Hello everybody and sorry to add my question at this time where clearly a large amount of new users flow in!
TL;DR : When I disable P2P mode, and run on a server with a public IP, then no ICE / STUN / TURN / NAT-holepunching process or anything else should be used, right? The ciients only connect to the server, and there are no other connections to double-check?
The long read:
I’m using Jitsi Meet since several years, sometimes on meet.jit.si and sometimes on my own instance. For me, and for friends on fiber or “good” providers, it works quite well (massive cheers to the developers!). I have one problem: dear family member K. is on a rather slow DSL line, with an Internet provider that I have no trust in, and while the experience never was great, it is now quite bad, and downright unusable when several people join in. Most people can hear each other, but there is always somebody that K. cannot see or hear. Or the others can’t hear her. When the audio connection stops, it never comes back. And even if it works, a delay of up to 10 seconds builds up during a 15-minute call.
Apart from the bad provider, I control all bits of the chain. Every client is on Debian 10 with Firefox or Chromium/Chrome. I tried many different set-ups:
- Jitsi Meet in Docker containers, behind a shared nginx front-end;
- Jitsi Meet directly installed on a cloud VM from Debian packages;
- I even once threw clients and server into the same Wireguard network, to avoid any connectivity problems.
I suspect, especially with the recent raise in traffic, that I cannot trust any external service or mechanism, no STUN, no TURN, no NAT hole-punching. To limit debugging, I would like all clients to only communicate with the server, not with each other. Especially, I want that sending bandwidth stays the same, whether there are 2 or 4 participants. By disabling P2P mode in the web config, I should get that, right?
I know that it should technically be possible for K., as Skype (on n:1) and sometimes even Nextcloud Talk (on 1:1) work quite well!
Thank you very much for reading,