Markdown Version | Transcript | Recording 1 | Recording 2 | Recording 3 | Session Materials

MOQ

Summary

The Media over QUIC (MOQ) Working Group held its first session at IETF 126 in Vienna. The session focused on updates to the core transport protocol (draft-ietf-moq-transport), low overhead container formats (draft-ietf-moq-loc), end-to-end secure objects (draft-ietf-moq-secure-objects), and authorization design space challenges.

Key outcomes included agreement to spin off MOQ management and diagnostics work into a separate repository, a consensus direction to move ahead with a first Working Group Last Call (WGLC) for the secure objects draft once it is frozen alongside transport, and significant design team progress on defining challenge-response requirements for authentication.

Key Discussion Points

1. Welcome and Agenda Bashing

• Presenter: Martin Duke
• Slides: Chair Slides Monday
• Discussion:
  • A live experiment of watching MOQ over MOQ (WebRTC-to-MoQ translated by Meetecho) was shared during the meeting.
  • Cullen Jennings formally objected to the chairs' determination that there is consensus to push the top-end filter and Sender-Side Track Switching (SSTS) proposals into separate drafts, arguing that there was a split (roughly 50/50) in the working group and that the topic deserves more face-to-face agenda time.
  • Magnus Westerlund explained that managing these complex additions in separate GitHub repositories and trackers is the most pragmatic way to make progress without stalling the core transport draft.
  • Mo Zanaty suggested that the working group should focus on identifying the specific objections of those who did not favor merging the proposals.
  • The Area Director, Zaheduzzaman Sarker, noted that while the discussion is reasonable, it is the chairs' responsibility to call consensus and face-to-face time should not be consumed by debating consensus calls; dissenters may appeal if they believe the call is invalid.

2. MOQT Updates Since IETF 125

• Presenter: Alan Frindell
• Slides: MOQT Updates Since IETF 125
• Relevant Draft: draft-ietf-moq-transport
• Discussion:
  • Since the last cycle, 72 PRs have landed, primarily around range filters, URL schemes, roles, and media types.
  • Range Filters: Now merged, allowing filtering based on priority, subgroup ID, object ID, and integer properties. Property filters in a subscription match data-plane object properties, while property filters in a SUBSCRIBE_TRACKS match control-plane track properties.
  • Control Plane & Data Plane Changes: SUBSCRIBE_NAMESPACE was split (discovery vs. wildcard SUBSCRIBE_TRACKS). Multiple subscriptions to the same track are now permitted, treated as independent flows with duplicate objects. Delivery timeout was divided into distinct object-level and subgroup-level timeouts.
  • Progress Speed: Alan Frindell noted that the core draft's progress is bottlenecked by poor asynchronous engagement on GitHub. He urged the community to prioritize review times on the transport draft to ensure it stabilizes, warning that if transport does not ship, none of the other WG drafts can progress.

3. Low Overhead Container (LOC)

• Presenter: Mo Zanaty
• Slides: LOC
• Relevant Draft: draft-ietf-moq-loc
• Discussion:
  • Revisions -03 and -04 introduced public properties in the header and private properties inside the payload.
  • An audio config property (type 0x0f) mapping to WebCodecs decoder configs was added to address formats like AAC.
  • Registry Collisions: Standard property registries have collided across drafts (such as between LOC and draft-ietf-moq-secure-objects). Will Law queried if LOC and MSF should use the application-defined range to prevent collisions. Mo Zanaty noted that LOC relies on short 1-byte ranges. Rowan Wheeler highlighted that because relays may need to inspect certain properties (e.g., for speaker selection), having them standardized is beneficial.
  • Colin Perkins recommended looking at historic RTP parameters to ensure the WG doesn't miss standard video/audio properties that will need to be added down the line.
  • Open Issue: The WG is still debating the requirement for a "frame duration" property to assist decoders when dealing with non-constant video frame rates.

4. Secure Objects

• Presenter: Cullen Jennings
• Slides: Secure Objects
• Relevant Draft: draft-ietf-moq-secure-objects
• Discussion:
  • Updates include refinements to the threat model—specifically regarding bad relays trying to amplify attacks against downstream subscribers—and standardizing canonicalized security fields to fixed-width 64-bit integers.
  • Publisher priority was removed from end-to-end authenticated fields because relays can unilaterally override or ignore it anyway.
  • Object ID constraint: Object IDs are restricted to 32 bits to prevent the reuse of nonces under AES-GCM (which uses a 96-bit nonce). If an application attempts to use a larger Object ID, it will trigger an error. Alan Frindell questioned if draft-ietf-moq-transport should formally constrain Object IDs to 32 bits to mitigate this.
  • WGLC Readiness: Cullen Jennings proposed moving the draft to Working Group Last Call and parking it until the core transport registry is stabilized. Suhas Nandakumar supported this path. Mo Zanaty asked if the design works well when encapsulated inside or outside other wrappers (such as FEC). Cullen confirmed that the current design supports both generic inner and outer wrapping.

