[littlestymaar]

In a 1 to 1 setting, it obviously doesn't work. But in a mesh setting with n connected peers, when you want to add a new peer, you can perform the first signaling step via a server to connect with one member of the mesh, and then do all the subsequent signaling through this peer. This way, to set-up a mesh with n users you only need n-1 signaling messages processes by your server (down from n(n-1)/2 if you perform everything through a centralized signaling server).

In practice, I know nobody who does that though. I don't think the extra complexity is worth it if your mesh are never really big, and with WebRTC, you rarely encounter situations where you have more than a few peers in the same mesh (Google Chrome even used to struggle a lot if you had more than a few dataChannel opened on the same page, while Firefox handled hundreds of connexions without issue).

[SahAssar]

Ah, yeah, I was thinking of the initial connection.

I guess the mesh needs to be pretty small so that you can have a fully connected mesh or you'd need some way to deal with netsplits and a gossip-style protocol for discovery of new nodes, right?