Audio/Video Transport Core Maintenance (avtcore) Working Group

CHAIRS: Jonathan Lennox
Bernard Aboba

IETF 119 Agenda
Location: Bribane, AU
Session: I
Room: P1

Date: March 19, 2024
Time: 09:30 - 11:30 Brisbane time

IETF 119 info: https://www.ietf.org/how/meetings/119
Meeting link: https://meetecho-sin.ietf.org/client/?session=31925
Notes: https://notes.ietf.org/notes-ietf-119-avtcore
Slides:
https://docs.google.com/presentation/d/1T7xa3v-jor0UblFX-83rZIHbhMvWeArJQORsam9AtKs/

Notetakers: Magnus Westerlund, Spencer Dawkins, Mo Zanaty


  1. Preliminaries (Chairs, 15 min)
    Note Well, Note Takers, Agenda Bashing, Draft status, IANA
    registries

    Note taker: Magnus Westerlund

    Spencer gave his usual pitch for people to join the Notes page and
    do corrections, and Jonathan encouraged people to look at the notes
    and make sure that what YOU said was recorded correctly.

    Jonathan said that the chairs have created an AVTCORE GitHub
    organization.

    Murray asked if AVTCORE expects interactions from RTCWEB WG as they
    want to shutdown.
    Bernard noted an issue with JSEPbis found in the HEVC Profile
    draft. This is being discussed.

    Chairs mentioned the RTP Payload format IANA registry and the fact
    that we haven't been able to find an RFC that actually defines the
    registry. Magnus is developing a draft to close the registry and
    update RFC 8088 to not require new RTP payload format to add to the
    registry. Zahed asked if it was necessary to publish this document
    at all. But to update RFC 8088 it would be needed. The chairs also
    want to ensure that we have consensus for the change.

  2. Galois Counter Mode with Secure Short Tags (GCM-SST) (J.P. Mattsson,
    15 min)
    https://datatracker.ietf.org/doc/html/draft-mattsson-cfrg-aes-gcm-ss

    J.P. went through the history of GCM-SST, documented in
    draft-mattsson-cfrg-aes-gcm-ss Section 1.
    During standarization of AES-GCM, Ferguson pointed out two
    weaknesses in the GSM authentication function.
    Rogaway was critical of AES-GCM with short tags and recommended
    disallowing tags shorter than 96 bits.
    While Nyberg pointed out workarounds, NIST instead specified
    additional requirements for use in Appendix C.
    Since then, Appendix C has been criticized and NIST has announced it
    will remove it.

    If the Nyberg changes are adopted, is AES-GCM with short tags
    sufficiently secure and performant?

    Chairs of CFRG will confer about potential adoption. In CFRG, there
    were comments assuming that it would
    only be used for video, not audio. If CFRG review is positive and
    adoption proceeds, is there interest?

    Harald Alvestrand: Google Meet is interested in 32-bit tags.

    Mo Zanaty: interested. But what is the hardware implementation
    impact?
    J.P.: This depends on how it is implemented. If there are hardware
    accelerated functions, they may be used.

    Jonathan: there seems to be interest in this work. The WG will wait
    for CFRG to progress it before adoption.
    It is good to make the interest clear to CFRG to show that it is
    meaningful for them to define AES-GCM short tags.

  3. RTP Payload Format for Visual Volumetric Video-based Coding (V3C)
    (L. Ilola, 10 min)
    https://datatracker.ietf.org/doc/html/draft-ietf-avtcore-rtp-v3c

    The WG was updated on how a number of issues have been handled.
    There are still some open issues.

    Jonathan: V3C specifies rarely implemented txmodes (MRST, MRMT). If
    only using SRST, things become simpler. Is it possible to leverage
    EVC SDP simplifications? Is anyone using DON?

Stephan Wenger: DON has been seen in the wild, but don't include it
unless there is a real use case.

  1. RTP over QUIC (M. Engelbart, J. Ott, S. Dawkins, 20 min)
    https://datatracker.ietf.org/doc/html/draft-ietf-avtcore-rtp-over-quic

    Mathis Engelbart presented updates and open issues.
    Changed doc to experimental status and added Directions for Future
    work
    Removed multiplexing of protocols other than RTP/RTCP.
    Added discussion on coalescing RTP packets.
    Future extensions can build on top of RoQ to multiplex other
    protocols.
    Example: data channel draft (draft-engelbart-quic-data-channels)
    Uses "qdc" as the ALPN

    Bernard: Why run the WebRTC data channel over QUIC? The document
    requires a separate ALPN. Doesn't
    that mean that data channels cannot share congestion control state
    with real-time media?
    Also the data channel API is poorly designed, since it delivers
    messages via events (within a browser).
    This means that overall receive buffers (QUIC + event queue) can
    grow without exerting back pressure.

    Peter Thatcher: Agree that using the current API would be a mistake
    but multiplexing data and RoQ
    (without introducing a distinct ALPN) would make sense.

    Spencer said from his perspective, we have been thinking about
    multiplexing other protocols with RoQ in general, and feedback about
    specific protocols that would be usefully multiplexed with RoQ, like
    the feedback Bernard and Peter are giving in this session, is
    extremely helpful.

    "Are we almost finished with this specification?"

    There are still lots of open questions to answer in experiments. But
    it is time to put down the pen.

