Abstract

The rapid evolution of the Internet of Things (IoT) and the increasing prevalence of interconnected devices have significantly expanded the challenges in real-time systems, particularly concerning traffic planning for Time-Sensitive Networks (TSN). These advancements necessitate innovative solutions to ensure Quality of Service (QoS) under dynamic conditions. A critical step in addressing these challenges involves developing innovative scheduling approaches that must be thoroughly verified beyond theoretical models to ensure their practical applicability. Verifying these schedules requires creating and assessing a virtual testbed as it replicates the intricate behavior of real-time systems in a setting that offers control, flexibility, and scalability. Such a virtual environment is essential for fine-tuning traffic scheduling strategies, ensuring their effectiveness and compliance with stringent real-time constraints before implementation in actual operational contexts.

We utilize Docker containers and Linux networking functionalities to emulate real-world network scenarios to analyze traffic plans generated by TSN scheduling algorithms, particularly the Greedy Flow Heap (GFH) algorithm. We also develop a comprehensive software framework within this virtual environment that transforms theoretical scheduling concepts into practical, executable traffic flows, simplifying the intricacies associated with network experiment setups. In the setup, the Earliest TxTime First (ETF) queuing discipline (qdisc) is implemented, enabling the scheduler to replicate the timing precision necessary for real-time schedules. The delta parameter within the ETF qdisc is important as it determines the buffer time before a packet's scheduled transmission time. It acts as a 'fudge factor', allowing us to accommodate the inherent latencies of a virtual environment.

We identified four key performance indicators for our experiment: frame drops, frame transmission accuracy, processing delay, and end-to-end (E2E) latency. Our results reveal a notable increase in frame loss with reduced ETF qdisc delta values, and we found an optimal delta value of 5*10^7 ns. The empirical tests demonstrated a requisite initial stabilization period for the system, after which frame transmission accuracy achieved a minimum value of 20 µs. The research also highlighted the system's scalability, with larger network topologies showing decreased processing times due to efficient traffic distribution. The importance of network topology in influencing E2E latency is also evident, particularly concerning the number of switches a frame traverses. The observed deviation in frame transmission precision, ranging from 20 microseconds at best to 10 milliseconds at worst, and the processing delays at switches reaching 25 milliseconds in some instances, suggest that an emulation-based validation of the GFH algorithm has its limitations.