Clarification about the “Load” field in the server status report

[mahdidev70]

I would like to better understand how the “Load” field in the server status report works.

What does this value represent exactly?

How is it calculated or measured?

Why does it sometimes show a lower load value for a server that actually hosts more users and sessions compared to another server with fewer sessions

[defnull]

The Load is just the number of participants in all meetings, with the exception that new meetings that were created less than 15 minutes ago are always counted with a minimum of 15 users. Moderators are also counted twice towards the load for whatever reason.

[mahdidev70]

What is the benefit or purpose of calculating the load this way?
@defnull

[defnull]

The new-meeting-penalty helps to prevent the 'trampling herd' problem when lots of meetings are created in a short time frame, e.g. in a teaching scenario where classes start at certain times. Empty meetings do not increase the load much, and new meetings are usually empty for a couple of minutes, so the server with the lowest current load would receive almost all new meetings until users actually start to join. To prevent that, the balancer just assumes that new meetings will grow within a few minutes and account for the higher future load.

If 15 load-points or 15 minutes are are a reasonable assumption depends a lot on your actual usage pattern. It would be nicer if the penalty would slowly decrease, and depend on the actual average meeting size. There are other BBB load balancers out there with more sophisticated algorithms.

I have no idea why moderators are counted twice, though. That's odd.

[mahdidev70]

@defnull
That was useful information.
Are the default 15 users and 15 minutes configurable?
What are the names of the other load balancers, and how do they differ?

[defnull]

Yes, via LOAD_JOIN_BUFFER_TIME and LOAD_MIN_USER_COUNT.

Out of my head there are:

• b3scale: Written in go
• b3lb: Written in Python/Django
• BBBLB: My own project. It's very new and not ready for production yet, but I cannot not mention it here.
• Pilos: Not a load-balancer, but a front-end with built in load balancing.

There are more out there, but those mentioned above are probably your best options if you are looking for alternatives.