Schedulability analysis for Controller Area Network (CAN) with FIFO queues priority queues and gateways

Abstract

Controller Area Network (CAN) • Section 4.6 has been added, providing formal proofs that the schedulability tests given in Sections 4.1, 4.2 and 4.3 are sufficient (Theorems 2 and 3) and selfsustainable (Theorems 4 and 5). This section also shows how more precise analysis can be achieved when the priorities of messages in a FIFO queue span those of messages in a priority queue or another FIFO queue, which is often the case in practice. Extended version• In Section 5.2, we have added a formal proof that transmission deadline monotonic priority ordering is optimal when all messages have the same maximum transmission time (Theorem 7).• In Section 7, we have extended the experimental evaluation to show how the performance degradation due to FIFO queues depends on the number of messages in each queue.• Sections 6.1 and 7.1 have been added, exploring the effects of implementing one or more FIFO queues in gateway nodes that are responsible for transferring messages from one network to another.

Analysis and optimization of message acceptance filter configurations for controller area network (CAN)

“…In industrial applications, message ID and thus priority assignment has often followed an ad-hoc approach, with message IDs allocated based on the ECU supplier and the type of signals contained in the message. This has led to priority assignments that leave automotive networks unschedulable at bus utilizations of more than about 30-35% [4], when more appropriate priority assignment would allow bus utilizations of around 80% [8] before any deadlines are missed.…”

“…Further work has explored the issues that can arise if the prioritybased arbitration mechanism is circumvented, for example by the use of non-abortable transmit bufers [20], or FIFO queues [8,9]. In practical applications, using an optimal priority assignment policy is not in itself enough, since the ordering generated could leave the system only just schedulable, and thus vulnerable to deadline misses in the event that there is an increase in errors on the bus.…”

A review of fixed priority and EDF scheduling for hard real-time uniprocessor systems

“…However, priority assignment can have a very large effect on schedulability as shown by Davis et al [30], hence it is important whenever possible to use an optimal priority assignment policy.…”

“…Precise schedulability analysis has been derived for Controller Area Networks [36] that makes certain assumptions in terms of the behaviour of the nodes connected to the bus (perfect priority queues); however, in practice these assumptions may not always hold. Recent work has sort to address particular features of the hardware such as nonabortable transmit requests [51], and the fact that FIFO queuing policies are often used [30], [31]. These extensions make the analysis more flexible and better suited to commercial use.…”

On priority assignment for controller area network when some message identifiers are fixed

“…There are however many ways in which the communications stack, device driver, or CAN controller hardware can be implemented that do not match these assumptions. Issues include: Non-abortable transmit buffers [19], [16]; delays in refilling a transmit buffer [15]; the use of FIFO queuing policies [11]; work-conserving but otherwise ill-defined queuing policies [10], for example those using multiple levels of queues, or internal CAN controller message arbitration based on transmit buffer number. In this paper, we assume that the middleware and device drivers have been carefully designed to meet the assumptions of the classical analysis [6], as has been done for example with the Volcano Target Package 2 .…”