Jitsi has been having issues with rather large meetings (>30 people for example) and looking at the graphs, it's obvious the eth0 can't handle it, one way to at least increase the threshold is to de-dockerize jitsi because docker containers introduce their own internal network and add a network overhead, that's why using docker is not recommended in high I/O services.
Let's have a node without docker and see how it goes.
I can attest to the fact that Jitsi/meet.wmcloud.org bogs down when you get to 20-25 people, which is a shame. This Jitsi service has been very useful for the movement during the pandemic so we don't rely on commercial services like Zoom, Google, Microsoft, et al. We have regular meetings of certain user groups that are small, but for larger gatherings we are forced to pay for platforms or borrow the logins of others. If there's any way to expedite this it would be great, since it would pave the way for other services like Calla, which is a virtual town platform based on Jitsi.
Maybe docker is not the problem here? {T268393#6686651} I applied some changes that might help. We should test it somehow.
Also, given the https://jitsi.github.io/handbook/docs/devops-guide/devops-guide-quickstart installing it on buster seems complicated (with requiring OpenJDK 8)
I just had enough and did it. It's actually quite easier than it used to be because new versions of jitsi support open-jdk 11 now. I couldn't get prosody support in place so authentication needs a bit of work but I'll do that later.