Guaranteeing Hard Real-Time Requirements of In-Vehicle Multi-hop Communication Over Ethernet 2012-01-0193
Vehicle manufacturers currently use Ethernet for fast batch transfers when updating software in ECUs in a vehicle. But Ethernet is also planned for real-time traffic as well; in the short term future for streaming of video and audio and potentially in the long term also for Drive-by-wire functions.
Ethernet today uses the same frame format as the original Ethernet from the 1970s did but except from that it bears little resemblance to its original form. Ethernet today uses switches and hence the argument often raised against its use for carrying hard real-time traffic - that the shared medium can cause unbounded delay due to collisions - is not applicable today. In addition, normal Ethernet switches today support prioritization of traffic which allows an engineer to assign a high priority to urgent time-critical traffic so that its queuing delay is not affected by lower priority (supposedly less time-critical) traffic.
Although the high bit-rate and the ability to control queuing delays thanks to prioritization are attractive for carrying real-time traffic, this alone cannot guarantee that real-time requirements are fulfilled. It is necessary to prove with mathematical rigor that given a traffic model and given the topology and configuration of the network, all real-time requirements will be met. Such proof techniques are known for the CAN bus and they have been adopted in design tools. Such proofs are also known for star networks based on switched Ethernet for simple traffic models but no design tool is currently based on them.
Therefore, in this paper, in order to help designing this kind of analysis tool, we list desired functionalities of a tool for proving that timing requirements are fulfilled, and discuss important considerations for the analysis. It should support the ability to analyze multi-hop traffic over switched Ethernet networks and allow designers great freedom in how flows are described in order to reduce pessimism of the analysis due to modeling.