I’m trying to improve the documentation and we’re also working on more settings and other optimizations to improve performance of Talk.
In theory, it should be possible to get about 20 participants in a single call without the HPB, but doing this requires a number of things. One limit is that a single video stream is about 1 mbit/sec so you can imagine that with 5 participants, each is already sending 4 streams of 1 mbit/sec and that quickly stops scaling. EDIT: Often, the problem is on the decoding side as well, decoding 5-6 video streams of 640x480 at 30 FPS is a lot of work and overloads systems. We are looking at how we can lower that, with decreasing quality and finding codecs that are commonly hardware-accelerated. Help welcome.
We definitely want to improve that, there are issues open for this but our backlog is sadly pretty big. We make progress quickly, just look at the changelogs from releases, 8.0.6 fixed some bottlenecks and we’re doing more.
However, you can do something yourself: simply not use video. That saves a lot of bandwidth, of course. You suddenly only send 50kbit/sec instead of 1 mbit/sec. If 1 participant has video or screensharing and the rest audio-only, you should be able to have that 15 to 20-ish people in a call. (EDIT: earlier I wrote you can’t disable video during a call, you will have to disable it, leave and join again, but that was actually fixed in a rewrite of the calling code earlier, we just tested it and 8.0.x should allow you to just disable video and lower the bitrate by ~20x)
I know it isn’t ideal and we will have to add the ability to simply lower the bandwidth being used (eg use 200kbit/sec video instead of 1 mbit/sec - we are working on this and might be able to bring this in the 8.0.x series still) and other features during a call but that is simply work we have not yet done and seriously, help is welcome. The talk team has loads of customers pushing the team to implement things those customers need. They pay the bill, you know, hard to ignore them.
In the mean time, obviously Jitsi is a nice solution and yes, if you can set that up, use it… There’s no need to hate on open source solutions for not doing everything you want. Pull up your sleeves and help out if you need them to do something instead.
Anyhow, I’m creating and discussing a PR here to the read.me that gives some tips on scaling up, feedback is welcome:
If you want to help, here a few issues that will make a big difference: