Skip to main content

IEEE 802.1Qbv TAS: the time-aware shaper, gates on a timetable

· 8 min read
PhD candidate, LyRIDS (ECE) × CEDRIC (Cnam)

The Time-Aware Shaper (TAS) is the TSN tool that gives the strongest timing guarantees, and the hardest one to configure. The mechanism in the bridge is simple: a gate per queue and a timetable. The difficulty lies in computing the timetable.

The IEEE 802.1Q bridge: architecture and data-plane actionsIEEE 802.1Q bridge ↗Control planeSRP (Qat), TSN configuration (Qcc), path control (Qca)frame inframe outIngressPSFP (Qci)FRER (CB)Relayforwardingexplicit treesQueuingCBS (Qav)ATS (Qcr)Tx selectiongates (Qbv)CQF (Qch)MAC / PHYpreemption(Qbu, 802.3br)Common time: gPTP (802.1AS)
Figure 1. Where the Time-Aware Shaper acts: a gate in front of each queue at transmission selection, driven by the common time base of 802.1AS. Strict priority or the credit-based shaper still choose among the queues whose gates are open.Source: own drawing, after the bridge architecture, data-plane actions and QoS function chain in J. Farkas, TSN Basic Concepts, DetNet–TSN workshop, 11 November 2018, slides 7, 8 and 10.
Abbreviations used in this post (9)
AbbreviationStands forMeaning
CNCCentralized Network ConfigurationThe 802.1Qcc entity that computes paths, schedules and reservations for the whole network and configures the bridges.
GCLGate Control ListThe repeating timetable of gate states used by the Time-Aware Shaper (and by stream gates in PSFP).
IEEEInstitute of Electrical and Electronics EngineersThe professional association whose Standards Association publishes the IEEE 802 LAN/MAN standards.
PCPPriority Code PointThe 3-bit priority field of the 802.1Q tag (priorities 0 to 7).
PTPPrecision Time ProtocolIEEE 1588, a protocol that synchronises clocks over a packet network using timestamped message exchanges.
SDNSoftware-Defined NetworkingAn architecture that separates the control plane, run by a logically centralised controller, from the data plane of the network elements (RFC 7426).
TASTime-Aware ShaperIEEE 802.1Qbv scheduled traffic: a gate per queue, opened and closed by a repeating gate control list.
TSNTime-Sensitive NetworkingThe set of IEEE 802.1 standards that give Ethernet bounded latency, time synchronisation and high reliability.
VLANVirtual LANA logical LAN within a bridged network, identified by a 12-bit VLAN ID in the 802.1Q tag.

1. Definition​

Scheduled traffic, IEEE 802.1Qbv (now part of IEEE 802.1Q), adds a transmission gate to each queue of an egress port [1][3]:

  • a queue whose gate is closed is ignored by transmission selection;
  • a queue whose gate is open competes as usual, under strict priority or a shaper.

The gate states follow a gate control list (GCL). Each entry holds the open/closed state of every gate and a duration. The list repeats with a fixed cycle time, from a base time expressed in the synchronised time of 802.1AS. The result is a time-division schedule on each link, with windows reserved for given priorities [2].

The GCL is built centrally. In the configuration model of 802.1Qcc, the CNC computes a schedule for the whole network and installs a list on every bridge [4]. The SDN post describes that controller.

2. Slices, infringement and the guard band​

Two cycles of an 802.1Qbv schedule, each with a short time slice for VLAN priority 3 followed by a longer slice for all other priorities
Figure 2. An 802.1Qbv schedule. Each cycle starts with time slice 1, reserved for VLAN priority 3, followed by time slice 2 for all other priorities.Source: Figure 1- Example IEEE 802.1Qbv schedule.svg by Ulgorash, from the Wikipedia article “Time-Sensitive Networking”, Wikimedia Commons, CC BY-SA 4.0.

In Figure 2, frames of priority 3 never wait behind other traffic. During slice 1 their queue is the only one with an open gate. But the gate only decides when a frame may start. A frame that has already started is not stopped when its gate closes.

