I’ve just setup the “high performance backend” (HBP) using the AIO Docker image on Nextcloud Hub 10 (31.0.6)
Today I did a test video call, adding one user at a time and watching top.
Here’s my (crude) results:
| Users | Average load (30s) | eturnal CPU | janus CPU |
|---|---|---|---|
| 1 | 0.07 | ||
| 2 | 0.22 | ||
| 3 | 0.33 | 25% | |
| 4 | 0.25?? | 50% | 12% |
| 5 | 0.78 | 73% | 17% |
| 6 | 1.5 | 117% | 21% |
I know without HPB the resource use locally, client-side scales something like r = c×(users - 1)² which is what makes it so inappropriate for more than 3 people.
But what is the expected resource use with HPB, on the server?
janus - I’m not sure what that is, but it seems to increase use linearly-ish.
eturnal - I’m guessing this is the TURN server. This went up a lot at the end.
The good news was that the call was experienced to be working well with the 6 participants. 
But I’d be interested if anyone has more info on how to expect resource use to scale.