Reducing CAN latencies by use of weak synchronization between stations

Abstract

Scheduling frames with offsets has been shown in the literature to be very beneficial for reducing response times in real-time networks because it allows the workload to be better spread over time and thus to reduce peaks of load. Maintaining a global synchronization amongst the stations induces substantial overhead and complexity in networks not providing a global time service such as CAN. Indeed, on CAN, a global clock is rarely implemented in practice and each station possesses its own local clock. Without a global clock, the de-synchronization between the streams of frames created by offsets remains local to each station and thus less efficient. In a previous paper [1], we developed a method to compute latency upper bounds for set of messages with offsets when the inter-node synchronization is not perfect. On a simplified test case, we obtained a reduction of 65% of the delay using a clock accuracy of only 1ms. In this article, we extend the method to consider a realistic case study (mixing periodic and asynchronous flows, considering errors and tacking into account the synchronization protocol).