Anecdata point: I had a lot of trouble setting up Nextcloud Talk with my own STUN & TURN servers. Even after I'd gotten that going, NCT video calls have been extremely unreliable, with it fully working only maybe 1 in 5 times, or worse. I still don't know why. Could be dependent on the kind of Internet each side is connected to (home network behind ISP router vs. coffee shop vs. corporate network with potential firewall, etc. etc.)

Where are you hosting your turn server and what software are you using?

I'm using coturn for TURN and STUN, and they are on the same VPS that hosts the Nextcloud code, version 13. If you actually want to help me solve the problems, send me a DM on github (username Pistos), and I can give you more of the technical details.

From what I understand of STUN and TURN, there is absolutely no point running STUN an TURN on the same server as Nextcloud - if you do, you could just as well skip running them in the first place. STUN and TURN only help if they are on a dedicated box, as close to a internet root server as possible, ideally in a big datacenter somewhere on the internet backbone. They can't be behind firewalls and layers of routers...

So your performance will already improve if you delete your local STUN and TURN and use the Nextcloud STUN as our free STUN runs in a big data center. Then you have to find a TURN somewhere that you can run also close to the backbone, or try using without...

Yes, running STUN and TURN properly is a pita.

I'm no super expert so check what I'm saying, but this is based on my best understanding of what these do and I had to write our documentation for customers about our Spreed High Performance Backend which, among other things, does STUN and TURN ;-) (Nextcloud marketing dude here)

> no point running STUN an TURN on the same server as Nextcloud

I just prefer to self-host as much as possible.

I reached out on to your IRC. I'll help, if I can.

I don't think Github has DMs.

Yep, you're right. My mistake. But we've been touch by other means.