Closed kixelated closed 1 year ago
Those two questions are related. Consumer should influence the decision because:
Those metric/behavior can not be inferred from simple ACKs of transport layer. It would be great that the protocol can provide an application-layer feedback channel from the consumer to the producer.
The current draft has a brief section on ABR. The claim is that the producer should choose renditions because it has better insight into the state of the network.
I don't agree with this claim as I mentioned before. There are a lot of things that do happen on the client side that the sender will never be aware of unless it is explicitly told by the client. Either the client provides all the other relevant details (some were listed by you and @VMatrix1900) to the sender so it can do a better decision or the sender tells the client what the current network conditions are and the client makes the decision and reports back to the sender.
Yeah, I think I should remove this section. It's too much of an opinion rather than something to act upon. I just wanted a soapbox to complain about low-latency client-side ABR.
Closing because the new draft does not specify track management yet. The plan is to support contribution and distribution.
The current draft has a brief section on ABR. The claim is that the producer should choose renditions because it has better insight into the state of the network.
Does the producer have enough information to choose the rendition, and should any additional information required be part of the protocol? For example, the current buffer size.
Additionally, should the consumer influence the decision? For example, the player is muted/backgrounded and we should save bandwidth by choosing from lower renditions.