I have managed to set up two jibri instances for my Jitsi server on two separate ubuntu machines: One with a i7-3770 CPU and 32 GB of Ram, and another on the machine specified in the header. Thanks to BBaldino and Emrah for their pointers on this.
On the Opteron machine, I have observed performance while recording and see only a few CPUs running on loads of between 30 and 60 per cent. Single thread performance on Opteron CPUs is just over half of what an i7-3770 delivers. I have read that such Opteron processors are not fast enough to manage ffpeg performance on jibri of the common EC2 instances everybody refers to where 4 CPUs would do the job.
So my question is: how can I set up Docker instances adjusting jibris to use 8 instead of 4 CPU threads and allocate an appropriate portion of RAM to each instance to avoid crashes during recordings?
I would like to limit the server to some 8 instances and use the resources as best as possible. I would set up just one user per jibri instance on jitsi prosody and aim to make available usernames and passwords to chairs for their online meetings to record. When they are done, I want to change passwords to stop unwanted recordings that might run down my hard disk space. After transferring recordings to the learning environment and cleaning up the recording directory, passwords would be unavailable to users until I redefine them and communicate them to students.
I have no experience setting up Docker and am still struggling with basic things. I have checked the manual on github but would want more details on network settings, creating the snd-aloop stuff, which I tried and failed. I will try again in July but would want to read up on to know what I am doing before I mess things up and get frustrated trying and failing time and again.
Can anyone point me towards instructions on how to achieve this? Any thoughts and pointers would be very welcome.