Flexible window-based scheduling with critical worst case latency evaluations for real time traffic in time sensitive networks

Deterministic and low latency communications are increasingly becoming essential requirements for several safety-critical applications, such as automotive and automation industries. Time-sensitive networking (TSN) is a new Ethernet-based framework introduced to support these applications. TSN differentiates mixedcriticality traffic into three different categories: time-triggered (TT), Audio/Video Bridging (AVB), and best effort (BE). The TT flows are scheduled using a predefined gate control list (GCL) in each selected node targeting deterministic and low latency, extremely low jitter, and no congestion loss. The unscheduled traffic (AVB and BE) share the remainder bandwidth using the credit-based shaper (CBS), with a deterministic latency requirement for AVB but less than TT traffic and no QoS requirements for BE. Implementing a suitable predefined schedule in all selected nodes is a complex and vital problem. The main challenge is how to guarantee TT requirements without missing AVB deadlines. First, complete isolation between TT windows leads to wasting bandwidth and missing QoS requirements for AVB traffic. Moreover, nonoptimized window offsets will degrade the end-to-end latency performance for the associated TT queues, leading to less bandwidth availability for unscheduled transmissions. Also, implementing all GCLs in the selected path based on TT evaluations without considering their impacts on the AVB performance results in improper scheduling designs. Accordingly, three related phases are introduced in this thesis to cover these points as follows. The first part introduces a flexible window-overlapping scheduling (FWOS) algorithm that allows the TT windows to overlap in GCL implementations. An analytical model for the worst-case end-to-end delay ( ) is derived for TT traffic using the network calculus (NC) approach and evaluated using a vehicular use case, considering the overlapping among TT windows by three different metrics: the priority of overlapping, the position of overlapping, and the overlapping ratio ( ). For each given latency deadline, the FWOS algorithm determines the maximum allowable that obtains the highest unscheduled bandwidth without missing the TT latency deadlines. Even under a non-overlapping scenario, FWOS obtains less pessimistic latency bounds than the latest related works. The second part proposes an optimized flexible window-overlapping scheduling (OFWOS) algorithm that optimizes the offset difference ( ) between the samepriority TT windows in the adjacent nodes. Using -based GCL implementations, the bound for TT traffic is formulated using NC for a targeted priority queue and assessed with under non-overlapping and overlapping-based scenarios. OFWOS obtains more reductions than the previous related works, leading to more flexible overlapping between TT windows in each node. A new scheduling constraint is implemented to control the overlapping, targeting more relaxed GCL implementations with guaranteed TT latency deadlines. In the third part, the worst-case AVB latency under overlapping-based TT windows (AVB-OBTTW) algorithm is presented to examine the OFWOS effects on AVBlatency performance, where represents an AVB queue, i.e., { }. Separate analytical models are derived using NC to calculate for AVB- with preemption and non-preemption modes. Both models are evaluated under back-to-back and porosity configurations with light and heavy load conditions. Compared to the latest related works, AVB-OBTTW reduces for AVB- flows by different percentages depending on values. The lowest bound is obtained with the maximum allowable that meets TT latency deadlines using the OFWOS algorithm. Thus, combining OFWOS and AVB-OBTTW evaluations can be a helpful guide for TSN designers to implement tighter and more trusted GCL schedules.

Text FK 2022 72 IR.pdf Download (1MB)

Actions (login required)

View Item