Content Stitching for Media over QUIC
draft-englishm-moq-content-stitching-00
This document is an Internet-Draft (I-D).
Anyone may submit an I-D to the IETF.
This I-D is not endorsed by the IETF and has no formal standing in the
IETF standards process.
| Document | Type | Active Internet-Draft (individual) | |
|---|---|---|---|
| Author | Mike English | ||
| Last updated | 2026-11-02 | ||
| RFC stream | (None) | ||
| Intended RFC status | (None) | ||
| Formats | |||
| Stream | Stream state | (No stream defined) | |
| Consensus boilerplate | Unknown | ||
| RFC Editor Note | (None) | ||
| IESG | IESG state | I-D Exists | |
| Telechat date | (None) | ||
| Responsible AD | (None) | ||
| Send notices to | (None) |
draft-englishm-moq-content-stitching-00
Media Over QUIC M. English
Internet-Draft Cloudflare
Intended status: Informational 2 November 2026
Expires: 6 May 2027
Content Stitching for Media over QUIC
draft-englishm-moq-content-stitching-00
Abstract
Media over QUIC Transport (MOQT) requires that an Object is always
the same bytes, no matter who asks for it or where it is retrieved
from. That property is what makes MOQT content cacheable by relays.
It also means that when different subscribers need to receive
different content, the straightforward approach is to publish a
separate track for each of them, even when most of what those tracks
carry is the same. This document describes content stitching:
building the tracks that subscribers receive out of Objects taken
from other tracks, so shared Objects can be cached once while each
audience still receives a track of its own. It names the pieces
involved, describes how they fit together, works through some
examples, and draws a line between a stitcher and a relay. It does
not define a format or mechanism for stitching.
About This Document
This note is to be removed before publishing as an RFC.
The latest revision of this draft can be found at
https://englishm.github.io/draft-englishm-moq-content-stitching/
draft-englishm-moq-content-stitching.html. Status information for
this document may be found at https://datatracker.ietf.org/doc/draft-
englishm-moq-content-stitching/.
Discussion of this document takes place on the Media Over QUIC
Working Group mailing list (mailto:moq@ietf.org), which is archived
at https://mailarchive.ietf.org/arch/browse/moq/. Subscribe at
https://www.ietf.org/mailman/listinfo/moq/.
Source for this draft and an issue tracker can be found at
https://github.com/englishm/draft-englishm-moq-content-stitching.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
English Expires 6 May 2027 [Page 1]
Internet-Draft MoQ Content Stitching November 2026
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 6 May 2027.
Copyright Notice
Copyright (c) 2026 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/
license-info) in effect on the date of publication of this document.
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document. Code Components
extracted from this document must include Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3
3. Stitchers and Relays . . . . . . . . . . . . . . . . . . . . 4
4. Composition Mechanics . . . . . . . . . . . . . . . . . . . . 5
4.1. Identity and Numbering . . . . . . . . . . . . . . . . . 5
4.2. Where the Uniqueness Lives . . . . . . . . . . . . . . . 5
4.3. Transitions Between Ingredients . . . . . . . . . . . . . 5
4.4. Object Properties . . . . . . . . . . . . . . . . . . . . 6
4.5. Signals and Lead Time . . . . . . . . . . . . . . . . . . 6
4.6. Placement and Cost . . . . . . . . . . . . . . . . . . . 6
4.7. Creating Composed Tracks . . . . . . . . . . . . . . . . 7
5. Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5.1. Substituting a Span of Groups . . . . . . . . . . . . . . 7
5.2. Responding to a Signal . . . . . . . . . . . . . . . . . 8
5.3. Composing Individual Objects . . . . . . . . . . . . . . 9
6. Security Considerations . . . . . . . . . . . . . . . . . . . 10
7. Privacy Considerations . . . . . . . . . . . . . . . . . . . 11
8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 11
9. References . . . . . . . . . . . . . . . . . . . . . . . . . 11
9.1. Normative References . . . . . . . . . . . . . . . . . . 11
English Expires 6 May 2027 [Page 2]
Internet-Draft MoQ Content Stitching November 2026
9.2. Informative References . . . . . . . . . . . . . . . . . 11
Use of Generative AI . . . . . . . . . . . . . . . . . . . . . . 11
Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 11
1. Introduction
MOQT [MoQT] identifies every Object by its Track Namespace, Track
Name, Group ID, and Object ID, and requires that an Object "must be
an identical sequence of bytes regardless of how or where it is
retrieved" (Section 2.1 of [MoQT]). It also requires that two tracks
in the same scope containing different data have different names
(Section 2.5 of [MoQT]). This is what lets a relay answer any
subscriber from its cache without knowing anything about who that
subscriber is.
