While I don’t think this is all unknown or new, let me share our tests with Jitsi(Meet). We have been looking for audio/video chat tools for both schools and business due to current Doomsday scenarios.
This was on jvb1 and some 6-7 years old java since it was the state-of-the-art until very recently.
The fist obvious program to test was Jitsi. I have installed a plain unhacked server on a container (yes, really, and works pretty well), set up monitoring (it was working only with java8), and let it go. Without much bullshitting:
- with about 10 participants
- the server load was almost negligible
- the bandwidth was about 1 Mbps per participant on the server (which wasn’t important since the server sits on some multi-gigs link)
- the clients ran more-or-less without problems.
(Some stats give really weird numbers, for the record. If anyone interested in the Netdata chart plugin, just ping me on matrix as @grin:grin.hu or on #jitsimeet:matrix.org.)
However. After raising participants above the magical 15, the server was still very happy, the bandwith was okay, but the clients started dying. Even with one videostream and 15 non-video partially muted participants they repeatedly lose connection, reconnect, an often the video (the one and only) get grayed out and stays there while the voice goes on (only fo some), and only client reload fixes it. The mobile client sometimes lose video “permanently” since it’s not that easy to restart it as the desktop version. It seems that slow participants make the fast ones disconnect, too.
A fellow looked at the desktop webpage code and mentioned several ugly problems, like buffering pouring insane amounts of data on the other participants after reconnecting and similar issues, but they seem to be hard to fix, though not impossible - but I’m just the messenger here.
So, anyway, it didn’t work well above 15 people, also had various problems on various browsers (like not seeing cam/mic on Safari/Mac, sometimes not seeing cam on Firefoxen).
[As for contrast BigBlueButton works pretty well and stable (still I want to make a test with more than a dozen participants, my install is way too fresh), but it’s very different as far as I see. For larger conferences we probably have to use BBB, unless there are mobile clients which are clunky there.]