Media over QUIC Overview

1.1. Motivation

Today's media delivery landscape uses separate protocols for different stages of the pipeline. Content is typically ingested using protocols like RTMP, then repackaged at origin servers for distribution via HTTP-based protocols such as HLS or DASH. This separation introduces repackaging overhead, limits end-to-end latency, and creates complexity at the boundaries between protocol domains.¶

MoQ addresses this by providing a single protocol that can carry media from original publisher to end subscriber, including through intermediate relay nodes. The design is driven by four goals:¶

Latency:

TCP-based protocols suffer from head-of-line blocking and are slow to respond to congestion. MoQ leverages QUIC's multiplexed streams and partial reliability to minimise queuing latency and react quickly to changing network conditions.¶

Leveraging QUIC:

QUIC provides parallel streams with independent loss recovery, unreliable datagrams, stream priorities, and flow control. MoQ maps its data model onto these features to achieve flexible trade-offs between latency and reliability.¶

Convergence:

A single protocol from contribution through distribution eliminates the need for intermediate repackaging. The same MoQ objects produced by an encoder can be delivered to end subscribers without transformation at relay nodes.¶

Relays:

Scalable media delivery requires intermediaries that can cache and fan out content. MoQ treats relays as first-class protocol participants: object metadata is structured so that relays can route, prioritise, and cache content without accessing the media payload itself.¶

The resulting protocol is a general-purpose pub/sub delivery layer. The media-specific aspects — how audio and video samples are packaged, how tracks are described in a catalog, how adaptive bitrate switching works — are defined by separate streaming format specifications that build on MOQT. Applications outside media delivery can use MOQT directly with their own object formats, without adopting any of the media-specific layers.¶

1.2. Document Scope

This document is informational. It does not define protocol behaviour or create new requirements. Its purpose is to:¶

• Orient readers to the MoQ architecture and its components.¶

• Explain how the components relate to each other.¶

• Point readers to the relevant normative specification for each topic.¶

The covered specifications are:¶

• MOQT (Media over QUIC Transport) [I-D.ietf-moq-transport] — the publish/subscribe transport protocol.¶

• MSF (MOQT Streaming Format) [I-D.ietf-moq-msf] — a streaming format defining media packaging, catalog, and timeline.¶

• CMSF (CMAF-compliant MSF) [I-D.ietf-moq-cmsf] — an extension of MSF for CMAF/ISO-BMFF packaged media.¶

• MoQ Secure Objects [I-D.ietf-moq-secure-objects] — end-to-end authenticated encryption for content objects.¶

• Privacy Pass Authentication [I-D.ietf-moq-privacy-pass-auth] — privacy-preserving token-based authorization.¶

• Common Access Tokens for MoQ (C4M) [I-D.ietf-moq-c4m] — bearer token authorization scheme.¶

• LOC (Low Overhead Container) [I-D.ietf-moq-loc] — a minimal media container format.¶

1.3. Terminology

This document uses the following terms. Definitions attempt to be consistent with [I-D.ietf-moq-transport] but in case of discrepancy rely on the MOQT specification.¶

Client:

The party initiating a Transport Session.¶

Server:

The party accepting an incoming Transport Session.¶

Publisher:

An endpoint that handles subscriptions by sending objects from a track.¶

Subscriber:

An endpoint that subscribes to and receives tracks.¶

Original Publisher:

The initial publisher of a given track — the entity that creates the media content.¶

End Subscriber:

A subscriber that consumes the content and does not forward it to other subscribers.¶

Relay:

An entity that is both a publisher and a subscriber, forwarding content between other endpoints. Relays are neither the original publisher nor the end subscriber.¶

Track:

A time-ordered sequence of groups, identified by a Full Track Name. A track is the unit of subscription.¶

Group:

A collection of objects within a track that forms a join point. Groups are independently useful — objects within a group should not depend on objects in other groups.¶

Object:

The smallest addressable data unit in MoQ. An object is an immutable sequence of bytes identified by its track, group ID, and object ID.¶

Subgroup:

A sequence of objects within a group that share a QUIC stream. Objects in a subgroup have consistent priority and dependency relationships.¶

Transport Session:

A QUIC connection or WebTransport session over which MOQT operates.¶

2.1. Live Streaming Media Delivery

In live streaming, an original publisher encodes and publishes media content that is delivered through one or more relays to a potentially large number of end subscribers.¶

 Original                                            End
 Publisher --> Relay --> Relay --> ... --> Relay --> Subscribers
                                                     (many)

Key characteristics:¶

• One-to-many distribution, typically through a tree of relays.¶

• Target end-to-end latency ranges from under one second to several seconds, depending on the application.¶

• Subscribers may join at any time and must be able to start playback from a recent group boundary.¶

• Adaptive bitrate (ABR) switching between alternate encodings of the same content allows subscribers to adapt to varying network conditions.¶

• The original publisher produces a catalog describing available tracks (resolutions, bitrates, codecs) so subscribers can make informed selections.¶

2.2. Real-Time Conferencing

In conferencing, multiple participants each act as both original publisher and end subscriber. Each participant publishes their own audio and video tracks and subscribes to tracks from other participants.¶

 Participant A ---+            +--- Participant C
                  |            |
                  v            v
              +-------+    +-------+
              | Relay |<-->| Relay |
              +-------+    +-------+
                  ^            ^
                  |            |
 Participant B ---+            +--- Participant D

Key characteristics:¶

• Many-to-many communication with low participant counts (tens to hundreds).¶

• Target latency is interactive: below 500 milliseconds end-to-end.¶

• Each participant subscribes to a subset of available tracks (e.g., active speakers only).¶

• Scalable video coding (SVC) or simulcast may be used, with the relay or subscriber selecting appropriate layers.¶

• End-to-end encryption is particularly important since relays should not have access to media content.¶

2.3. Content Contribution and Primary Distribution

In content contribution (ingest), a professional encoder at a production site publishes high-quality media to a relay operated by a distribution service (e.g., a CDN).¶

 Encoder/       CDN Ingress      CDN Edge
 Producer  -->  Relay       -->  Relay(s)  -->  Subscribers

Key characteristics:¶

• The ingress relay receives the contributed content and may transcode it into multiple representations (resolutions, bitrates).¶

• After transcoding, the distribution service becomes the original publisher of the new representations — the transcoded tracks are distinct from the contributed track.¶

• Reliability is prioritised over latency for the contribution link.¶

• The same MOQT protocol is used for both the contribution link (encoder to ingress) and the distribution links (edge relays to subscribers), achieving the convergence goal.¶

3.1. Roles

A MOQT deployment involves endpoints in the following roles:¶

Publisher:

Sends objects belonging to one or more tracks. A publisher responds to subscriptions and delivers matching objects to subscribers.¶

Subscriber:

Requests tracks and receives objects. A subscriber issues SUBSCRIBE or FETCH requests to obtain content.¶

Original Publisher:

The endpoint that creates and first publishes a track's content. The original publisher is responsible for the media encoding, object payload, and track properties.¶

End Subscriber:

The final consumer of a track's content. An end subscriber decodes and renders the media.¶

A single endpoint may act in multiple roles simultaneously. In conferencing, each participant is both an original publisher (for their own media) and an end subscriber (for other participants' media).¶

3.2. Relays

Relays are intermediary nodes that are both publishers (to downstream subscribers) and subscribers (to upstream publishers). They terminate MOQT sessions on each side independently — a relay is not a transparent proxy.¶

Relays provide:¶

• Fan-out: A single upstream subscription serves multiple downstream subscribers requesting the same track.¶

• Caching: Objects can be stored and served to subscribers that join later, without fetching from the original publisher again.¶

• Geographic distribution: Relays can be deployed close to subscribers to reduce latency.¶

Relays are content-agnostic: they forward object payloads without interpreting or modifying them. This enables end-to-end encryption where relays cannot access media content, while still performing their routing and caching functions based on object metadata (track names, group IDs, priorities, properties).¶