5. Auth Design Team Update - IETF 126

• Presenter: Mike English
• Slides: Auth Design Team Update - IETF 126
• Relevant Drafts: draft-ietf-moq-transport, draft-ietf-moq-c4m, draft-ietf-moq-privacy-pass-auth
• Discussion:
  • The design team was formed to solve the challenge of executing proof-of-possession (which requires a challenge-response exchange) on the wire.
  • Mike English detailed three auth topologies under consideration:
    1. Hop-by-hop delegated auth: Edge relays handle client authentication. Simple, but increases trust assumptions.
    2. Direct client-to-publisher validation with subscriber sampling: High complexity; the relay must forward validation tokens upstream for a subset of subscribers.
    3. Policy delegation: The publisher embeds validation rules in a token, which the relay executes.
  • Ted Hardie warned against designs that require passing subscriber databases or hash tables to relays, noting it introduces massive privacy issues and forces subscribers to reveal information they would otherwise keep private.
  • Chris Lemons and Jonathan Lennox suggested that if media is already end-to-end encrypted, the severity of a relay leaking unauthorized data payloads is significantly lower.
  • Cullen Jennings noted that in scenarios where the original publisher is a client without a trusted domain name or TLS certificate, a robust token binding scheme is required.
  • Spelling on the wire: The design team is evaluating whether challenge-response exchanges should occur inline on the subscription control stream or over a separate session-wide authentication control stream. Suhas Nandakumar and Chris Lemons favored a DPoP-style challenge-response flow.

6. MOQT Issues and PRs, Part 1

• Presenter: Alan Frindell
• Slides: MOQT Issues and PRs, Part 1
• Relevant Draft: draft-ietf-moq-transport
• Discussion:
  • Blocked Issues: Five issues remain blocked awaiting input from specific assignees.
  • Diagnostics & Metrics: Alan queried where relay diagnostics and management considerations should be documented. The room agreed that these should be moved out of the core transport draft to a separate document to save editor bandwidth. Alperen Temel (altanai) volunteered to co-author this new draft.
  • Timeouts: The draft currently advises using timeouts to protect resources, but the group discussed whether these should be negotiated explicitly on the wire. Ted Hardie, Ian Swett, and Colin Perkins expressed that "less is more." Only timeouts that require explicit application-layer remediation or prevent user-visible performance issues (e.g., subscription termination) should be signaled on the wire.

Decisions and Action Items

• Decision: Spin off the management, metrics, and diagnostics considerations of MOQT into a separate applicability/manageability document in a new repository.
• Action Item (Alperen Temel): Create the initial -00 draft for diagnostics and relay management, coordinating with Alan Frindell and Ian Swett.
• Decision: Remove "publisher priority" from the list of authenticated fields in draft-ietf-moq-secure-objects.
• Action Item (Cullen Jennings): Prepare draft-ietf-moq-secure-objects for a Working Group Last Call, to be held in a "freeze/parking lot" state pending the core transport registry completion.

Next Steps

• Auth Design Team: Continue evaluating the challenge-response signaling mechanisms (inline on subscription vs. a dedicated session-level authentication stream) and present progress by the October interim.
• Core Transport: WG members are tasked with clearing outstanding blocked issues on draft-ietf-moq-transport and focusing on asynchronous GitHub reviews to stabilize the protocol before the next hybrid interop.

Session Date/Time: 23 Jul 2026 14:30

MOQ

Summary

The MOQ working group met at IETF 126 to continue drafting and refining the Media over QUIC protocol suite. The session focused primarily on burning down issues and pull requests (PRs) related to the core transport specification, draft-ietf-moq-transport.

Key topics included:

• Sender-Side Track Switching (SSTS): Renamed from Dynamic Track Selection (DTS), incorporating modular algorithms and discussing draft integration.
• Location Filters: Simplifying and unifying syntax for specifying start and end positions.
• Transport-Layer Issue Resolutions: Addressing subscription notifications, parameter propagation in PUBLISH, track alias reuse, flow control, and version/extension negotiation.
• Live Agent Performance and Feedback: Examining experimental data and requirements for application- and transport-layer feedback in low-latency live media agent interactions.

