Formula to capacity plan Jitsi Network, and Jibri video recording local storage (Defaults)?

While I see general information for Jibri installs, and use of ffmpeg, and changing resolutions in the config.js between 720, 180, 1080, etc. I’m having trouble pinning down what general formula I can use to determine expected file size of recordings.
I need to know: min/max/avg for network traversal costs (Jitsi) and how much capacity to plan for with Jibri recording sessions locally. I know there are many configuration variations available, but is there some kind of default we can start with so I can then extrapolate other resolutions and options from that baseline?
So, for sake of trying to pin this down from so many variables, lets assume a typical scenario of:

1 Jitsi meeting room with 20 users, all sending video (and taking turns talking one at a time). Everyone is on broadband and not having connectivity issues, and running on reasonably powerful systems, so they can run at whatever the quality default settings are typically for a default Jitsi setup. (720p? 1080p?)

This room is recording through Jibri to a hard drive connected to the Jitsi and Jibri single server.
The meeting (and recording) lasts for 10 minutes.

I see numbers all over the place without consistency, so hoping can pin something reasonable down for capacity planning.

Looking for a reasonable ballpark formula I can start with for:

  1. Determining how the network bandwidth is going to be consumed Mbps.
  2. What default resolution, (quality/throughput, etc.) the recorded video will be with default install.
  3. What would the total filesize of the finished 10 minute default settings recording be in MB or GB.
  4. What differences would there be in the numbers between the older H264 versus the newer VP9?

With that information I can extrapolate a lot from there for other variant scenarios.

Appreciate help trying to pin this down to something useful.

Thanks kindly!

The latest version of Jibri records in 1080p resolution by default.

It’s virtually impossible to determine with exactitude what the file size of a recording will be because it depends on several factors, some of which you’ve already mentioned. I’ve seen this topic discussed somewhat (not necessarily with regards to Jibri) and one bit no one seems to talk about, which is actually very important, is the activity that’s being captured. If there’s a lot of people in the frame and a lot of movement in the frame (people, object, clouds, anything…), then ffmpeg will need to do more work and this will in turn impact the demand on the processor and also the file size. You can actually test this on your own: set up a controlled environment, in the first try, sit still and record for say, about 2 mins. Then try again right after and this time, move around as much as possible (while still in the frame) for 2 mins. You’ll see that the file sizes will differ - the 2nd recording will be larger. Same subject, same conditions, different activity level. This difference becomes more obvious the longer the recording.

VP9 files are larger.

1 Like

Yes, I understand this is challenging, and understand at least in general how the algorithms work with the pixel update, adjacent, diffs, etc. to make significant differences in how much data needs to be recorded between a video with very little visual change and something with lots of changes, but in the aggregate I am hoping some people, hopefully here, have performed aggregate data analysis that allows such trending information to be acquired between min, max, and avg.
I do have in my backlog a plan to run these tests myself to get these numbers with canned simulated user videos, but folks were trying to get estimates for storage (and network traversal) from me weeks before I am ready, because they are trying to get the pricing worked out, and so I was hoping to have some basics to work with. If no one has any suggestions, I’ll throw some very wide numbers out until I have time to do the formal analysis next sprint.

Thanks for the feedback. Thanks for that general info about the impact of VP9 related to this topic.

Welcome any additional suggestions that might work toward some numbers.
Thanks kindly to all!