Server requirements for 20 users with video

Hi all,
Do anyone know where i can find the requirements for server installation ? I want to deploy one instance for 20 users but I don’t know about CPU, disk, RAM …
Thanks in advance


I also looked for this, but didn’t find anything specific. From our very limited,experience of the last weeks: We run a standard install from the provided packages, on Debian 10, no tweaks. Server is a virtual machine (KVM), with 2 dedicated Xeon 6230 cores.

With 6 to 7 users, all on standard home broadband, we see around 75% CPU utilization (100% = one core), and 20 to 30 MBit sustained traffic. I’d assume this machine to be fine for up to 10 or 12 people, but will upgrade to more CPU cores for bigger meetings.

If anyone else has experience with “small” installations like this, I’d be happy to hear if the numbers we see make sense, or if there might be configuration errors somewhere.

1 Like

I’ve talked to other people setting up jitsi servers, and they see similar performance (a current-generation Xeon core serving about 8 people) on their system. I’m not completely sure this scales with the number of users, but for now I would assume so.

From what I’ve gathered, there’s only a few things that can be tweaked:

  • get all users to use Chrome (or derivates), so simulcast works, and the videobridge can drop the high-bandwidth streams whenever necessary
  • tweak the config.js (in debian: /etc/jitsi/meet/[domain]-config.js, “resolution” and especially “constraints” section) for lower video resolution. Standard there is 720, for us 480 is good enough. Besides reducing CPU and bandwidth on server, it will help people on slow connections if others join w/o simulcast enabled (is there a way to only force non-simulcast browsers to the low resolution?).
  • (especially) if you are in a virtualized server environment, make sure hardware encryption support is actually enabled and accessible to the VM:
  • try switching to videobridge2 (in the unstable repository), which might have some performance tweaks. We did the switch w/o problems, but I don’t really have performance comparison numbers.

Thank you very much for your feedback on the installation. I am willing to test local servers install for local meeting.

I ve read that with videobridge it should not comsume more than 50% for 1000 concurrent conection and 500 mbps… my goals are minor in participants so I want to test it.

Hola Palandia. Cuéntame cómo resulto tu prueba, quiero montar este server para un colegio. Mi idea es poner a 32 profesores con 45 alumnos cada uno y no he logrado calcular lo que necesito en hardware para montar la solución final.

Hi all,

I need a hardware recommendation for a school solution for 30 classes with each 25 participants too.
I did some tests with a 4 core VM on an Intel Xeon Gold 6242 (2.8 Ghz) on an Ubuntu standard installation. With 10 participants (video/audio) we had a workload about 22%. Another performance test on a VM with 8 cores (typ b2-30) from the OVH Cloud provider also with 10 participants we had a workload about 28%. The memory consumption on both VMs war under 2 GB.
So I’m confused about my results and additionally the performance report about 1.000 sessions on one single server. With 30 classes and 25 participnts I have 750 sessions, right? So I need only one server too?
I would like to place the server(s) to a European or German cloud provider. During the weekends, in the night and in the school holidays the server can be off.

Has anyone an idea, a solution or another hint?


I’ve found this load-test here

Hey, so I ran this with 3 rooms and about 25 folks in 2 rooms and 10 folks in the final room. It was 2 vCPUs and 8GB of ram on GCP. Everything worked flawlessly. It looks like there is no real relationship between number of users and the overall CPU load. What it seems like to me is that you will want to do an installation where the routing server and the video bridges are on/in separate VMs so that you can scale up as needed. I am building out a custom setup with high fidelity audio and video designed for dance lessons, music lessons, and dance events.

I will keep this updated as I figure stuff out.

I will say this, stay away from any sort of metered bandwidth solution as it will cost you a bundle unless you setup the p2p options.



@KentimusReximus thanks for the advice! What server provider are you using (if I may ask?)

Google Compute Cloud… this is dangerous though because GCP is metered bandwidth.

Are there any cloud compute options which aren’t metered? Every provider I’ve checked offers only metered solutions - e.g. DigitalOcean, Vultr, UpCloud, AWS, etc. :confused:

Sure! Here is one:

All you need to do is search for unmetered VPS and you will probably find a few.

I am looking at an unmetered solution that will backup into GCP incase I need server capacity right away and I am okay with paying some metered bandwidth while I figure out if I need to buy another server at unmetered VPS.

Let me know if you have anymore questions!

1 Like


there are a few Swiss VPS providers, for example, which offer unlimited networking, starting under 10 CHF/month.

I don’t have (yet) experience with these, but here’s one, for example:


Do you remmember how much dedicated bandwich had for support these three rooms?
Can I have your e-mail for contact you?

we are studying to implement a soluction for around 50 people talking through 25 rooms.


Lo que necesitas es configurar un nginx con varios videobirdges

Hi Kentimus

Do you have already some more experience?
I am currently serving a dancing school, and having for ex 15 participants in 1 session.
Above 10 participants I see dropping videos from participants.
The server is like 2CPU 2GB ram and 5GBPS bandwidth where the capacity on any of these parameters isn’t burned yet.

What experiences do u have?

Just a tip: you can always add another machine or two for work-balancing later, if your current setup cannot handle the load. The protocol to do this is called octo and is really not hard to implement.

You also need to setup the following stuff: (found here on the forum somewhere):

"enable octo to use multiple bridges for one conference:

That is possible, You will need to enable octo on your bridges.

in the jicofo/ add



(The SplitBridgeSelectionStrategy is for testing only, it will split each conference immediately)

in config.js add the octo part to the testing key:

testing: {
    octo: {
            probability: 1

and set a region in the deployment info:

deploymentInfo: {
// shard: “shard1”,
// region: “europe”,
// userRegion: “asia”
userRegion: “myRegion”

(this can be more complicated if you have multiple regions)

Finally, in config files of the videobridges:

Octo configuration

org.jitsi.videobridge.octo.BIND_ADDRESS=<< private_ip >>
org.jitsi.videobridge.octo.PUBLIC_ADDRESS=<< public_ip >>

(also, open port 4096/10000 between the videobridges)

Note: I have had this running for test purposes, we haven’t deployed this production yet."

P.S.: Also read these comprehensive tuning tips to get the most out of your machines:


Check startVideoMuted and channelLastN in /etc/jitsi/meet/YOURDOMAIN-config.js

Hello, in my case I am considering using Contabo for my implementation for a school, I already have it working with another project (No Jitsi) and it really does not have a transfer limit, and they offer good connection speed.

I had poor experience today with about 10 participants (some kept dropping out). I am using a shared VPS at Hetzner, with about 1.5 Gbits/s bandwidth, 3 CPUs, 4 GBytes RAM, 40 GBytes SSD. I think the issue was with the CPUs. It is very difficult to tell what’s going on, with people not being sure if their microphone and/or camera is on or off, not appreciating the echo may be because their independent microphone and audio are both on, etc. But, ultimately, my platform has to take the blame. The CPU hit 100% a few times according to Hetzner’s graphs. For my modest needs, even at Hetzner’s very-reasonable prices, I cannot justify the next step up in processing power. Participants were from across US, Ireland, and England.