[jitsi-dev] jvb benchmark with hammer - cannot reproduce jitsi team setup


I am trying to benchmark jvb (running on Kubernetes). I am using hammer to simulate the users. I've checked the jitsi team results - https://jitsi.org/Projects/JitsiVideobridgePerformance . However I cannot reproduce the scenario. I am measuring network_in and network_out on both hammer and videobridge - and the bitrates do not match to what is presented by jitsi team.
The measurements I obtained are somehow strange - network_out = 2*network_in (on both jvb and hammer, any number of users). Also the values on jvb and hammer are roughly the same (however looking at the figure if jvb_network_in=N then --> hammer_network_in= N(N-1)). So I have the following questions:

1) How does Last-N and simulcasting work with hammer? Enabled/disabled/supported/not-supported?

2) What configuration (which parameter, which configuration files) should I use to get to the same setting as in picture at https://jitsi.org/Projects/JitsiVideobridgePerformance

3) Is the statement jvb_network_in=N --> hammer_network_in=N(N-1) correct (if last-N disabled) ?

4) What about the video resolution/video quality/audio quality? Is this fixed by hammer or is it adjusted to network conditions?

5) How about streaming from browsers - is video resolution/video quality/audio quality dynamically adjusted to network conditions (WebRTC congestion control) or not?

6) I am using the hammer-provided video/audio resources - https://github.com/jitsi/jitsi-hammer/tree/master/resources . Is description of these (resolution, fps, size, encoding etc.) of these resources available somewhere?



Jernej Trnkoczy
Faculty of Civil and Geodetic Engineering
University of Ljubljana
Phone: +386 1 4768 588
e-mail: jernej.trnkoczy@fgg.uni-lj.si<mailto:jernej.trnkoczy@fgg.uni-lj.si>