A best-effort frame started late in its slice runs into the next cycle and infringes the time-critical slice
Figure 3. Time slice infringement. A best-effort frame that starts late in slice 2 is still being sent when slice 1 of the next cycle begins, and delays the time-critical traffic.Source: IEEE 802.1Qbv - time slice infringement.svg by Ulgorash, from the Wikipedia article “Time-Sensitive Networking”, Wikimedia Commons, CC BY-SA 4.0.

The remedy is a guard band: before each protected slice, no new frame of the other queues may start. To be safe the guard band must be as long as the largest interfering frame. For a full-size Ethernet frame with preamble and inter-frame gap, that is about 12 µs at 1 Gbit/s and about 123 µs at 100 Mbit/s.

The same 802.1Qbv schedule with a guard band at the end of each best-effort slice
Figure 4. The same schedule with a guard band (G) at the end of each slice 2. The link stays idle during G unless a frame short enough to finish before slice 1 can be sent.Source: TSN IEEE 802.1Qbv Example with guard band.svg by Ulgorash, from the Wikipedia article “Time-Sensitive Networking”, Wikimedia Commons, CC BY-SA 4.0.

The guard band is bandwidth lost on every cycle. Frame preemption reduces it from a whole frame to a small fragment.

Time-gated queues with gates opening and closing for critical and non-critical traffic
Figure 4b. Scheduled traffic on the original slide: each queue has a gate (G), open or closed according to a periodic schedule; closing the non-critical gates in advance protects the critical frames.Source: J. Farkas, “TSN Basic Concepts,” DetNet–TSN workshop, IEEE 802.1, 11 November 2018, slide 11; cropped, with neighbouring slide text, logos or navigation icons blanked.
A transmission gate after each queue of an 802.1Q port, driven by a gate control list
Figure 4c. The structure added by 802.1Qbv, as drawn by the task group: a transmission gate after the transmission selection algorithm of each queue, all driven by one gate control list.Source: J. Farkas, “Introduction to IEEE 802.1: Focus on the Time-Sensitive Networking Task Group,” IEEE 802.1, 15 May 2017, slide 31; cropped, otherwise unchanged.
Express gate and preemptable traffic at an egress port, with and without a guard band
Figure 4d. The guard band on the original slide: the express gate opens for frames from two ports, and preemptable transmission is held (HOLD) and released (RELEASE) around it.Source: J. Farkas, “Introduction to IEEE 802.1: Focus on the Time-Sensitive Networking Task Group,” IEEE 802.1, 15 May 2017, slide 33; cropped, otherwise unchanged.

3. What a schedule has to know​

Network modelbridges and end stations; per link: propagation delay, processing delay, line rateegress queues for scheduled traffic (at most 8), maximum GCL length (8 to 1024)synchronisation error between clocksTraffic modelper stream: release time, period, payload, deadline, jitterrelease time fixed by the scheduler (fully) or by the talker (partially schedulable)deadline implicit, constrained or arbitrary; jitter zero or boundedScheduling modeldelay: no-wait, wait-allowed, or window-basedqueue isolation: none, FIFO, frame-based or stream-basedfixed routing or joint routing and scheduling; optional fragmentation or preemptionScheduler in the CNCoutput: one gate control list per egress port
Figure 5. The inputs of a TAS scheduling problem. A study of 17 methods sorts them into network, traffic and scheduling models. Each method picks one option per dimension, so two methods often solve different problems.Source: own drawing, after the system models of Section II of C. Xue, T. Zhang, Y. Zhou, M. Nixon, A. Loveless, S. Han, Real-Time Scheduling for 802.1Qbv Time-Sensitive Networking (TSN): A Systematic Review and Experimental Study, arXiv:2305.16772.

A 2023 systematic review and experimental study of TAS scheduling [4] starts from a precise statement of what a scheduler is given. Some of the inputs are practical constraints that are easy to forget:

  • GCL length is bounded. Bridges support only a limited number of entries, typically between 8 and 1024.
  • Priority-to-queue mapping may differ from bridge to bridge, so the same stream can sit in different queues along its path.
  • Traffic models differ. A fully schedulable model lets the scheduler choose the release time of each stream at the talker. A partially schedulable model takes release times as given.
  • Delay models differ. In a no-wait schedule, frames are forwarded without queuing at any hop. Wait-allowed schedules let frames wait in egress queues and have a larger solution space. Window-based models reserve windows for groups of streams and allow some jitter.
  • Queue isolation. If two streams share a queue, the FIFO order must match the schedule. Otherwise a lost or delayed frame shifts the others. Methods enforce FIFO, frame-based or stream-based isolation, with stream-based the strictest.