    Peter Thatcher: how far can the RoQ and SDP for RoQ specification go
    if there are no solutions for peer to peer QUIC? Will the WG be
    happy with a specification that only supports client to server? What
    if P2P falls through the cracks between AVTCORE and QUIC WG?

    Bernard: RoQ is just a transport mapping. Just because something
    isn't mentioned doesn't mean it is prohibited.
    RFC 3550 doesn't reference ICE.

    Spencer: Confirmed with Jonathan that SDP is not blocking for WG
    last call but likely is blocking PUBREQ.

    Bernard: Why? RFC 3550 doesn't mention SDP. There is an informative
    reference to Spencer's SDP draft, but why should that block
    publication?

    Spencer notes that the authors have considered further developments
    that are ongoing or expected to happen and what potential impact
    that can have.

    Spencer pointed out that the current revision of RoQ has a new
    section 13
    , on Future work, added at Bernard's suggestion during
    the last interim. This describes things we have legitimate issues
    for, but do not have a path for yet (like NAT Traversal), or do not
    have experience with (like connection migration and QUIC multipath).
    Spencer doesn't think it's worth blocking RoQ publication for these
    topics, and they can be described in other specifications while we
    are getting implementation and deployment experience.

    Joerg Ott stated that we can go to WG last call before IETF 120, and
    work on that feedback in a timely way. "This work has already
    dragged out way too long" (Spencer notes that the individual draft
    -00 was submitted in July 2021, which is far enough in the past that
    one might reasonably expect we are finished).

    Magnus stated that we should go to WG last call. Signalling is
    seperate, and RoQ can be extended for future capabilities. Peer to
    Peer is all about establishing the QUIC connection, while RoQ is
    about mapping onto a QUIC connection, so it's fine for them to
    progress independently.

    From Lucas Pardue in Zulip: +1 WGLC soon

  2. HEVC Profile for WebRTC (B. Aboba, 10 min)
    https://datatracker.ietf.org/doc/html/draft-ietf-avtcore-hevc-webrtc

    Safari TP 189 offers HEVC on a send/recv m-line, because it supports
    both encode and decode, packetization and
    depacketization. Chromium or FF does not yet have HEVC recv on by
    default but when they do, question is whether
    browsers can negotiate successfully. Does this require 2 m-lines
    (send-only + recv-only) or will offering a single
    send/recv m-line suffice? Assume we are talking about offering both
    H.264 and H.265.

    Randall Jesup feels it is ok to offer both codecs on a send/recv
    m-line even if one is really recv-only. This does not violate RFC
    3264. However, a problem will arise if the Answerer removes other
    non-H.265 codecs so it can only receive H.265 which the Offerer
    cannot send. Will that happen?

    Harald says send/receive m-lines are a problem, but we must deal
    with them.

    Bernard: please provide any additional comments in the GitHub
    threads.

  3. RTP Payload Format for SFrame (P. Thatcher, 10 min)
    https://datatracker.ietf.org/doc/html/draft-ietf-avtcore-rtp-sframe

    Jonathan Lennox noted that start and end indicators in some header
    extensions, like AV1 Dependency Descriptor or Frame Marking, may not
    tolerate blind encapsulation.

    Zahed Sarker (as individual) wants to see implementation
    exerperience before progressing this.
    This is a security draft, after all.

  4. RTP Payload for Haptics (H. Yang, 10 min)
    https://datatracker.ietf.org/doc/html/draft-hsyang-avtcore-rtp-haptics

    Jonathan Lennox indicated that the chairs will proceed with calling
    for adoption.

  5. RTP Payload Format for Advanced Professional Video (Y. Lim, 10 min)

    https://datatracker.ietf.org/doc/html/draft-lim-apv

    Stephan Wenger noted CELLAR WG does lossless codecs which may align
    with this.

    David Tauban asked if hardware implementation is a goal. It is.

    Mo Zanaty noted that the IETF NETVC WG did not succeed, and this has
    raised questions about
    whether IETF is a viable SDO for development of royalty free codecs.
    Most new work in this area
    seems to be occurring in AoMedia which has developed a framework for
    this kind of work.

    Stephan Wenger would prefer a 30 year old codec to a new one, to be
    confident of low IPR risk.

    Cullen Jennings said an Informational RFC that documents an existing
    design rather than creates a new design would be a viable path.

    Zahed Sarker asks if there are people interested to participate in
    this work. If so, then a WG makes sense. If no interest, then
    Independent Submission makes sense.

    Murray Kucherawy noted that he's handling Cellar over to Zahed on
    Wednesday of this week.

    Spencer noted that the model of "documenting what's out there now,
    and then working on specifications for improvements" is the way
    Cellar was chartered - we produced FFV1 Video Coding Format
    Versions 0, 1, and 3,
    RFC 9043, and are now working on FFV1
    version 4.

  6. RTP Payload Format for sub-codestream J2K streaming (P. Lemieux, 10
    min)
    https://datatracker.ietf.org/doc/html/draft-ietf-avtcore-rtp-j2k-scl

    Some progress on hardware implementation.

  7. Wrapup and Next Steps (Chairs, 10 min)

Jonathan Lennox asked for doc shephards. We will not pub-req Sframe and
RTP over QUIC.

Spencer Dawkins offered to help shephard.

Altanai Bisht asked for shephard responsibilities, and may consider
volunteering. Cullen said he would be happy to help Altanai shepherd the
first document.

Jonathan Lennox noted we will have adoption calls for Haptics and Region
of Interest, and WGLC on Sframe and ROQ.