Relays are enforcement points for access and authorization. In many deployments that have any restriction on access of the content the relay is expected to be the node that receives proof of authorization and will be responsible to verify and periodically re-verify this proof before delivering content.¶

Relays are also not necessarily application-agnostic when it comes to deployment and operation. The configuration of relay topologies, authorization policies, caching strategies, and resource limits are application-specific concerns. A relay serving a live streaming platform operates differently from one supporting a conferencing service, even though the protocol mechanics are identical. Moreover, even though relays need not parse media payload formats in order to forward content, they may still apply application-aware policy based on visible metadata, naming conventions, and authorization state.¶

For further discussion of relay operational considerations, including denial-of-service resilience, see [I-D.englishm-moq-relay-dos].¶

3.3. Deployment Topologies

MOQT supports several deployment topologies:¶

Point-to-point:

A publisher connects directly to a subscriber with no intermediate relay. Suitable for contribution ingest, AI produced dissemination, or testing.¶

 Publisher <--> Subscriber

Single relay:

A publisher connects to a relay, which serves one or more subscribers. The relay provides fan-out and caching.¶

 Publisher <--> Relay <--> Subscriber(s)

Relay tree:

A publisher connects to an origin relay, which fans out to edge relays closer to subscribers. This is the typical CDN deployment model.¶

 Publisher <--> Origin Relay <--> Edge Relay(s) <--> Subscriber(s)

Relay mesh:

Relays interconnect with multiple peers, subscribing and publishing bidirectionally between each other. Each relay both receives content from and forwards content to its neighbouring relays. This topology suits conferencing and multi-publisher scenarios where content flows in multiple directions simultaneously and there is no single origin.¶

 Publisher A <--> Relay 1 <----> Relay 2 <--> Subscriber C
                   ^    ^       ^    ^
                   |     \     /     |
                   |      v   v      |
                   |     Relay 3     |
                   |       ^         |
                   v       |         v
         Subscriber B  Publisher D  Subscriber E

In all topologies, each hop is an independent MOQT session. Relays manage subscriptions on each session independently and can aggregate multiple downstream subscriptions into a single upstream subscription.¶

MOQT supports graceful session migration: a relay that needs to shut down sends a GOAWAY message directing its peers to reconnect to a different server. The peers then establish new sessions and re-issue their subscriptions, allowing relay maintenance or replacement, but continuity depends on object availability, timing, and application/session policy.¶

4.1. Hierarchy

The data model has four levels:¶

 Track
   └── Group (join point)
         └── Subgroup (stream-mapped unit)
               └── Object (smallest addressable payload)

Track:

A named, time-ordered sequence of groups. Tracks are the unit of subscription — a subscriber requests a track by its Full Track Name and receives objects from that track.¶

Group:

A temporal collection of objects within a track. A group represents a join point: a subscriber can begin receiving content at any group boundary without needing prior groups. For video, a group typically corresponds to a Group of Pictures (GOP) starting with a random access point.¶

Subgroup:

A sequence of objects within a group that are sent on a single QUIC stream. Objects in a subgroup share priority and have dependency relationships consistent with in-order delivery. Subgroups enable the priority mechanism to favour certain objects (e.g., base-layer frames) over others within the same group.¶

Object:

The basic data element — an addressable, immutable sequence of bytes. Each object is uniquely identified by its Full Track Name, group ID, and object ID. Objects within a group are ordered by object ID.¶

4.2. Track Naming

Every track is identified by a Full Track Name consisting of two parts:¶

Track Namespace:

An ordered tuple of one or more fields ( up to 32) (e.g., "example.com", "meeting123", "alice"). The namespace provides hierarchical structure that relays use for prefix-based routing and subscription matching.¶

Track Name:

A byte sequence identifying a specific track within the namespace (e.g., "video-hd", "audio", "catalog").¶

The combination of namespace and track name is unique within a given scope. A scope is defined by the server or set of servers that a client connects to — specifically, those sharing the same host and path in the connection URI. Within a scope, no two distinct tracks can have the same Full Track Name, which means it can serve as a cache key at relays.¶

