Evaluating Message Transmission Times in Controller Area Network (CAN) without Buffer Preemption Revisited

Abstract

Controller area network (CAN) is a serial, broadcast bus, and is currently the de-facto standard for in-vehicle data transmission. CAN is the most widely used automotive communication network having the advantages of being low-cost and providing flexible and robust communication with bounded delay. CAN networks are mainly used for safety-critical real-time applications that have strict timing requirements in communication. This time-criticality makes the CAN schedulability analysis that has been studied over two decades crucial. Most of the previous studies on CAN schedulability assume an idealized communication model, with an unlimited number of transmit buffers in the communication controller. In this paper, we address the CAN schedulability analysis for the case of a single transmit buffer without buffer preemption when the buffering overhead can be ignored. We revisit the existing worst-case response-time analysis for this case that is based on the early, flawed CAN analysis, and we present a revised schedulability analysis. Moreover, we evaluate and compare the analyses by means of example message sets.

Eacan

“…Since the qth instance of the message is released at qT i , qT i is subtracted from the completion time of the qth instance to calculate the response time as shown in Equation (4). These WCRT analyses have been extended to address various practical issues, such as the limited size of TxObject [18], FIFO queuing in the device driver [10], non-abortable TxObject [20], and non-negligible time for copying a message into TxObject [19]. In particular, several studies [2,6] focused on the impact of transmission errors on the response time.…”