Key Discussion Points

1. Sender-Side Track Switching (SSTS)

Will Law presented the update on Sender-Side Track Switching, formerly known as Dynamic Track Selection (DTS).

• Design Changes: The design has been split into a core framework and individual, extensible algorithms. This avoids ossification and allows custom switching behaviors. Algorithm 0 has been defined as a baseline, simple implementation.
• DDoS Protection: Per-session limits were removed because they are easily bypassed (e.g., by opening multiple sessions). Instead, relays must protect themselves internally by monitoring resource usage (such as the ratio of incoming to outgoing data) and terminating abusive sessions/subscriptions.
• Draft Integration vs. Separate Draft: A procedural discussion took place regarding how to publish this work.
  • Ian Swett expressed concern over editors having to unilaterally decide whether to merge a large PR into the base transport draft.
  • Colin Perkins proposed a hybrid approach: merge the structural modifications (such as how SSTS integrates into the transport layer) into draft-ietf-moq-transport, but extract specific algorithms (like Algorithm 0) into a separate document.
  • Magnus Westerlund noted there was previous consensus to adopt the functionality, but not specifically on how to package it.
  • Decision: The working group will delay a formal consensus call on the PR to allow further discussion of packaging options during an upcoming virtual interim.

2. Location Filters

Mo Zanati presented Location Filter (PR 1809), which aims to unify location ranges in the base transport protocol.

• Proposed Simplification: The PR eliminates the filter type enum entirely. Instead, locations are defined simply by optional start and end coordinates:
  • Omitting the end object implies all objects in that group.
  • Omitting the end group and object implies an open-ended subscription.
  • Omitting the start object represents a relative start group (e.g., starting at the current or previous group).
  • Setting length to zero removes the active filter.
• Integration with Fill Fetch: This syntax will also replace the enums currently proposed in the fill fetch PR (PR 1673), ensuring that live subscribe, fetch, and fill operations share a consistent spelling for location ranges.
• Discussion: Alan Frindell and Suhas Nandakumar supported the clean, lego-like structure of the new syntax, noting that it resolves prior ambiguities. Victor Vasiliev agreed that unifying the three ways of specifying ranges is highly beneficial.
• Decision: The working group agreed to merge PR 1809 into draft-ietf-moq-transport.

3. MOQT Transport Issues and PRs

Ian Swett and Alan Frindell led a walk-through of active issues and PRs for draft-ietf-moq-transport:

• Publisher Subscription Updates (PR 1803 / "Subscription State Update"):

  • Discussion on how a publisher can unilaterally notify a subscriber of state changes (e.g., track switches or falling out of a top-N filter).
  • Ian Swett proposed an unacknowledged "Subscription State Update" message on the bidirectional control stream.
  • Colin Perkins cautioned against control-plane amplification if a publisher sends millions of updates simultaneously. Mo Zanati noted that unacknowledged messages are ideal for preventing message storms.
  • Martin Duke suggested that this mechanism could be expanded to simplify the "publish done" state machine.
  • Decision: The PR will be updated to rename the message to "Subscription Notify" (or similar), with explicit constraints on when it can be sent (primarily in response to subscriber-initiated conditions). Martin Duke will contribute text to explore simplifying the "publish done" mechanism.

• Parameters in PUBLISH:

  • Discussion on whether to echo all subscription parameters (e.g., priority, filters) back in the PUBLISH message.
  • Martin Duke requested a comprehensive table in the draft detailing which parameters are allowed in which messages. Alan Frindell noted an issue already exists for this. Suhas Nandakumar pointed out that security parameters like authorization tokens should not be echoed.
  • Decision: Ian Swett will write a PR allowing appropriate subscription parameters (excluding security tokens) in PUBLISH and will draft the parameter summary table.

• Track Alias Reuse:

  • Discussed whether track aliases can be reused for different tracks once a prior subscription ends.
  • Alan Frindell favored allowing it to save byte overhead ("handing them the foot gun is just fine"). Colin Perkins recommended adding cautionary text so reviewers understand the race conditions.
  • Decision: Keep the status quo (allow reuse) but add cautionary text in the draft warning about the associated hazards.

• Resource Consumption / Flow Control (Issue 869):

  • Discussed subscription-level flow control to prevent individual subscriptions from starving others.
  • Victor Vasiliev argued against flow control at this level, suggesting application priorities should handle bandwidth starvation. Mo Zanati suggested that if flow control is added, it should only apply to stream limits, not data limits.
  • Decision: Move subscription-level flow control out of the base transport draft and develop it in a separate extension draft.