4.3. Properties

Tracks and objects can carry additional metadata called Properties. Properties are visible to relays and can influence how content is distributed. There are two scopes of properties — track-wide and per-object — and within either scope, individual properties can optionally be marked as immutable to support end-to-end authentication.¶

Track Properties:

Metadata associated with an entire track. Examples include default publisher priority, group order preference, maximum cache duration, and delivery timeouts. Track Properties are communicated in control messages (PUBLISH, SUBSCRIBE_OK, FETCH_OK).¶

Object Properties:

Per-object metadata carried in object headers. MOQT defines the framework for object properties, including gap indicators (Prior Group ID Gap, Prior Object ID Gap). Media-specific properties such as timestamps and frame markings are defined by LOC and MSF for use with encoded audio/video samples.¶

Immutable Properties:

Either track or object properties may be placed inside an Immutable Properties container. Once set by the original publisher, these cannot be modified or removed by relays. This enables end-to-end authentication: subscribers can verify that immutable values have not been altered in transit.¶

Properties are registered in IANA registries and use a Key-Value-Pair encoding. Relays that do not understand a property must forward it unchanged.¶

4.4. Object Immutability and Cacheability

A fundamental property of MOQT objects is immutability: once an object is published, its payload must never change. Two received objects with the same Full Track Name, group ID, and object ID must contain identical bytes.¶

This guarantee enables:¶

• Caching at relays without cache-invalidation complexity.¶

• Deduplication when the same object arrives from multiple upstream sources.¶

• End-to-end integrity verification — a subscriber can detect if an object has been tampered with.¶

An object can transition from "exists" to "does not exist" (e.g., cache expiry), but its contents cannot be modified. The protocol signals object status (normal, end-of-group, end-of-track) to communicate lifecycle information.¶

5.1. Session Establishment

MOQT runs over either native QUIC or WebTransport:¶

• Native QUIC: The client connects using the TLS Application-Layer Protocol Negotiation (ALPN) identifier "moqt" and identifies the server using a "moqt://" URI. The authority and path are communicated via Setup Options.¶

• WebTransport: The client establishes a WebTransport session over HTTP/3, using an HTTPS URI derived from the moqt:// URI.¶

Both provide stream multiplexing and secure transport, though the deployment model differs — particularly in browser-based environments where only WebTransport is available. Applications should treat them as functionally similar substrates with some environment-specific constraints.¶

After the transport connection is established, each endpoint opens a unidirectional control stream and sends a SETUP message. The SETUP exchange negotiates the protocol version and extensions (Setup Options) before any other messages are exchanged.¶

5.2. Publishing and Subscribing

MOQT provides several mechanisms for publishers and subscribers to exchange content:¶

SUBSCRIBE:

A subscriber requests newly published objects for a track. The publisher responds with SUBSCRIBE_OK and begins forwarding matching objects as they are produced. Subscription filters (Largest Object, Next Group, AbsoluteStart, AbsoluteRange) control which objects are delivered.¶

PUBLISH:

A publisher initiates a subscription by pushing content to a subscriber. This inverts the usual direction — the publisher offers a track rather than waiting for a subscription request.¶

FETCH:

A subscriber requests a specific range of previously published objects. Unlike SUBSCRIBE, which delivers objects as they are produced going forward, FETCH retrieves objects that already exist. A subscriber that joins a track mid-stream typically needs both: a SUBSCRIBE to receive new objects, and a FETCH (called a Joining Fetch) to retrieve the recent objects needed to start decoding — for example, the beginning of the current video group of pictures.¶

Namespace Discovery:

Publishers advertise available namespaces via PUBLISH_NAMESPACE. Subscribers discover content by sending SUBSCRIBE_NAMESPACE (to learn about matching namespaces) or SUBSCRIBE_TRACKS (to receive PUBLISH messages for tracks within matching namespaces). Relays use namespace prefix matching to route subscriptions to the appropriate upstream publishers.¶

All subscription interactions follow a state machine: Idle → Pending → Established → Terminated. Either endpoint can terminate a subscription — the subscriber via cancellation, the publisher via PUBLISH_DONE.¶

