Mo Q Media Over QUIC
Main.MoQMediaOverQUIC History
Hide minor edits - Show changes to output
Changed lines 1-3 from:
Notes: Media Over QUIC
Chair: Mike English
Scribe: Will Law
Chair: Mike English
to:
Notes: Media Over QUIC \\
Chair: Mike English \\
Scribe: Will Law
Chair: Mike English \\
Scribe: Will Law
October 31, 2023, at 05:30 pm
by - formatting
Changed lines 2-4 from:
Chair: Mike
Scribe: Will
Scribe: Will
to:
Chair: Mike English
Scribe: Will Law
Scribe: Will Law
Changed lines 21-22 from:
to:
Daniel - what about SVC? Mike - with MOQ, you can set stream priorities for layers, use that for SVC.
Will - what decisions is MOQ facing?
Changed lines 30-32 from:
Call to action: more implementations.
to:
Will - How long until production usage? Still needs advertising, DRM, captions. Several years.
Call to action: more implementations.
October 31, 2023, at 05:25 pm
by - Will Law's notes from MoQ Session
Added lines 1-30:
Notes: Media Over QUIC
Chair: Mike
Scribe: Will
Mike: should we do it?
* Yuriy, does not solve resource provisioning, agreement on capacity. Use CDNs to solve today. Solved ST1/ST2 (1970s).
* Yuriy, cannot solve contribution side, needs protocol. ABR is messy, should not exist.
* Mike - can’t guarantee resources, that is the nature of the internet.
* Ali - real-time rate adaptation, trade off latency and quality. MoQ gives us more tools in toolset.
* Mike - webrtc does not have any cachability. MoQ is good to get to broadcast scale. MOQT is a transport protocol. Streaming formats sit on top
* Will - one network for real-time, interactive and VOD will be more cost efficient than separate networks.
* Elijah - we have to deal with ABR in real life.
* JP - we have 3 protocols in our “protocol”, reliable, GOP-per-stream, and datagrams, reduces latency. If you care about fairness, need a different solution.
* Ali - how do you decide which to use? JP - does not change during session. What level of CPU overhead? About 5%. On QUIC you can understand jitter, calc acceleration of jitter, change encode bitrate.
* Will - datagrams only solve latency if there is congestion? JP - yes. Relay does cache datagrams.
* Yuriy - if we enable unreliable comms (datagram), then codecs should be error concealing, modern codecs do does not do this. Will anyone vendor write error concealing decoders? Prob not.
* Will, with HEVC, can you use tiling to compartmentalize errors? Yuriy, yes, and there is slicing too.
* JP - for audio, is unreliable datagrams.
* Yuiry - dynamic switching of bitrates. Switching at GOP boundaries is inefficient. Look at switching at other opportunities within the codec.
* Daniel - what about SVC? Mike - with MOQ, you can set stream priorities for layers, use that for SVC.
* Will - what decisions is MOQ facing?
* Yuriy, concealment
* Mike, see clear separation of media mapping and transport objects.
* JP - how do you implement QUIC for CDN.? Scale will be difficult. Elijah - see big uptick in QUIC traffic. Many networks are not ready for it.
* Will, Akamai will prefer WebTransport as public interface.
* What congestion controller is being used? BBRv2 seems common.
* Will - How long until production usage? Still needs advertising, DRM, captions. Several years.
Call to action: more implementations.
Chair: Mike
Scribe: Will
Mike: should we do it?
* Yuriy, does not solve resource provisioning, agreement on capacity. Use CDNs to solve today. Solved ST1/ST2 (1970s).
* Yuriy, cannot solve contribution side, needs protocol. ABR is messy, should not exist.
* Mike - can’t guarantee resources, that is the nature of the internet.
* Ali - real-time rate adaptation, trade off latency and quality. MoQ gives us more tools in toolset.
* Mike - webrtc does not have any cachability. MoQ is good to get to broadcast scale. MOQT is a transport protocol. Streaming formats sit on top
* Will - one network for real-time, interactive and VOD will be more cost efficient than separate networks.
* Elijah - we have to deal with ABR in real life.
* JP - we have 3 protocols in our “protocol”, reliable, GOP-per-stream, and datagrams, reduces latency. If you care about fairness, need a different solution.
* Ali - how do you decide which to use? JP - does not change during session. What level of CPU overhead? About 5%. On QUIC you can understand jitter, calc acceleration of jitter, change encode bitrate.
* Will - datagrams only solve latency if there is congestion? JP - yes. Relay does cache datagrams.
* Yuriy - if we enable unreliable comms (datagram), then codecs should be error concealing, modern codecs do does not do this. Will anyone vendor write error concealing decoders? Prob not.
* Will, with HEVC, can you use tiling to compartmentalize errors? Yuriy, yes, and there is slicing too.
* JP - for audio, is unreliable datagrams.
* Yuiry - dynamic switching of bitrates. Switching at GOP boundaries is inefficient. Look at switching at other opportunities within the codec.
* Daniel - what about SVC? Mike - with MOQ, you can set stream priorities for layers, use that for SVC.
* Will - what decisions is MOQ facing?
* Yuriy, concealment
* Mike, see clear separation of media mapping and transport objects.
* JP - how do you implement QUIC for CDN.? Scale will be difficult. Elijah - see big uptick in QUIC traffic. Many networks are not ready for it.
* Will, Akamai will prefer WebTransport as public interface.
* What congestion controller is being used? BBRv2 seems common.
* Will - How long until production usage? Still needs advertising, DRM, captions. Several years.
Call to action: more implementations.