pthatcherg: this is super cool; on the Matrix side we hadn't spotted that you'd written your own SFU. Would you have any objection to us trying to use it as a decentralised Matrix SFU, as per the design in https://github.com/matrix-org/matrix-doc/blob/matthew/group-...? (Obviously we'd do so within the constraints of AGPLv3).
cool! i suppose even with limited metadata with e2ee, you could probably do a quick study to get a sense for how many calls could switch (how many are 2 party, how many aren't behind nats, etc) and maybe even get a sense for how much could be saved in operation costs.
interesting privacy considerations now that i think of it though. p2p would trivially leak the existence of a call between two devices, but i suppose that given the right taps it would be possible to correlate through server streams based on timing data anyhow.
can imagine some sort of mode for just a few participants on mobile networks only where all the high bandwidth stuff remains fully peer to peer.