Dedicated server CPU usage

Our game calls for a limited number of players operating inside a large world, but currently there is no plan to allow users to host their own games. This means that we may have to run a high number of server machines and so would like to get as many instances of the dedicated server running as possible.

The game contains some reasonably complex physics objects, so I’m looking at the best ways to get CPU usage down on each dedicated server. As a server cannot (as far as I know) share assets or spawn multiple levels there is an inherent overhead for each instance which makes an impact when trying to cram as many players as possible onto one machine.

I’m currently in the process of running our game with various different physics settings and scenarios to try to measure to impact of different options, but are there any benchmarks showing the results of dedicated servers running on Windows and Linux in different use cases?

Also, what is the best way to profile the CPU of a running dedicated server? Is there any built-in mechanism for this or does it have to be done with a third-party profiler?

I’m not aware of any dedicated server benchmarks that are available, and in addition, the results will vary significantly for different games, depending on the characteristics of the specific one being benchmarked.

We do have a built-in stats system that is a good place to start with profiling. There’s a blog post about it here, it’s from a little while ago but the information is still relevant. You might want to use another profiler as well once you get into lower-level, more granular optimizations - the one built in to Visual Studio is pretty good for Windows.

Hi,

We think this post contains useful information which we would like to share with our public UE4 community. With your approval, we would like to make a copy of this post on the public AnswerHub which includes the discussion but strips out your username and company name. Please let us know if you are okay with this.

Thanks!

Yes, of course.