• Updates on PUBLISH_OK:

  • Currently, the protocol allows updating state variables (like location filters or priority) inside a PUBLISH_OK response.
  • Decision: Merge the proposal to disallow state updates in PUBLISH_OK. If a state update is needed, the sender must transmit an immediate REQUEST_UPDATE message instead.

• Which Token Caused the Error?:

  • If a request contains multiple authorization tokens and fails, the client does not know which token triggered the error.
  • Colin Perkins and Gwendal argued that clients need to know which token failed (especially during migrations from v1 to v2 tokens) to avoid infinite retry loops. Martin Duke and Alan Frindell debated whether this should be solved via integer indexes or punted.
  • Decision: Punt the issue to the Authorization Design Team.

• Structured MoQ Query Space:

  • Discussed whether to restrict MoQ URL query strings to standard key=value syntax. Colin Perkins and Mike Bishop favored constraining the space to improve extensibility, provided it remains convertible to HTTPS.
  • Decision: Revisit the issue on GitHub for further specification.

• Versions and Extensions:

  • Discussed the draft text requiring new MoQ versions to specify compatible extensions, and new extensions to specify compatible MoQ versions.
  • John and Ian Swett noted that requiring future MoQ versions to explicitly list all past compatible extensions is too restrictive. Mo Zanati suggested focusing on "setup options" since those are the actual objects registered with IANA.
  • Decision: Loosen the requirement for new MoQ versions to list compatible extensions, but retain the requirement for extensions to list compatible MoQ versions. Update the text to focus on setup options.

4. MoQ Performance and Feedback Requirements for Live Media and Agent Interaction

Yanmei presented MoQ Performance and Feedback Requirements for Live Media and Agent Interaction, sharing Alibaba's experimental results using MoQ for low-latency omnimodal AI agent interactions.

• Performance vs. WebRTC: MoQ demonstrated faster session setup (utilizing 0-RTT), better Quality of Service (QoS) via multi-stream prioritization, and improved Time to First Frame (TTFT) compared to WebRTC.
• Feedback Requirements: To optimize congestion control and media delivery, the application requires transport-layer feedback (e.g., object completion time, missed deadlines, receiver throughput) and QoE feedback (e.g., playout stalls, rebuffering).
• Relay Support: Proposed a two-tier feedback system: "last-mile" feedback consumed by local relays (L1), and "original" feedback forwarded to the media publisher. Relays can aggregate feedback over short periods (e.g., 5 seconds) to prevent feedback storms back to the origin.
• Discussion: Jordi Cenzano asked about the scale of the experiment, which Yanmei confirmed was between 100,000 and 1,000,000 connections. Mo Zanati suggested aligning this feedback aggregation with the top-tracks filter work to optimize upstream fan-in.

Decisions and Action Items

Session Date/Time: 24 Jul 2026 14:00

MOQ

Summary

The MOQ Working Group met at IETF 126 to discuss updates to its core streaming and transport specifications, review the latest interoperability testing results, and evaluate proposals for synchronized playback, real-time conferencing architectures, and selective track filtering. Key discussions focused on updating track serialization and configuration parameters, aligning interop testing goals, and refining the Top N track filtering design for inclusion in the base transport protocol.

Key Discussion Points

1. MSF & CMSF Updates

Will Law presented updates on draft-ietf-moq-msf and draft-ietf-moq-cmsf based on the MSF & CMSF presentation.

• Delta Update Mechanism: A newly merged PR allows dynamic track field updates via an override mechanism, avoiding the need to delete and recreate tracks.
• Initialization Data Properties: Discussions occurred around carrying initialization data via track properties or object properties.
  • Moe Zanaty noted that draft-ietf-moq-loc already registers track and object properties for binary configuration data.
  • Cullen Jennings and Rohan Mahy argued against using the "application-specific" (private use) range for standardized elements, suggesting they be moved to the standardized registry instead.
  • Suhas Nandakumar suggested dividing the space appropriately between track and object properties.
• Dissimilar Group Length Switching: Discussion on a PR allowing clients to switch across tracks with differing group lengths. Gwendal Simon noted that group ID alignment and switch-from parameters must be clearly mapped to ensure smooth transitions.
• Replacing Catalog with Null Track Name: A proposal to use the null/empty track name to represent the default catalog, simplifying URL paths.
  • Suhas Nandakumar and Moe Zanaty discussed the potential ambiguity when doing prefix matching with empty names, noting that a delimiter (like dash dash) might still be needed to prevent ambiguities.
• Subgroup Zero Constraint (PR 169): A proposal requiring all MSF objects to be mapped to subgroup zero was discussed. Moe Zanaty and Gwendal Simon expressed concern that this restriction would inhibit future temporal scalability designs.