The 17 methods then split along three lines: fixed routing or joint routing and scheduling, no-wait or wait-allowed, and exact solutions (SMT or ILP solvers) or heuristics [4].

A TSN bridge with ingress ports, switching engine, priority filter, queues with gates and a gate control list
Figure 5b. The bridge model of the scheduling study: ingress ports, a switching engine, a priority filter that maps the PCP to a queue, and egress queues gated by a gate control list (GCL).Source: C. Xue, T. Zhang, Y. Zhou, M. Nixon, A. Loveless, S. Han, “Real-Time Scheduling for 802.1Qbv Time-Sensitive Networking (TSN): A Systematic Review and Experimental Study,” arXiv:2305.16772, Fig. 1; cropped, otherwise unchanged.

4. Lessons from the experimental study​

The study ran all 17 methods on the same stream sets, in simulation and on an eight-bridge testbed [4]. Its main conclusions:

  1. No method dominates. Each method wins only under some settings. Comparisons in earlier papers were often valid only for the setting they used.
  2. Simple models scale better. No-wait models and simple heuristics, such as list scheduling, handled large workloads better than exact formulations.
  3. Joint routing and scheduling can backfire. On large networks it can do worse than fixed routing, because the problem becomes too large to solve.
  4. Flexibility at the talker pays more than flexibility in the bridges. Letting the scheduler choose release times helped more than letting frames wait in bridges.
  5. With few queues, prefer no-wait.
  6. Real hardware adds constraints the models omit. These include collisions between scheduled frames and PTP messages, a minimum gap between frames at end stations (about 50 µs on the testbed, far more than in bridges), and the time granularity of GCL entries.

5. Where TAS fits​

TAS can isolate a critical stream completely from all other traffic. A review of industrial technologies notes that a well-configured TAS gives extremely low delay variation and a transmission delay bound of the order of 100 µs. It also notes that a small time error can desynchronise the windows and delay scheduled traffic by a whole cycle, and that TAS requires central management [5]. A 2026 tutorial on decentralised reservation leaves TAS out of its models for that reason. It adds that a lowest-latency schedule is not the same as a schedule that admits the most streams, because TAS reserves the link for a stream even when that stream sends nothing [6].

For worst-case analysis, a queue behind a gate sees a service that is zero while the gate is closed and the full link rate while it is open. The service curves post derives that service curve, and the TSN shapers post uses it.

6. Summary​

PropertyTime-Aware Shaper
StandardIEEE 802.1Qbv, now in IEEE 802.1Q
Acts onTransmission selection at an egress port, one gate per queue
StateGate control list, cycle time, base time
Needs time synchronisationYes
Configured byUsually a CNC that computes a network-wide schedule
Main costGuard bands, central computation, sensitivity to time errors

References​

  1. IEEE 802.1, Time-Sensitive Networking (TSN) Task Group. Web page
  2. Wikipedia, "Time-Sensitive Networking," section on scheduling and traffic shaping. Article
  3. J. Farkas, "TSN Basic Concepts," DetNet–TSN workshop, 11 November 2018. Slides
  4. C. Xue, T. Zhang, Y. Zhou, M. Nixon, A. Loveless, S. Han, "Real-Time Scheduling for 802.1Qbv Time-Sensitive Networking (TSN): A Systematic Review and Experimental Study," arXiv:2305.16772, 2023 (v4, 2024), Sections I–III and VII. arXiv
  5. I. Behnke, H. Austad, "Real-Time Performance of Industrial IoT Communication Technologies: A Review," arXiv:2311.08852, 2023, Section III-D. arXiv
  6. A. Grigorjew, S. Geissler, J. Schmitt, T. Hossfeld, "Decentralized Latency Models and Information Flow for TSN Shapers – A Tutorial," arXiv:2610.11999, 2026, Section VIII. arXiv