Mixed sync +/- 100ms of latency or whatever is probably fine, for most purposes? If you really want everyone to be centralised on a single SFU then you’d set the permissions on the room appropriately.
It can be distributed SFU just the relative timings need to converge deterministically (possibly by servers agreeing on a virtual apex to artificially delay the streams to make up for topological differences) to make everyone hear the same thing.
So if it turns out that the streams need to be synced then we can delay them to do so - but i am a bit unsure on how important that really is. Are you worried about folks talking over each other if they think they successfully interrupted but then a remote user thought the same and they collide and backoff, like csma/cd?
> Are you worried about folks talking over each other if they think they successfully interrupted but then a remote user thought the same and they collide and backoff, like csma/cd?
Yes, that is really important for classroom use cases.
No it needs to be deterministic. "Just picking the lowest latency node" would result in sync failures i.e. not everybody hearing the same thing.