I only ever used audio-only sessions. So far the entire feature is in beta so we had deployment running since December with development and beta-testing going on. Now it seems OpenVidu server is using around 32 gigabytes of physical memory and 64G of virtual (allocated). It’s around 25% of the server’s available RAM.
Kurento media server is actually way better. It only takes 711MB of physical RAM but also around 34G of virtual.
We also had a test session today with 12 users using the voice-chat. Kurento’s CPU usage was going all the way to around 1.2 core (120%). The server has in total of 32 cores (64 hyperthreaded) so there’s still space to grow … but that said for audio-only it’s much more CPU usage than what I expected.
Here OpenVidu load testing: a systematic study of OpenVidu platform performance | by OpenVidu | Medium they conducted VIDEO + AUDIO testing and were able to get to around 182 people / browsers on a 16 core machine.
In my case everybody is both a subscriber and a publisher - but shouldn’t audio-only help?
Also is it possible for the server to merge all audio streams into one?
Is OpenVidu server known to leak memory?