Book section Open Access
ZENG,Haibo; JOSHI, Prachi; THIELE, Daniel; DIEMER, Jonas; AXER, Philip; ERNST, Rolf
In addition to traditional buses such as CAN or FlexRay, packet-switched Ethernet will be used in next-generation automotive communication architectures. Ethernet’s superior bandwidth and flexibility makes it ideal to address the high communication demands of, for example, Advanced Driver Assistance Systems (ADASs), infotainment systems, and ECU flashing. As a switched network, Ethernet provides a scalable, high-speed, and cost-effective communication platform, which allows arbitrary topologies.
Ethernet evolved from a shared bus communication medium with CSMA/CDbased link access scheme to a switched network. Frame collisions in CSMA/CD were resolved by a binary exponential backoff algorithm which picked a random delay until a retransmission could be started after a collision. This deemed CSMA/CD unsuitable for real-time systems with tight latency or jitter requirements. Switched
Ethernet made CSMA/CD obsolete. In switched Ethernet, contention is moved into the switches, where a scheduler has full control over each output port. This enables the implementation of elaborate link schedulers, which allow the derivation of realtime guarantees. Today, Ethernet installations (including the automotive domain) are almost always switched. Hence, in the following, we will refer to switched Ethernet
as standard Ethernet.
In the automotive context, Ethernet is anticipated to serve as an in-vehicle communication backbone, here it must be able to transport traffic streams of mixedcriticality. This requires Quality of Service (QoS) mechanisms, in order to provide deterministic timing guarantees for critical traffic. Standard Ethernet (IEEE 802.1Q) introduced eight traffic classes. These classes can be used to prioritize traffic, which is typically implemented by a Static-Priority Non-Preemptive (SPNP) scheduler at each output port in each switch and end-point. This limited number of classes requires that multiple traffic streams share a class, making streams of equal priority indistinguishable to the scheduler. Traffic within a shared class is usually scheduled in First In First Out (FIFO) order.
Compared to CAN or FlexRay, Ethernet exhibits complex timing behavior, as each switch output port is a point of arbitration, which adds delay to the overall end-to-end latency. While mature formal performance analysis techniques have been established for CAN and FlexRay, such techniques are even more required for Ethernet before it can be used in timing- and safety-critical systems. This will become even more important in the context of highly automated and autonomous driving.
In this section, we use Compositional Performance Analysis (CPA) (see Chapter 23 and ) to derive worst-case performance bounds for Ethernet.