2. MoQ Interop Report

Mike English presented the MoQ Interop Report, highlighting findings from the automated interop runner and manual wiki reports.

• Interoperability Status: Progress was made on several draft-18 implementations, though some regressions were noted compared to draft-16, mainly due to the complexity of transitioning to bidirectional streams.
• Test Integration: Alan Frindell (via chat/relay) emphasized the need to integrate data-plane test suites (such as verifying PUBLISH_DONE propagation and prefix matching) into the automated runner.
• Testing Roadmap: To address current gaps, Alan Frindell and Mike English proposed a virtual interop day targeting draft-18 in early September, with the goal of moving implementations to draft-20 for the Seattle meeting.
• Interop Levels: Jordi Cenzano proposed defining distinct "interop levels" (e.g., basic control messages vs. full media synchronization) to help categorize implementation capabilities.

3. Tempo: Synchronized Playout

Suhas Nandakumar presented tempo, a proposal for synchronized media playout across disparate clients (e.g., for live sports, multi-room audio, and emergency alerts).

• Protocol Metrics: The design relies on three metadata properties: capture timestamp, playout delay, and hop time (updated by relays).
• Architecture & Feedback Loop: The draft describes a system where clients send playback status back to a central synchronization server, which instructs the publisher to adjust playout delays.
  • Will Law and Ali C. Begen raised scalability concerns regarding feedback loops back to the publisher for large-scale events. They suggested relying on local client adjustments (speeding up or slowing down playback) and cumulatively summing hop times instead.
  • Cullen Jennings suggested that statistical feedback could be gathered efficiently using a Top N track filter.

4. Mocha: Video Conferencing over MoQ

Cullen Jennings presented mocha, showing how MoQ transport can support decentralized, multi-party video conferencing and chat without dedicated Selective Forwarding Units (SFUs) or application-layer media servers.

• Design & Specifications: The architecture coordinates namespaces, catalogs, and encryption keys using MLS-based end-to-end security via draft-ietf-moq-secure-objects and Common Access Tokens (draft-ietf-moq-c4m).
• Relay Optimization: The system shifts active speaker selection and track switching logic onto the MoQ relays using the Top N track filtering mechanism, optimizing bandwidth without needing specialized media processing.

5. Top N Tracks Filter

Moe Zanaty presented the Top N Tracks Filter, focusing on the design proposed in PR 1518.

• Use Cases: Essential for conferencing (active speaker selection), security camera grids, esports, and managing high-volume feedback tracks.
• Bandwidth Benefits: Relays propagate Top N filters upstream, ensuring pruning occurs close to the publishers and significantly reducing relay-to-relay traffic.
• Simplified Design: Recent updates restricted the maximum selected tracks to 255 (allowing 1-byte encoding) and simplified the relay state machine. Relays evaluate publisher-provided object metrics to dynamically select and forward active tracks.
• Status: Magnus Westerlund noted there is working group consensus to work on this feature. The discussion centered on whether to merge it into the base transport draft (draft-ietf-moq-transport) or publish it as an extension. Will Law and Rohan Mahy supported making it a core feature.

Decisions and Action Items

• MSF & CMSF Properties: Will Law and Moe Zanaty to collaborate on moving track and object configuration properties out of the private application space and registering them as standard codepoints.
• Private Use Terminology: Working group to update the terminology in draft-ietf-moq-transport to clarify that "application-specific use" is intended for private use.
• Subgroup Zero Mapping: Will Law to follow up on PR 169 to address concerns regarding the restriction of MSF mapping exclusively to subgroup zero.
• Virtual Interop Day: Mike English has the authority to schedule a virtual interop day for draft-18, targeted for early September.

Next Steps

• Top N Tracks Filter: Moe Zanaty to create a clean, dedicated PR containing only the Top N tracks filter (separating it from other range filters).
• Mailing List Call: Once the clean PR is available, the chairs will initiate a consensus call on the mailing list to determine if the Top N filter is ready to be merged directly into draft-ietf-moq-transport.

Related Documents

draft-16, draft-18, draft-20, draft-ietf-moq-c4m, draft-ietf-moq-cmsf, draft-ietf-moq-loc, draft-ietf-moq-msf, draft-ietf-moq-privacy-pass-auth, draft-ietf-moq-secure-objects, draft-ietf-moq-transport