Ideal Specifications, large event

Hi Folks,

I know these sorts of questions are painful but i’d rather ask the experts. We will be running a self-hosted conference on Jitsi with over 50 rooms and 1000 clients connecting at once.

Is a single, powerful dedicated server with 16 threads, 128GB RAM and a 10Gbps internet connection going to cut it? Or are we better off spreading the load over many virtual machines?

Let me know your thoughts and suggestions. Really appreciate your time in advance.

I should add that these users are from all around the world. Is it better in that case to have a few VMs in a few countries and somehow loadbalance them?

You’ll need multiple JVBs to support your scenario. Of course there’s value in having the bridges as close as possible to where the users are connecting from as that potentially helps reduce RTT. In that case, unless each room will be hosting people all from the same region, you’ll also benefit from implementing Octo.

Could someone help elaborate some more on this? what specifics elements i’ll need to allow seamless operation? I’ve only deployed single instances before so would appreciate some guidance. Even if you point me to some guides for each element i need that will be enough, but specs wise if someone can give me a rough idea on what i need for the VMs globally that would be great.

jitsi cluster

anyone wanna get paid to put this together?

A topic in paid-work category will work

what are the ideal specs per VM? Say we have 300 users in Australia, 200 in the US, 200 in Europe and 300 across africa and Asia. How many VMs woukd you deploy per region and with what specs based on these numbers?

A single 8-core (real cores, not threads/vCPUs) JVB in each region would be adequate for those numbers if it has enough bandwidth (real bandwidth, not marketed port speeds) and is well set up. Would recommend at least two for redundancy, of course, and it may be worth splitting across multiple locations within each of those regions for latency reasons, especially in Asia, but if you know where your users are that gets easier.

If you need a conference to span multiple JVBs due to conference size, you need secure, high-performance connectivity between the JVBs for Octo, too. It can be best to shard by region, if conferences tend to be localised to a region, so that you only need to deal with that backend transfer within a region rather than internationally.

Feel free to DM if you’d like to discuss in detail, large Jitsi deployments are my focus.