Notes: Media Over QUIC
Chair: Mike English
Scribe: Will Law
Mike: should we do it?
- Yuriy, does not solve resource provisioning, agreement on capacity. Use CDNs? to solve today. Solved 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.