There are a number of applications where subscribers receiving what
is nominally the same content need to receive different bytes, for
some or all of the time. Within MOQT's rules, each distinct version
has to be its own track. If each version is published as a complete
track in its own right, caches end up holding every version in full.
With N versions, the cache footprint is roughly N times larger, even
when only a small fraction of the content actually differs between
them. Ideally, cache cost would scale with how much content is
actually unique, not with the number of versions.
HTTP-based adaptive streaming deployments often handle this in the
manifest, pointing different clients at different lists of the same
cacheable resources. MOQT has no manifest in the delivery path, and
unlike HTTP request headers, MOQT request parameters do not change
what a track contains. So the same problem has to be addressed at
the level of tracks and Objects.
This document describes one way to do that, which we call content
stitching. The goal is to describe building blocks: the entities
involved, what each is responsible for, and the properties that
follow from putting them together. It intentionally does not define
how a stitcher is configured, how signals are carried, or a format
for describing compositions. Some of that might be worth
standardizing later, if there turns out to be a real need for
different implementations to interoperate on it.
2. Terminology
This document uses the terms Application, Track, Group, Object, Full
Track Name, Original Publisher, End Subscriber, and Relay as defined
in [MoQT].
Composed track: A track produced by a stitcher from Objects taken
English Expires 6 May 2027 [Page 3]
Internet-Draft MoQ Content Stitching November 2026
from one or more ingredient tracks. A composed track is an
ordinary MOQT track with its own Full Track Name. Subscribers do
not need to know that a track they receive is composed.
Ingredient track: A track that a stitcher reads from in order to
produce composed tracks. An ingredient track might feed many
composed tracks or only one. Ingredient tracks can also be
generated dynamically. In most deployments, ingredient tracks are
not made available to End Subscribers directly.
Recipe: The rule a stitcher follows to build a composed track from
its ingredient tracks. Recipes are defined by the Application. A
recipe might be a fixed pattern known in advance, or it might
respond to signals.
Signal: Information that tells a stitcher to change how a composed
track relates to its ingredient tracks. Signals might be carried
in ingredient tracks (for example, as Object Properties), in a
separate track, or outside of MOQT entirely. This document does
not specify how.
Stitcher: An Application that applies recipes to ingredient tracks
to produce composed tracks.
Audience: The set of End Subscribers that receive a given composed
track. An audience can be large, or it can be a single End
Subscriber.
3. Stitchers and Relays
A stitcher is the Original Publisher of the composed tracks it
produces. From the point of view of each ingredient track, it is an
End Subscriber: it does not forward that track to anyone. [MoQT]
defines a Relay as an entity that is neither the Original Publisher
nor the End Subscriber of the tracks it handles, so a stitcher is not
a Relay. It is an Application.
In practice, stitchers will often be deployed alongside relays,
possibly in the same process, because that is where the shared
ingredient Objects are already cached. That is a reasonable
operational choice, but it is only that. Nothing in this document is
an expectation of MOQT relays, and a relay that does no stitching at
all is a perfectly good relay.
Downstream of the stitcher, a composed track is just a track, and
relays handle it like any other.