5.3. Priorities and Delivery

When network capacity is insufficient to deliver all subscribed content, MOQT's priority system determines what gets delivered first:¶

• Subscriber Priority: Set per-subscription by the subscriber to express relative importance among its own subscriptions.¶

• Publisher Priority: Set per-object by the publisher to express relative importance within a track (e.g., base layer vs. enhancement layer).¶

• Group Order: Ascending (oldest first) or descending (newest first) delivery of groups within a subscription.¶

Under congestion, lower-priority content may be delayed or dropped entirely. Two timeout mechanisms support controlled data loss:¶

• Object Delivery Timeout: If an object cannot be sent within this duration after its first byte was produced, it is discarded.¶

• Subgroup Delivery Timeout: If a completed subgroup's data is not fully acknowledged within this duration, the stream is reset — meaning any undelivered data on that stream is abandoned and will not be retransmitted.¶

Objects can be sent on QUIC streams (reliable, in-order within the stream) or as QUIC datagrams (unreliable, low-latency). The choice is the Object Forwarding Preference, set per-object by the original publisher.¶

Client side Adaptive Bitrate (ABR) is possible by either ending a subscription and replace it with another track that is another represenation at another bit-rate than the previous one. Steering these changes to be done at the next group boundary.¶

The WG is working on sender side ABR where the publisher uses its understanding of the congestion situation to select the set of tracks to be published to the subscriber.¶

5.4. Extensibility and Session Management

MOQT is designed for evolution without breaking existing implementations:¶

Setup Options:

Negotiated per-session in the SETUP exchange. Unknown options are ignored, allowing new features to be introduced without breaking older endpoints.¶

Message Parameters:

Per-message metadata included in control messages. Parameters are peer-to-peer and are not forwarded by relays.¶

Properties:

Per-track and per-object metadata that is forwarded and cached by relays, unlike Message Parameters. Properties use IANA-registered type codes. Unknown properties must be forwarded unchanged.¶

GREASE (Generate Random Extensions And Sustain Extensibility):

Reserved code points that implementations may send to exercise their peers' handling of unknown values. This ensures that deployed software does not break when new extensions are introduced.¶

GOAWAY supports graceful session migration and can reduce disruption during relay maintenance or replacement, but continuity depends on object availability, timing, and application/session policy.¶

6.1. Role of a Streaming Format

A streaming format specifies:¶

• How encoded media samples are mapped to MOQT objects and groups.¶

• A catalog format describing available tracks and their properties.¶

• Rules for time-alignment and ABR switching between tracks.¶

• Timeline information mapping MOQT group/object IDs to media presentation time.¶

• Any additional metadata tracks (events, logs, metrics).¶

Without a streaming format, a subscriber receiving objects would have no way to discover what content is available, how to decode it, or how to synchronise multiple tracks.¶

6.2. MSF: MOQT Streaming Format

The MOQT Streaming Format (MSF) [I-D.ietf-moq-msf] is the primary streaming format defined for MoQ. It uses LOC (Low Overhead Container) [I-D.ietf-moq-loc] packaging, where each encoded media sample (audio frame or video frame) is placed in a separate MOQT object. This one-sample-per-object mapping allows the transport to make independent delivery decisions for each frame — for example, prioritising base-layer video frames over enhancement layers, or dropping an expired frame without affecting the delivery of subsequent frames.¶

MSF defines:¶

Catalog:

A JSON track named "catalog" that describes all available tracks in a namespace. The catalog includes codec information, bitrates, resolutions, language, and relationships between tracks (render groups for synchronised playback, alternate groups for ABR switching). Delta updates allow efficient signalling of track additions and removals.¶

Media Timeline:

An optional track that maps MOQT group/object IDs to media presentation timestamps and wallclock times. This enables seeking and synchronisation. A template mechanism supports regular patterns without requiring an explicit timeline track.¶

Event Timelines:

Optional metadata tracks carrying time-synchronised application data (e.g., sports scores, ad insertion markers, GPS coordinates).¶

Publish Tracks:

Tracks defined in the catalog to which the subscriber can publish data back (e.g., QoE metrics, logs).¶

MSF supports content with target latencies ranging from real-time (under 500ms) to standard live (several seconds) to video-on-demand.¶

6.3. CMSF: CMAF-Compliant Streaming Format

CMSF [I-D.ietf-moq-cmsf] extends MSF to support CMAF (Common Media Application Format) packaged media. It is designed for workflows that use ISO-BMFF containers and existing DRM systems.¶

Key differences from base MSF:¶

• Objects contain CMAF Chunks rather than raw codec samples. A CMAF Chunk is the elementary media segment structure used by ISO-BMFF-based streaming (the container format behind DASH and HLS fMP4), consisting of metadata and media data boxes.¶

• Initialization segments (CMAF Headers) are carried in the catalog as base64-encoded inline data.¶

• Content protection uses Common Encryption (CENC) with commercial DRM systems (Widevine, PlayReady, FairPlay) rather than the application-layer encryption defined by MoQ Secure Objects.¶

• SAP-type event timelines signal stream access points for seeking.¶

CMSF targets deployments that need interoperability with existing DASH/HLS ecosystems and hardware-based content decryption modules.¶

6.4. Layering

The relationship between the components is:¶

 +------------------+
 | Application      |
 +------------------+
 | Streaming Format |  <-- MSF or CMSF
 | (catalog, rules) |
 +------------------+
 | Media Container  |  <-- LOC or CMAF
 +------------------+
 | MOQT             |  <-- transport protocol
 +------------------+
 | QUIC / WebTrans  |
 +------------------+
 | UDP / IP         |
 +------------------+

The streaming format layer defines the media semantics. The container layer defines how individual samples are serialised. MOQT provides the delivery machinery. Applications select the appropriate combination for their needs.¶

7.1. Hop-by-Hop Security

MOQT runs over QUIC or WebTransport, both of which use TLS 1.3 for the underlying connection. This provides:¶

• Confidentiality and integrity of all data on each hop.¶

• Server authentication (and optionally mutual authentication).¶

• Protection against on-path attackers between adjacent nodes.¶

However, hop-by-hop security does not protect content from relays. A relay terminates the TLS session and has access to all MOQT message contents, including object payloads, unless additional end-to-end protection is applied.¶

7.2. End-to-End Object Encryption

MoQ Secure Objects [I-D.ietf-moq-secure-objects] provides end-to-end authenticated encryption of media object payloads. It is based on the SFrame mechanism ([RFC9605]).¶

The scheme:¶

• Encrypts the object payload and optional encrypted properties.¶

• Authenticates the group ID, object ID, Full Track Name, and immutable properties.¶

• Derives per-track keys from a shared base key using HKDF (a standard key derivation function that deterministically produces cryptographic keys from input material).¶

• Forms a unique nonce (a value used exactly once) for each object from the group ID and object ID, ensuring that no two objects are encrypted with the same parameters.¶

Relays can still route and cache objects based on unencrypted metadata (track names, group IDs, priorities, object properties) but cannot access the media content. The architecture separates metadata required for routing/caching from metadata that can be authenticated or encrypted end-to-end. The exact protected elements depend on the object security scheme and application profile.¶

Key distribution is out of scope for the MoQ specifications — applications use external key management systems (e.g., MLS-based key distribution for conferencing [RFC9750],[RFC9420], or out-of-band provisioning for streaming).¶

7.3. Authorization

In general, MOQT defines where authorization material can be carried in protocol exchanges, while the semantic interpretation and verification procedure depend on the selected authorization scheme and the relay or publisher’s local policy.¶

MOQT provides a generic AUTHORIZATION TOKEN parameter that can be included in control messages (SUBSCRIBE, FETCH, PUBLISH, PUBLISH_NAMESPACE, and others). Both publishers and subscribers use this single mechanism to present credentials; relays verify the token before forwarding content or establishing upstream subscriptions. Tokens can be registered with session-scoped aliases to avoid retransmitting large values on every message.¶

The working group has defined two authorization schemes that operate over this mechanism:¶

