Recommended Server Specs for 2020?

I’d like to get an update to some of the older recommendations on the forum for server capacity:

  1. Are there general guidelines for server capacity/performance for a certain number of simultaneous users? (October 2018 recommendations)
  2. Are there general guidelines for the bandwidth required for a certain number of simultaneous users? (July 2017 recommendations)

Is there any newer data for a Jitsi server in 2020? Anecdotal reports welcome in this thread. Please use the format: simply select the following text and click “Quote”

I am successfully hosting X simultaneous users on hosting provider using a server with Y cores and Z Gbytes of RAM. Add other qualifying information, like estimates of bandwidth consumption or the point where it gets overloaded…

I will summarize the information in the OP as time goes on. Many thanks.

3 Likes

OK I’ll start…

I am successfully hosting ~5 simultaneous users on https://www.scaleway.com/en/dedibox/ using a server with 2 x 1.8GHz Intel Atom cores and 4 GB of RAM. Haven’t managed to break it yet. Will report back when that happens.

1 Like

I am successfully hosting 3 simultaneous users on a cloud virtual machine at cacloud.com running minimal Debian on 1 core and 0.5 GB of RAM. Not using recording, streaming, or PTSN interface. This was one test, planning to try with more users soon.

I run jitsi on a special discount VPS very similar to brmdamon’s one

1vCore, 0.5GB RAM, no bandwidth guarantee, Debian minimal with deb-setup, nginx

Works fine for up to 5 persons with camera, when more and more joined (up to 10), some persons got kicked. Some users Chrome, some other or Android-app.

1 Like

I’m also running on a 1vCore (Intel® Xeon® Gold 5120 CPU @ 2.20GHz, running on VMware ESXi) server from 1und1 with 0.5 GB RAM.
We experienced some lags yesterday when reaching more than 6 users, allthough cpu utilisation was at 65 % and memory at 75% max. Logs are full of "Logs are full of “Cannot find SSRC for RTX, retransmitting plain. SSRC=3444644245” - but this doesn’t seem to be utilisation related.

1 Like

We are running a docker deployment on Flexible Engine (s3.xlarge.2, 4 vCPUs | 8 GB | CentOS 7.5).

We successfully hosting 28 users (with only one video stream) but we ban firefox, most of use use electron client. Max CPU usage 25%, Max Outband Incoming Rate 300KB/s, Max Outband Outgoing Rate 2MB/s)

1 Like

This is a very useful guage. Visit https://docs.easyjitsi.com/docs/benchmarks

Who is the original author? https://jitsi.org/jitsi-videobridge-performance-evaluation/

The official jitsi team i guess

Yes, this is an interesting study. I am glad you cited it. But I would like to keep this topic focused on summaries of people’s experience with the number of simultaneous sessions on a variety of hosting provider / # CPUs / RAM combinations, as requested in the original post. Thank you.

I am successfully hosting ~23 simultaneous users in one room on Hetzner using a server with 4 physical cores with HT and 64 Gbytes of RAM. RAM usage never came close to maximum as far as I am aware. Participants did not use video during that test. I’ve also run 10 simultaneous users with video in one room on similar hardware. Bandwidth was never a problem on the server. Maximum CPU usage was never above 190% (Linux, maximum would be 800%)

The CPU used was an older but highly clocked Xeon/i5 (if I compare the core clock of my Xeon to older Xeons with 8+ cores)

Jitsi Meet room size seems to be more limited by the client performance, as the many WebRTC streams can be quite taxing to decode.

As we do not track our users, I do not know if there were any other rooms running while the big rooms were active.

I am trialing a cloud SSD server with 20GB disk, 1000GB bandwidth, 1 CPU (core unspecified) and 2Gb RAM. Yesterday it worked acceptably with 4 people. When we got up to 5 the resolution dropped, and with 6 the person with a Chromebook dropped off. We were using Firefox, Chromium, and Linux, Windows and Android devices. Safari on a Mac could not join and the user changed to Windows 10 and Firefox.

After the event, I discovered that CPU usage peaked at 99%.

I am now investigating a dedicated server as below and will report back.
Intel 8 Core C2750 Atom
Single CPU, Entry Level Budget Server
from£20/mo
8 x 2.4Ghz CPU Cores
8GB Memory
500GB (HDD) Hard Drive
10TB Bandwidth

We are running our own instance of jitsi on a dedicated server at OVH
Specs:
Intel Xeon E3-1270. 8 cores 16 threads, running at stable speeds over 4ghz (performance governor enabled in linux)
32gb of RAM
SSD boot disk + 2TB storage
1000d/500u Internet link

The setup is configured to always aim at the 1080p resolution by default.
P2P is disabled by default for reason I won’t detail here. (so with 2 participants, it still runs off the server)

Tested So far
We had few meetings with around 100 participants and the server was not even working… we had a 20% load on one core and all others were staying below 2% with some occasional peaks at 8% on a single core.

We are going to have a real crazy test next week, as we anticipate 2500+ participants at an event… I am crossing fingers that the server will hold (and am trying to devise a real test we could run to make sure!)

1 Like

500 upload is like only 100 participants getting hd streams, very roughly…

Hum… Its time for the F word then! :frowning:

How did that go?

1 Like

Summary so far: From the reports above, here’s a rough summary of the number of simultaneous clients a Jitsi server can support:

  • Garden variety VPS servers or Docker containers: three to maybe a half-dozen simultaneous clients.

  • Big servers (say, 4 cores, 64GBytes RAM): 20-30 simultaneous clients.

  • Additional Videobridges: Lots and lots of clients… (see the next item)

  • Bandwidth is not usually a problem.

  • Clients may matter. Some people report that Firefox slows things. This was true in April, is it true today?

Request: Please use the format below if you’re going to report your experience. To do this, simply select the text below and click “Quote”

I am successfully hosting X simultaneous users on hosting provider using a server with Y cores and Z Gbytes of RAM. Add other qualifying information, like estimates of bandwidth consumption, the point where it gets overloaded, or whether different clients matter…

Anyone else want to chime in? Thanks.

1 Like

We ended-up firing-up 6 extra Videobriges (4 cores 16gb RAM, 5Gbs Internet) on temporary VPS servers for the event.
When we “openend the gates” at 12:00, 1 000 users connected in the first five minutes, giving a huge load on the main Jitsi server but it worked flawlessly… Overall, we had 1 700 users, communicating and jumping from room to room at all time, the event lasted 5 hours.
In the end, a couple dozen users experienced issues with the platform and we had a live chat support setup to help them… Most of the issues were related to hardware, software or settings issues on the user’s side. The rest was due to issues we had never encoutered before and could not identify (but given that around 1600 users connected without issue, it was most certainly a user-side issue)

awesome ! I can only imagine the nervous tension…
Did you have time to look at the bandwidth vs video quality effectively delivered to clients ?