Design and Analysis of Adaptive Traffic Scheduling Mechanisms for Time-Sensitive Networks

PhD Supervisors: Emmanuel Grolleau, Angeliki Kritikakou, Aakash SONI

Research Laboratories:

• INRIA, Rennes
• LIAS/ENSMA, Poitiers
• LYRIDS-ECE, Paris

Location: Paris, France

Interview Dates: February 2025 - Mars 2026

Keywords: Time-Sensitive Networking, Time-Aware Shaper, Deterministic Ethernet, Adaptive Scheduling, Real-Time Systems, Network Calculus, Robustness, Machine Learning, Industrial Networking

1. General Context

Time-Sensitive Networking (TSN) extends the Ethernet standard to support deterministic communication, guaranteeing predictable and bounded delays in distributed systems where time is critical. This evolution of Ethernet is a cornerstone technology in fields such as automotive systems, avionics, and industrial automation, where data transmissions must meet stringent temporal constraints [1].

Among the mechanisms defined by the TSN standards, the Time-Aware Shaper (TAS) [2] plays a key role. TAS governs the precise timing of frame transmissions according to a Gate Control List (GCL), a cyclic schedule that opens and closes transmission gates in synchronization with a global clock. In this way, the network behaves as a carefully orchestrated system, ensuring that time-critical messages are transmitted in a collision-free and predictable manner.

While this approach achieves a high degree of determinism, it is built on a fundamental assumption: that all traffic behaves according to a worst-case scenario [3, 2]. Frame sizes, arrival patterns, and load conditions are considered at their theoretical extremes to guarantee safety margins. However, this assumption, though conservative, is rarely representative of reality. In practice, message sizes vary, traffic patterns fluctuate, and network load is often well below its maximum capacity [4].

2. Problem Statement

This mismatch between theoretical assumptions and real-world behavior introduces inefficiencies and new forms of unpredictability. When the actual traffic is lighter or more irregular than expected, the precomputed GCL no longer matches the network’s real-time needs. Idle windows may appear in the schedule, degrading bandwidth utilization; frames may arrive earlier or later than planned, affecting synchronization; and latency or jitter can increase in subtle, unintended ways.

Ironically, a decrease in network load, which intuitively should improve performance, can destabilize a schedule that was designed for maximal contention. As a result, systems optimized for worst-case scenarios can exhibit degraded performance under typical operating conditions.

This observation points to a fundamental open question in TSN research: how can we reconcile strict temporal determinism with adaptivity to real-world variability? The challenge is to make deterministic networks more resilient and responsive to traffic fluctuations, without compromising their timing guarantees.

3. Research Objectives

The central aim of this PhD thesis is to explore how adaptive and robust scheduling mechanisms can be developed to improve the performance and reliability of TSN systems. Specifically, the research will focus on extending the Time-Aware Shaper beyond static, worst-case configurations toward dynamic, traffic-aware scheduling strategies.

The thesis will investigate how deviations from worst-case assumptions, such as smaller frames, sporadic arrivals, or transient idle periods, affect system behavior, and how scheduling mechanisms can be designed to detect and exploit these deviations. The goal is to create a new class of self-adjusting mechanisms that remain temporally safe yet dynamically optimized for the observed traffic.

Through this inquiry, the thesis seeks to move from a strictly worst-case approach to one that also incorporates average-case reasoning, capturing not only the limits of determinism but also the opportunities for adaptation.

4. Research Approach

The research will proceed through a combination of modeling, algorithmic development, and experimental validation.

• The first axis concerns the design of adaptive scheduling mechanisms. Several directions will be explored, including local adjustments of GCL, adaptive resizing of time slots, and distributed coordination strategies between network nodes. The main challenge lies in designing algorithms that are both computationally efficient and fully compatible with TSN standards, ensuring that any adaptation preserves the strict determinism required for real-time communication.
• The second axis will focus on theoretical modeling and analysis. Existing frameworks such as Network Calculus [5] and timed automata need to be extended to capture the behavior of TAS under variable or uncertain traffic conditions. New robustness metrics will be developed to measure the sensitivity of deterministic schedules to deviations from their design assumptions and to quantify the performance stability of adaptive mechanisms.
• Finally, the third axis involves experimental evaluation of the proposed methods. Simulation platforms such as OMNeT++ will be used to assess the impact of adaptive scheduling on latency, jitter, and bandwidth efficiency.

5. Expected Contributions

This research is expected to produce several significant outcomes. On the theoretical side, it will contribute to a deeper understanding of how variability affects deterministic communication, leading to new analytical models for robustness and adaptivity in TSN. On the practical side, it will result in novel algorithms for dynamic and context-aware scheduling that preserve timing guarantees while improving performance and resource utilization.

More broadly, the thesis aims to open a new research direction in self-adaptive deterministic networking, bridging the current divide between rigid, worst-case scheduling and flexible, real-world network behavior.

6. Required Skills

The desired profile should have knowledge in real-time scheduling and/or embedded networks, as well as be comfortable with programming tools. A good level of spoken and written English is required.

https://www.ece.fr/lecole-2/le-centre-de-recherche/

The desired profile should have knowledge in real-time scheduling and/or embedded networks, as well as be comfortable with programming tools. A good level of spoken and written English is required.