• Privacy Pass Authentication [I-D.ietf-moq-privacy-pass-auth] — provides privacy-preserving, unlinkable authorization using Privacy Pass tokens. It supports fine-grained access control through namespace and track name matching rules, and defines mechanisms for continuous re-authorization over long-lived sessions — either by obtaining multiple tokens in advance (batched issuance) or by exchanging a validated token for fresh ones directly with the relay (reverse issuance flow).¶

• Common Access Tokens (C4M) [I-D.ietf-moq-c4m] — provides a bearer token scheme using structured tokens with claims that express authorization scope. C4M follows a model familiar from HTTP-based systems.¶

The choice between schemes depends on application requirements: Privacy Pass prioritises subscriber anonymity, while C4M offers simpler integration for deployments with existing identity infrastructure.¶

The streaming format catalog can signal authorization requirements per-track (via the authInfo field in MSF), allowing subscribers to discover what credentials they need before attempting to subscribe.¶

7.4. Privacy Considerations

MoQ's relay-based architecture means that intermediaries necessarily observe certain metadata in the course of routing and caching content. Key privacy considerations include:¶

• Relay visibility of metadata: Track namespaces, track names, group IDs, and object properties are visible to every relay in the delivery path. This metadata can reveal information about content and participants even when payloads are end-to-end encrypted.¶

• Traffic analysis: Object sizes, timing patterns, and subscription behaviour can allow relays or on-path observers to infer what content is being consumed or who is communicating.¶

• Subscriber unlinkability: Without privacy-preserving authorization (such as Privacy Pass), relays can correlate a subscriber's identity across multiple sessions or subscriptions.¶

MOQT provides mechanisms that applications can use to mitigate these risks, including padding (to obscure traffic patterns) and token schemes that support unlinkable authentication. However, achieving strong privacy requires deliberate application design — for example, using generic namespace structures and consistent object sizes.¶

8.1. Application Design Choices

Defining a MoQ application requires decisions in each of the following areas:¶

Transport version:

Which MOQT version the application requires, as identified by the protocol's ALPN string (the identifier used during TLS connection setup to select the application protocol). This determines the available protocol features.¶

Streaming format and catalog:

How media or data is packaged into MOQT objects, and how available tracks are described to subscribers. An application may adopt an existing format (MSF, CMSF) or define a new one.¶

Container and codecs:

What container format wraps the encoded samples (e.g., LOC, CMAF) and which codecs are supported.¶

Security:

Whether end-to-end encryption is required (and which scheme), and which authorization scheme(s) are used to control access.¶

Relay deployment model:

How relays are deployed and operated — tree, mesh, or hybrid — and what trust and authorization policies apply between nodes.¶

Extensions:

Any mandatory MOQT extensions (Properties, Setup Options) that endpoints must support. Applications that need relay-visible metadata beyond what MOQT defines can register new Properties or use the application-reserved type ranges.¶

If existing components do not cover an application's needs, the MoQ framework supports extension without modifying the base protocol: new streaming formats can define their own catalog structures and packaging rules, new Properties can be registered via IANA, and new event timeline types can be defined for synchronised metadata.¶

8.2. Existing Application Definitions

The MoQ working group has defined two complete application specifications that illustrate the design choices above:¶

MSF (MOQT Streaming Format):

Uses MOQT + LOC packaging + a JSON catalog. Targets real-time and live streaming with LOC's minimal per-sample overhead. Supports ABR switching via time-aligned alternate groups, media timelines, event timelines, and publish tracks for subscriber-to-platform communication. Defined in [I-D.ietf-moq-msf].¶

CMSF (CMAF-compliant Streaming Format):

Extends MSF with CMAF/ISO-BMFF packaging and Common Encryption (CENC) for DRM integration. Targets deployments that interoperate with existing DASH/HLS ecosystems and hardware content decryption. Defined in [I-D.ietf-moq-cmsf].¶

Both specifications make concrete choices for every area listed in Section 8.1, producing fully interoperable application definitions.¶

8.3. Reading Guide

The following table maps common goals to the relevant specifications:¶