IETF 116 Yokohama - MOPS Session

Monday, March 27, 2023. 3:30pm JST

Chaired by: Leslie Daigle and (Technical Advisor) Glenn Deen (with
WG co-chair Kyle Rose remote)

Note taker: Ali Begen, even without the bribe of chocolate.

Note well.

Media Operations Use Case for an Augmented Reality Application on Edge Computing Infrastructure

Presented by Renan Krishna

• Section 5.1 presents the changes for the tput values used by various
  XR applications.
• XR traffic has long-range (term?) dependency, self similarity -->
  long bursts of traffic in multi-client environments
• Section 5.2 presents the updated XR performance metrics in terms of
  latency. Network requirements proportionally increase with the
  number of users.

Authors are asking for WGLC. The chair is asking for comments from the
room.

• Cullen J. says this is a moving target and will never be in a
  perfect shape, having said that he is supportive of a last call.
  Chris Lemmons seconds that.

TreeDN

Presented by Lenny Giuliano

• Problem statement: Scale of live streaming not same as on-demand
  streaming. Need for low latency. Join rates (concurrency?) vary
  vastly.
• Multicast is successful in many places but not so much on the open
  internet, it is an "all or nothing" problem, perceived to be
  difficult but network replication methods are here to saves us from
  these troubles.
• TreeDN: Uses native and overlay concepts to deal with parts of the
  network that do not support multicast. Native: SSM and Overlay: AMT
  (RFC 7450).
• CDNs with multicast --> TreeDN. Better network utilization, may
  offer Replication-as-a-Service (RaaS). If AMT is already supported,
  lower barrier to deploy. Supposedly democratizes and decentralizes
  content sourcing (as opposed to a few companies controlling the
  entire content distribution)
• TreeDNs can be used for live streaming or large software
  distributions like OS updates.
• Next steps: adding diagrams, gap analysis (what else is needed for
  proper product deployment), scoping the work (current target vs.
  future work).

Comments/questions:

• Alan Frindell is asking about TLS, how is TreeDN supposed to support
  encryption/privacy? It is not Lenny G.'s expertise area but there
  are companies (?) that implemented/showed support for this. Alan F.
  suggests adding some text into the draft to this effect. Reading:
  https://github.com/squarooticus/draft-krose-multicast-security
• Alan F. is asking about adaptive bitrates (adaptive streaming). How
  does it fit into multicast distribution as it is not client-driven
  anymore?
• Cullen Jennings: Current CDNs use authorization for billing for
  distribution and storage. Multicast QUIC should be revived?
• MOPS would be a good place to discuss most of the CDN,
  storage/distribution, billing related matters as MOQ is trying to
  create a new protocol for media delivery.
• Jordi Cenzano: Multicast uses UDP as opposed to TCP. FEC can be used
  for loss recovery. Live streaming at large scales might benefit from
  non-unicast loss-repair methods.
• The author is asking the WG to read and comment on the draft. No
  prior knowledge of multicast is needed.
• Leslie D. suggested and the author agreed -- the scope of the
  document relevant to this WG is to focus on capturing the specific
  problem space (including realtime streaming), and the overall
  architecture of addressing it.

Updates from SVTA

Presented by (cool) Glenn Deen

• SVTA uses the stuff IETF produces, works with the operators and
  providers. The goal here is cross pollenation.
• SVA was renamed to SVTA, 100+ member companies (Microsoft just
  joined). 10 working/study groups on all aspects of streaming.
• SVTA projects that might be of interest to the IETFers: Open
  caching, multicdn delivery, QUIC video, distributed/request tracing,
  use of DNS in streaming.
• Segments 2023 conference in May (New Orleans, LA) followed up by the
  SVTA members meeting. Also the fall meeting will be in Prague right
  before the IETF 118 (3 days).
• Contact info for the SVTAers.

Video Reqs in AR/VR

Preented by (cooler) Cullen Jennings

• Some futuristic scenarios showing lightfields. Capture device -->
  Cloud (30 Mbps) --> Headsets (6 Mbps). Viewport changes require new
  data from the cloud.
• If 5G low-latency edge compute is utilized, the bandwidth
  requirement goes up and latency drops.
• Viewport updates need to be < 10 ms. Higher pixels/degrees on the
  headset, faster updates. Image lag might cause puking. RTT to cloud
  will be > 10 ms so there are things we cannot achieve yet. User
  experience is very much dependent on the use
  case/application/user/content.
• Texture-mapped polygons: Single bit of noise might create a big
  error in the rendering. It is easier to hide the noise in point
  clouds (but this requires higher bandwidth).
• Light fields: The orange on a green bottle example, shot from
  different angles. There are things light fields may render better
  than polygons or point clouds.
• NeRF: Neural radiance fields.
• Media for object manipulation and hand gestures.
• MOPS should come up with better definitions for interactive and live
  media. There are things to be discussed about 3D non-video media
  types. Some lessons learned from WebEx holograms. Streaming vs.
  WebRTC type of stuff seems to be converging.

Comments/questions:

• Renan K.: Are you using the existing codecs? Cullen J. says they are
  using H.264 and will soon use AV1.

Update from MOQ

Presented by Alan Frindell

• MOQ's goal is to build protocols leveraging QUIC for real-time and
  near-real-time media.
• The WG has been busy with understanding the use cases, terminology,
  requirements, scenarios.
• Quite a bit of experimentation from Bernard Aboba, Ali Begen, Suhas
  N. and Jordi Cenzano.
• MOQ started with three independent but similar proposals, working
  towards a base/common protocol, to be discussed this Thursday.
• Seeking consensus on a number of issues, looking into adoption and
  the WG needs high bandwidth at this stage --> more interims
• Alan F. bribing chocolate for note takers!