Improving Latency Analysis for Flexible Window-Based GCL Scheduling in TSN Networks by Integration of Consecutive Nodes Offsets

Abstract

Time-Sensitive Networking (TSN) is an upcoming set of Ethernet standards designed for real-time and safety-critical Internet of Things (IoT) applications in automotive, aerospace and industrial automation domains. With the combination, complexity and flexibility of flow control mechanisms in TSN connected systems, the performance analysis for mixed-critical messages is becoming a difficult challenge. The flexible window-based Gate Control List (GCL) scheduling model has been proposed as a relaxation to assumptions on frames-to-window allocation, mutually exclusive gates opening, and scheduled end systems and switches, which offers more flexibility in the configuration of GCLs. In this paper, we are interested in providing a reliable verification method based on the network calculus theory to drive GCL configurations for TSN networks. To the best of our knowledge, this is the first performance analysis method suitable for the general flexible window-based GCLs in entire TSN networks, by reflecting the relative positional relationships of windows for same priority queues on consecutive nodes and constructing the window limitations into the shaper curve, in order to reduce the pessimism of the latency bounds. We validate the proposed method through Industrial IoT synthetics test cases and two large realistic cases, showing the significant reduction in pessimism on delay bounds, and the correctness and scalability by comparing with results from the previous work and simulation results.

Critical Offset Optimizations for Overlapping-Based Time-Triggered Windows in Time-Sensitive Network

“…Moreover, the algorithms were implemented with node-based design without considering the offset difference between the same priority windows in the adjoining nodes [5], [6], [34], [35]. Zhao et al [7] proposed an analytical model based on completely isolated TT windows by considering the offset difference between the same class windows in adjoining nodes. Also, after achieving a critical latency improvement compared to [7], Shalghum et al [8] optimized the overlap between TT windows based on three aspects; the priority of the overlapped queues, the position of the overlap, and the overlapping ratio.…”

“…Zhao et al [7] proposed an analytical model based on completely isolated TT windows by considering the offset difference between the same class windows in adjoining nodes. Also, after achieving a critical latency improvement compared to [7], Shalghum et al [8] optimized the overlap between TT windows based on three aspects; the priority of the overlapped queues, the position of the overlap, and the overlapping ratio. Although the proposed models in [7] and [8] have reduced the worst-case TT latency compared to the models in [5], [6], the window offsets were statically selected without critical optimization with related network metrics and latency behavior.…”

“…Also, after achieving a critical latency improvement compared to [7], Shalghum et al [8] optimized the overlap between TT windows based on three aspects; the priority of the overlapped queues, the position of the overlap, and the overlapping ratio. Although the proposed models in [7] and [8] have reduced the worst-case TT latency compared to the models in [5], [6], the window offsets were statically selected without critical optimization with related network metrics and latency behavior. The algorithms in [7] and [8] also did not confirm that the selected window offsets meet the lowest worst-case end-to-end latency bound.…”

“…Although the proposed models in [7] and [8] have reduced the worst-case TT latency compared to the models in [5], [6], the window offsets were statically selected without critical optimization with related network metrics and latency behavior. The algorithms in [7] and [8] also did not confirm that the selected window offsets meet the lowest worst-case end-to-end latency bound. Thus, a critical optimization is required to evaluate the whole possible range of offset difference between the same priority windows on the adjacent nodes.…”

“…Under fixed priority traffic assumption, window offsets will undoubtedly affect the end-to-end delay for the associated queues. Although the proposed algorithms in [7], [8] have proved the impact of window offsets on the worst-case latency bounds, no one has optimized the offset difference between the same priority windows in the adjoining nodes. Furthermore, the GCL schedules are offline based implemented, thus it is fundamental to evaluate and optimize the offset difference between the same priority transmission windows.…”