NC Talk performance just for chat . Does it need HPB, STUN or TURN servers?

Using NC 26.0.3 php_fpm in an instance that can run a few hundreds of simultaneous users.

As our organization already provides a videocall solution, we鈥檙e only interested on extending our NC instance features with the chat capabilities of Talk.

We鈥檝e enabled a test environment just for chat (disabled video and calls) with the DB trick explained here Disable Calling Functionality 路 Issue #1282 路 nextcloud/spreed 路 GitHub and seems to work fine, but I鈥檓 still worried about moving to production because of the performance issues that can raise as documentation mention on system requiremetns section (Server system requirements - Nextcloud Talk API documentation) 鈫 鈥淣extcloud Talk will significantly increase load on your server, depending on the number of concurrent active participants - no matter whether they use audio call, video call or just the chat.

So my question is, is there any advantage on configuring HPB, STUN or TURN server? As far as I鈥檓 reading it seems that only helps for video and calls. Is there any other setting that I should tune to avoid any performance issues like php-fpm values? Server tuning 鈥 Nextcloud latest Administration Manual latest documentation

I鈥檓 familiar with a small part of the chat API, but AFAIK the chat protocol doesn鈥檛 seem to flow via something like TURN; I think it鈥檚 just HTTP rather than a streaming type of protocol.

HPB, STUN and TURN do not improve performance for pure-chat usage in terms of server resources. In fact STUN and TURN would only be needed for video calling.
The HPB (or the signaling server to be precise) would allow you to use something like typing indicators (not quite sure right now if they would work on the web with internal signaling, for mobile clients external signaling would be required for that feature). That鈥檚 the only really advantage I would see right now using a external signaling server when being chat-only.