Disconnections with more than 15 people

In performance tests with torture, I found that more than 15 participants lead to disconnections.
The resources are not even used to the full.
The jvb uses 20% of the CPU, the input traffic is 1669.12 KiB and the output traffic is 3491.84 KiB.
Is this the normal case, if not what could be the reason?
Are there opportunities to hold larger video conferences?

How are you running the chromes from torture, on the same machine?

No, I have an AWS ECS cluster, the nodes run on Fargate.

Jitsi runs with docker-compose on a Hetzner server.
This is a German provider.

8 vCPU
16GB RAM

20% CPU, I suppose is overall CPU usage, do you have insight of only the prosody CPU usage, as it is single threaded and will be uysing one core and even hitting 100% may not have a visible reflection on overall cpu usage.

The 20% only refer to the JVB, the other components were so low that I had left them out for now.

Here is a complete example for 20 people, there are also disconnections here.

What do you mean by disconnections? Is it the video going off or clients are disconnected and reloaded?

Clients are disconnected and reloaded.
But the images from the clients do not run smoothly either.

This may mean bandwidth problems.

1 Like

In order to have more bandwidth, I booked a dedicated server.
The px92 with a bandwidth of 11.24 Tbit.


But my performance tests failed again with more than 15 customers. An example of 20 customers (17:30 - 17:41):
After 30 seconds there are a lot of disconnections and sometime later all clients come back.


Do you see some http error responses in nginx logs?

I just did a second longer test over 30 minutes with 40 customers.
It is interesting that the connection only breaks at the beginning, from 9:40 am I had a really stable connection.
With 60 clients, however, the entire time the connection is bad.
40 Clients:

60 Clients:

There are also warnings and errors.
Web:
2020/12/08 09:35:57 [error] 513#513: *23036 upstream timed out (110: Connection timed out) while reading response header from upstream, client: 18.158.187.7, server: _, request: β€œPOST /http-bind?room=test0 HTTP/2.0”, upstream: β€œhttp://172.21.0.3:5280/http-bind?room=test0”, host: β€œXXX.de”, referrer: β€œhttps://XXX.de/test0”

JVB:
EVERE: EndpointMessageTransport still not connected.
Dec 08, 2020 10:00:14 AM org.jitsi.utils.logging2.LoggerImpl log
java.lang.Exception: Performing a health check took too long: PT3.154S
Dec 08, 2020 10:01:27 AM org.jitsi.utils.logging2.LoggerImpl log
WARNING: 926520089 large jump in sequence numbers detected (highest received was 8748, current is 8866, jump of 117) , not requesting retransmissions
Dec 08, 2020 10:01:39 AM org.jitsi.utils.logging2.LoggerImpl log
WARNING: RTT suspiciously high (1957ms), capping to 1000ms.
Dec 08, 2020 10:02:01 AM org.jitsi.utils.logging2.LoggerImpl log

jicofo:
Jicofo 2020-12-08 09:59:43.225 WARNING: [38] org.jitsi.jicofo.JitsiMeetConferenceImpl.log() No jingle session yet for test0@muc.meet.jitsi/0976cbcb

prosody:

When I was kicked out I got the following error message.
2020-12-08-Browser-Error