Worst-Case Latency Analysis for IEEE 802.1Qbv Time Sensitive Networks Using Network Calculus

Abstract

Distributed safety-critical applications in industrial automation, aerospace, and automotive, require worst-case end-to-end latency analysis for critical communication flows in order to prove their correct behavior in the temporal domain. With the advent of time sensitive networks (TSNs), distributed applications can be built on top of standard Ethernet technologies without sacrificing real-time characteristics. The time-based transmission selection and clock synchronization mechanism defined in the TSN enable the real-time transmission of frames based on a global schedule configured through so-called gate control lists (GCLs). This paper has an enhancement of allowing a mixture of the priority-based scheduling and time-triggered, which expand the solution space for the GCLs. Then, it is necessary to analyze the latency bounds for the critical traffic in the TSN network. In this paper, we start from the assumption that the GCLs, i.e., the communication schedules, and the traffic class (priority) assignment for critical flows are given for each output port and derive, using network calculus, an analysis of the worst-case delays that individual critical flows can experience along the hops from sender to receiver(s). Our method can be employed for the analysis of the TSNs where the GCLs have been created in advance, as well as for driving the GCL synthesis that explores a larger solution space than previous methods, which required a complete isolation of transmission events from different traffic classes. We validate our model and analysis by performing experiments on both synthetic and real-world use-cases, showing the scalability of our implementation as well as the impact of certain GCL properties (gate overlapping and traffic class assignments) on the worst-case latency of critical communication flows.