English Expires 6 May 2027 [Page 4]
Internet-Draft MoQ Content Stitching November 2026
4. Composition Mechanics
4.1. Identity and Numbering
Each composed track has its own Full Track Name, so MOQT's
immutability rule applies to it exactly as it does to any other
track. As the Original Publisher, the stitcher assigns the composed
track's Group IDs, Subgroup IDs, Object IDs, Publisher Priority, and
Object Properties. None of these need to match the ingredient
Objects the composed Objects were taken from. A stitcher might keep
an ingredient's Group IDs when its ingredients are aligned by Group
ID, or it might number the composed track's Groups independently.
Either way, the composed track's Group IDs are expected to increase
with time, the same as any other track's (Section 2.3.1 of [MoQT]).
Immutability has a consequence for how recipes run. Once a stitcher
has published an Object in a composed track, that Object is fixed. A
recipe can change which ingredient future Objects come from, but it
can never revise what has already been published, so every decision
about a given composed Object has to be made before that Object is
published.
4.2. Where the Uniqueness Lives
It can help to ask where the uniqueness of a composed track actually
lives. In some cases it lives in the Objects: some ingredient tracks
are used by only one audience, and their Objects appear in no other
composed track. In other cases every ingredient track is shared, and
the uniqueness lives entirely in the recipe, in which shared Objects
each audience gets and in what order.
The cache cost of stitching follows from that answer. Unique Objects
cost storage. Unique recipes mostly do not.
4.3. Transitions Between Ingredients
Taking an Object out of one track and putting it in another does not
make it decodable. Some media has dependencies between Objects. A
frame that depends on a reference frame the composed track never
carried is useless, and so is a frame without the initialization data
its decoder needs. Codec configuration and frame rate also have to
line up across the ingredients that a composed track moves between.
English Expires 6 May 2027 [Page 5]
Internet-Draft MoQ Content Stitching November 2026
The simplest approach is for a composed track to move between
ingredient tracks only at Group boundaries, since MOQT Groups are
already meant to be independently useful join points (Section 2.3 of
[MoQT]). This works best when ingredients are prepared so that their
Group boundaries line up with each other. Where Objects within a
Group do not depend on each other, a recipe can work at a finer
granularity and select individual Objects.
Timing is a separate problem. An ingredient carries its own
timeline, which does not necessarily match the composed track's. If
timing information lives outside the payload (in Object Properties,
for example), the stitcher can set it for the composed track as part
of being its Original Publisher. If timing lives inside the payload,
then either the stitcher has to rewrite payloads, which costs
processing and breaks any end-to-end authentication of those
payloads, or the ingredients have to be prepared so that their
timelines already agree.
4.4. Object Properties
Because a composed Object belongs to a different track than the
ingredient Object it came from, the stitcher decides which Object
Properties it carries. Carrying everything over is rarely what is
wanted. Properties that only make sense in the context of an
ingredient generally should not appear in composed tracks, and that
includes signals. In some applications it matters that End
Subscribers cannot tell from the composed track where transitions
happen.
4.5. Signals and Lead Time
When a recipe responds to signals, how much warning a signal gives
matters. A signal that arrives well before the affected Objects need
to be published leaves time for the stitcher to choose ingredients,
subscribe to them, and receive their Objects. A signal that arrives
too late cannot change anything, since the affected composed Objects
may already be published. Recipes that respond to signals generally
need a fallback for that case, such as continuing with the current
ingredient.
4.6. Placement and Cost
Ingredient Objects shared across many composed tracks only need to be
cached once where the stitcher runs. Downstream of the stitcher,
though, each composed track is a separate track, and a relay that
caches a composed track caches all of it. Most of the benefit of
stitching therefore comes from placing stitchers as close to End
Subscribers as is practical. The larger an audience, the more a
English Expires 6 May 2027 [Page 6]
Internet-Draft MoQ Content Stitching November 2026
downstream cache of its composed track is worth. For an audience of
one, it is worth very little.
4.7. Creating Composed Tracks
A stitcher does not have to create every composed track in advance.
