I am running Jitsi docker platform on a single dedicated server, running both JVB, Jicofo and.prosody
I can see CPU increasing quickly with few participants talking to each other.
I would like to reach around 1,000 simultaneous participants having from 2 to 5 participants in each call.
Would you know more or less what bandwidth and capacity the server should have to handle such load?
The estimations are very very rough, as there are many moving parts, but 5 participant conference with chrome and simulcast, 720p max will use 17.5 Mb/s download and upload. So 1000 = 200*5 = 200*17.5= 3500 Mb/s upload and download. Currently adding some firefox browsers in the estimation changes the upload estimation a lot.
4 chrome + 1 firefox is let’s say 17.5 Mb/s download for the server and the upload will be 28 Mb/s. Then upload is 5600 Mb/s.
Thank you for your quick reply!
Some few more questions then if you dont mind!
1/ You are talking about Mbytes right not Mbits?
2/ If P2P is enable, for a 2 participants conference, I can consider there is almost not upload/download traffic, right?
3/ For a 3 and 4 participants conf. how much download/upload should we have?
4/ Is there any documentation somewhere explaining how to reduce bandwidth, CPU being used by playing/changing JVB settings?
5/ In term of CPU for this same load, 1000 conferences, how much CPU would you put ? and maybe how much RAM too?
Thank you for your time
This is Mbits, not MBytes
Not always, there are cases like symmetric nats or double nats then it will fallback to jvb, or you need to configure turn server.
If you consider chrome and simulcast everywhere, the calculation is the same as on 5.
No. Other than lowering the default resolution in config.js there is per user to select quality or audio-only mode.
We use c5.xlarge aws instances for our bridges. You will not be able to handle 1000 sessions in one bridge, first because of bandwidth. Normally the bandwidth is the cap for a bridge.
Thanks a lot! This helps a lot.