Skip to main content

IEEE 802.1Qat, 802.1Qcc and P802.1DD: stream reservation and admission control

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

A shaper bounds latency only if the traffic it serves stays within what it was configured for. Something must therefore decide, before a stream starts, whether the network can carry it. This post covers the distributed way of making that decision, in which the bridges themselves admit or reject streams. The centralised alternative is covered in the SDN post.

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 stream reservation acts: in the control plane of every bridge. It changes no forwarding stage. It decides which streams the shapers and queues below will have to carry.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 (10)
AbbreviationStands forMeaning
ATSAsynchronous Traffic ShapingIEEE 802.1Qcr: per-stream token-bucket shaping at every hop using local clocks only.
AVBAudio Video BridgingThe predecessor of TSN: gPTP, the credit-based shaper and stream reservation, combined for audio and video (profile: IEEE 802.1BA).
CBSCredit-Based ShaperThe IEEE 802.1Qav shaper that limits a traffic class to its reserved rate (idleSlope) with a per-queue credit. In MEF bandwidth profiles, CBS instead means Committed Burst Size.
CNCCentralized Network ConfigurationThe 802.1Qcc entity that computes paths, schedules and reservations for the whole network and configures the bridges.
CQFCyclic Queuing and ForwardingIEEE 802.1Qch: two queues per class alternate every cycle, so a frame advances one hop per cycle.
IEEEInstitute of Electrical and Electronics EngineersThe professional association whose Standards Association publishes the IEEE 802 LAN/MAN standards.
MACMedium Access ControlThe lower sublayer of the data link layer, which frames data and controls access to the medium. A MAC address identifies a network interface.
RAPResource Allocation ProtocolIEEE P802.1DD: a newer reservation protocol for centralised and decentralised TSN networks, with token-bucket traffic specifications.
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).
SRPStream Reservation ProtocolIEEE 802.1Qat: distributed admission control in which talkers advertise streams and listeners subscribe, reserving resources hop by hop.

1. Definition​

Stream reservation is distributed admission control. A talker declares a stream and its traffic, listeners declare interest, and every bridge on the path checks and commits its resources before the stream may send [2]. Three generations of standards do this [2][3]:

StandardProtocolStatus
IEEE 802.1Qat (now in 802.1Q)Stream Reservation Protocol (SRP), built on the Multiple Registration ProtocolPublished; the basis of AVB
IEEE 802.1Qcc (now in 802.1Q)SRP enhancements, plus the user/network interface and the centralised modelsPublished
IEEE P802.1DDResource Allocation Protocol (RAP), a publish–subscribe protocol over the Link-local Registration Protocol (802.1CS), for centralised and decentralised networksIn development [4]

2. SRP: advertise, then subscribe​

The protocol itself is described in the SRP post; this section summarises what matters for the analysis.

A talker advertise propagating from bridge to bridge; in each bridge a registration on the input port becomes a declaration on the output ports
Figure 2. A talker advertise spreading through the bridges. In each bridge the advertise is registered (R) on the port where it arrived and declared (D) on the other ports, so it reaches the intended listener and other end stations.Source: Talker-Advertise-b.pdf by Michael Johas Teener (Mikejt), from the Wikipedia article “Time-Sensitive Networking”, Wikimedia Commons, CC BY-SA 3.0; page 1 rendered from PDF to PNG, otherwise unchanged.

In SRP a talker sends a talker advertise carrying the stream identifier, its traffic class and frame rate, and an accumulated latency that each bridge increases. A listener that wants the stream answers with a listener ready, which travels back towards the talker. Each bridge it passes commits bandwidth for the stream. If a bridge lacks resources, the advertise becomes a talker failed and carries the reason [2]. Streams are identified by 64 bits: the talker's 48-bit MAC address and a 16-bit number [2].

The weakness of SRP was its analysis, not its signalling. The standards gave example latency formulas for the accumulated latency, but these lacked formal proof and were later shown to be inaccurate [4].

3. What a bridge must know to compute a local bound​

Talker announceburst 120 B, rate 175 kbit/sacc. max latency 500 µsTalker announcesame traffic specificationacc. max latency 1.5 msTalker announcesame traffic specificationacc. max latency 2.2 msTalkerBridge ABridge BListenerlistener ready travels back; each bridge commits its resources
Figure 3. The information a talker announcement carries in a decentralised reservation. The traffic specification stays the same from hop to hop; the accumulated maximum latency grows by the threshold of each delay segment.Source: own drawing, after A. Grigorjew, S. Geissler, J. Schmitt, T. Hossfeld, Decentralized Latency Models and Information Flow for TSN Shapers, arXiv:2610.11999, 2026, Fig. 7 (values from that example).

