this paper we provide a general probabilistic schedulability analysis technique which is applied specifically to CAN to determine the effect of random network faults on the response times of messages. The resultant probability distribution of response times can be used to provide probabilistic guarantees of real-time behaviour in the presence of faults. The analysis is designed to have as little pessimism as possible but never be optimistic. Through simulations, this is shown to be the case. It is easy to apply and can provide useful evidence for justification of an event-triggered bus in a critical system.