When the set of possible audiences is large, it can be more practical
to create a composed track when it is first subscribed to. How a
subscriber learns which composed track to subscribe to, and how that
is authorized, is up to the Application.
5. Examples
These examples are illustrative only. In them, a Full Track Name is
written as its Track Namespace Fields separated by slashes, then a
colon, then the Track Name; for example, example/in:main is the track
named main in the namespace with fields example and in. A Location
is written as {Group ID, Object ID}. This notation is specific to
this document.
5.1. Substituting a Span of Groups
There are two audiences, A and B. The ingredient track example/
in:main is shared by both. The ingredient track example/in:alt-1 is
used only by audience A, and has three Groups prepared to fit in
place of Groups 104 through 106 of main. Every Group in both
ingredients has four Objects.
Audience A receives example/out/7f3a:v and audience B receives
example/out/c21e:v. The stitcher numbers composed Groups starting
from 0.
English Expires 6 May 2027 [Page 7]
Internet-Draft MoQ Content Stitching November 2026
+================+========================+========================+
| Composed Group | example/out/7f3a:v (A) | example/out/c21e:v (B) |
+================+========================+========================+
| 0 | main Group 100 | main Group 100 |
+----------------+------------------------+------------------------+
| 1 | main Group 101 | main Group 101 |
+----------------+------------------------+------------------------+
| 2 | main Group 102 | main Group 102 |
+----------------+------------------------+------------------------+
| 3 | main Group 103 | main Group 103 |
+----------------+------------------------+------------------------+
| 4 | alt-1 Group 0 | main Group 104 |
+----------------+------------------------+------------------------+
| 5 | alt-1 Group 1 | main Group 105 |
+----------------+------------------------+------------------------+
| 6 | alt-1 Group 2 | main Group 106 |
+----------------+------------------------+------------------------+
| 7 | main Group 107 | main Group 107 |
+----------------+------------------------+------------------------+
Table 1: Composed Groups and the ingredient Groups they carry
Within each Group, composed Objects 0 through 3 carry the payloads of
Objects 0 through 3 of the corresponding ingredient Group. So {4, 2}
in A's composed track has the same payload as {0, 2} in alt-1, and
{4, 2} in B's composed track has the same payload as {104, 2} in
main.
The stitcher's cache holds 8 Groups of main and 3 Groups of alt-1, no
matter how many End Subscribers are in each audience. A relay
downstream that caches both composed tracks holds 16 Groups.
B's composed track carries exactly what main carries, under a
different name and numbering. Giving B main directly would avoid
that, but it would expose an ingredient track to End Subscribers, and
it would make visible which End Subscribers are getting something
different.
5.2. Responding to a Signal
Object {204, 0} of example/in:main carries an Application-defined
Object Property indicating that Groups 206 and 207 are to be
replaced. The composed track example/out/9b04:v numbers its Groups
as the main Group ID minus 100. When the signal arrives, the
stitcher has published composed Groups up through 104. Its recipe
selects example/in:alt-9 for this audience, and the stitcher
subscribes to it and receives its Groups 0 and 1.
English Expires 6 May 2027 [Page 8]
Internet-Draft MoQ Content Stitching November 2026
+================+================+======================+
| Composed Group | Taken from | Notes |
+================+================+======================+
| 104 | main Group 204 | signal Property on |
| | | {204, 0} not carried |
+----------------+----------------+----------------------+
| 105 | main Group 205 | |
+----------------+----------------+----------------------+
| 106 | alt-9 Group 0 | |
+----------------+----------------+----------------------+
| 107 | alt-9 Group 1 | |
+----------------+----------------+----------------------+
| 108 | main Group 208 | |
+----------------+----------------+----------------------+
Table 2: Composition in response to a signal
Composed Object {104, 0} does not carry the signal Property, so End
Subscribers cannot see the transition coming.
Suppose instead that Group 0 of alt-9 is not available by the time
composed Group 106 has to be published. A recipe with a fallback
might publish main Group 206 as composed Group 106. Once it has,
composed Group 106 is fixed. If Group 0 of alt-9 arrives a moment
later, it cannot replace what has already been published. Whether it
makes sense to use Group 1 of alt-9 on its own for composed Group 107
is up to the recipe.