The 2026 tutorial on decentralised latency models works out exactly what a bridge needs to bound the latency of its segment without global knowledge. It contributes this analysis to the RAP work [4]. Its list, paraphrased:

InformationWhenPurpose
Latency threshold ΔsegP\Delta^P_{\text{seg}} per segment and priorityConfigured beforehandUpper limit on the latency a segment may consume; added to every stream's accumulated maximum latency
Shaper of each upstream traffic class, and its parameters (CBS idle slope, CQF cycle)Exchanged with neighbours beforehandEstimate the burstiness of incoming traffic
Stream identifier and header fieldsPer stream, at reservationMatch frames to the stream and its priority
Traffic specification: burst bb, rate rr, minimum and maximum frame lengthPer streamBound the traffic volume: w(t)≤b+rtw(t) \le b + r t
Accumulated maximum and minimum latency L+L^{+}, L−L^{-}Per stream, updated at each hopBound the extra burstiness caused by delay variation
Upstream traffic classPer streamCompute the upstream queuing delay with the upstream bridge's own class mapping
Speed of the previous linkPer streamTighter bound using serialisation (optional)

At each hop the accumulated latencies are updated as

L+←L++ΔsegP,L−←L−+ℓmin⁡Rseg.L^{+} \leftarrow L^{+} + \Delta^P_{\text{seg}}, \qquad L^{-} \leftarrow L^{-} + \frac{\ell^{\min}}{R_{\text{seg}}}.

The pre-configured threshold is used rather than the actual current bound. The actual bound changes with every new reservation, whereas the threshold stays valid as other streams come and go [4]. The traffic specification follows the token-bucket model of the arrival curves post, which replaces SRP's frames-per-interval description [4].

4. The bound a bridge computes​

Two steps give the simplest bound of the tutorial, for strict priority without shaping [4].

Effective burst. A stream that has accumulated delay variation L+−L−L^{+} - L^{-} can arrive at a queue with a burst of at most

beff=b+r (L+−L−).b^{\text{eff}} = b + r\,(L^{+} - L^{-}).

This is the output-burstiness result of the fundamental bounds post, applied with the jitter accumulated so far.

Per-class bound. Let CC be the streams in the observed upstream class, HH the streams of higher classes, ℓmax⁡\ell^{\max} the largest lower-priority frame and RsegR_{\text{seg}} the link rate. Then the worst-case delay of the segment is bounded by

dsegP≤∑i∈H(bieff+ri ΔsegP)+∑i∈Cbieff+ℓmax⁡Rseg.d^P_{\text{seg}} \le \frac{\sum_{i \in H} \left(b_i^{\text{eff}} + r_i\,\Delta^P_{\text{seg}}\right) + \sum_{i \in C} b_i^{\text{eff}} + \ell^{\max}}{R_{\text{seg}}}.

Admission control accepts a new stream only if, after adding it, dsegP≤ΔsegPd^P_{\text{seg}} \le \Delta^P_{\text{seg}} still holds for every priority on the segment. The tutorial gives tighter network calculus versions and versions for CQF, ATS and CBS. All follow the same pattern: bound the workload from the traffic specifications and accumulated latencies, then divide by the service rate [4].

Centralized network / distributed user model: a CNC managing bridges, with user/network configuration at the edge
Figure 3b. The centralized network / distributed user model of 802.1Qcc as drawn by the task group: end stations exchange user/network configuration information at the edge, and a CNC manages the bridges.Source: J. Farkas, “Introduction to IEEE 802.1: Focus on the Time-Sensitive Networking Task Group,” IEEE 802.1, 15 May 2017, slide 38; cropped, otherwise unchanged.

5. Distributed or centralised?​

A distributed protocol has no single point of failure and copes with bridges whose internals a central controller would not know [4]. A review of industrial IoT technologies points out the other side. A reservation can fail at any bridge, and some failures only appear at run time. A small change in the network can reroute traffic and invalidate earlier reservations. This is the argument for a central controller that sees the whole network [5].

RAP is designed for both cases. Decentralised models can also join two centrally configured domains, as shown in the tutorial's motivating scenarios [4].

References​

  1. IEEE 802.1, Time-Sensitive Networking (TSN) Task Group. Web page
  2. Wikipedia, "Time-Sensitive Networking," sections on AVB and stream reservation. Article
  3. J. Farkas, "TSN Basic Concepts," DetNet–TSN workshop, 11 November 2018, slide 18. Slides
  4. A. Grigorjew, S. Geissler, J. Schmitt, T. Hossfeld, "Decentralized Latency Models and Information Flow for TSN Shapers – A Tutorial," arXiv:2610.11999, 2026, Sections I, V, VI-A and VIII. arXiv
  5. I. Behnke, H. Austad, "Real-Time Performance of Industrial IoT Communication Technologies: A Review," arXiv:2311.08852, 2023, Section III-D. arXiv