5.3. Composing Individual Objects
Where Objects within a Group do not depend on each other, a recipe
can select individual Objects rather than whole Groups. Group 12 of
example/in:shared has Objects 0 through 3. The ingredient track
example/in:u-91f0 is used for a single End Subscriber, and its Object
{3, 0} is to appear after Object 1 of shared Group 12.
English Expires 6 May 2027 [Page 9]
Internet-Draft MoQ Content Stitching November 2026
+===================+================+
| Composed Location | Taken from |
+===================+================+
| {12, 0} | shared {12, 0} |
+-------------------+----------------+
| {12, 1} | shared {12, 1} |
+-------------------+----------------+
| {12, 2} | u-91f0 {3, 0} |
+-------------------+----------------+
| {12, 3} | shared {12, 2} |
+-------------------+----------------+
| {12, 4} | shared {12, 3} |
+-------------------+----------------+
Table 3: Object-level composition
The stitcher assigns the composed Object IDs, so Object {12, 2} of
shared appears in the composed track as {12, 3}. The audience is a
single End Subscriber, so the composed track as a whole is unique to
them, but only one of its Objects is.
6. Security Considerations
If End Subscribers can subscribe to ingredient tracks directly, they
can receive content their recipe would not have given them, or bypass
the composition entirely. Where that matters, deployments need
access control on ingredient tracks, not just on composed tracks.
Signals change what End Subscribers receive. A forged or replayed
signal could cause a stitcher to publish content that was never
intended. Stitchers need some way to trust the signals they act on.
Section 2.1 of [MoQT] makes the Original Publisher solely responsible
for an Object's payload, including any end-to-end encryption or
authentication. A stitcher is the Original Publisher of its composed
tracks, but the payloads it publishes were produced by the publishers
of its ingredients. Where payloads are encrypted or authenticated
end to end, keys and authentication need to cover every ingredient a
composed track may draw from, and End Subscribers need to accept
payloads from all of them. A stitcher that cannot read payloads can
still compose tracks, as long as its recipes only depend on
information it can see.
Composed tracks are state on the stitcher. A stitcher that creates
composed tracks in response to subscriptions lets subscribers cause
it to create state, so it needs to authorize a subscription before
creating anything for it.
English Expires 6 May 2027 [Page 10]
Internet-Draft MoQ Content Stitching November 2026
7. Privacy Considerations
A composed track exists because some set of End Subscribers was
treated differently from others, so its Full Track Name can reveal
something about its audience. If a name encodes audience membership
in a readable way, anything that can see the name, including relays
and logs, learns it too. Opaque composed track names avoid this.
A composed track for a single End Subscriber is linkable to that End
Subscriber. In some applications that linkability is the point. In
others it is a side effect worth minimizing, for example by using
such tracks only when the recipe actually calls for them.
8. IANA Considerations
This document has no IANA actions.
9. References
9.1. Normative References
[MoQT] Nandakumar, S., Vasiliev, V., Swett, I., and A. Frindell,
"Media over QUIC Transport", Work in Progress, Internet-
Draft, draft-ietf-moq-transport-22, 1 October 2026,
<https://datatracker.ietf.org/doc/html/draft-ietf-moq-
transport-22>.
9.2. Informative References
[RFC9775] Perkins, C. S., "IRTF Code of Conduct", RFC 9775,
DOI 10.17487/RFC9775, March 2025,
<https://www.rfc-editor.org/rfc/rfc9775>.
Use of Generative AI
Anthropic's Claude was used to assist with organizing source
materials, structuring the document, and drafting text. All AI-
generated content was reviewed and approved by the author. This
disclosure follows the guidance in [RFC9775], applied voluntarily to
this IETF document.
Author's Address
Mike English
Cloudflare
Email: ietf@englishm.net
English Expires 6 May 2027 [Page 11]