www.omnipotents.com/forums

snake98 wrote: ↑Wed Feb 22, 2023 10:10 am Where do you see the nodes constructed and destroyed at?

https://www.omnipotents.com/utstats/mat ... match=3384

Yes at $50, but is a dedicated server worth it? I'm guessing there is contention somewhere in the network stack. Might not even be on the node. Snarf mentioned seeing much better performance on Landlords vs NFO which originated the tests with Roger's server. There are other weird things, like a mention in another thread that clients communicate with the server at the frame rate. If that's true, a high frame rate can DDOS the server, although I assume it intelligently drops surplus packets. Another question, is the server code multithreaded?

No it's not multithreaded, and the current CPU can't handle the load. It requires brute force single threaded performance, which costs $$$$$$$$$$$$ if you're not hosting it yourself.

Landlords *is* NFO also. I was wrong. It's weird it doesn't respond to ping like HoC and Omni do, but doing net traces it is definitely NFO.

Framerate is limited by your netspeed. Omni caps the netspeed at 13600. That means 13.6Kb/s * 32 players = 435.2Kb/s bandwidth requirements. The max netspeed the game supports is 20000, so 640Kb/s. Any reasonable network should be able to handle this game. When we were testing CEONSS dedicated linux server I ran a network monitor and bandwidth usage was pretty low.

Another factor is the server tick rate. 13600 = 13.6Kb/s when the tick rate is 30. Increase the tick rate and you increase the bandwidth requirements.

Everything in the game that goes over the network has a network priority. These are all sorted by priority each tick before being sent out the network. If there's not enough bandwidth or cpu time to process everything, lower items in the priority get dropped. This is why you might see missing tank turrets or link gun beams when the server is overloaded. A detailed description of UT2004 netcode can be found here:

https://docs.google.com/document/d/1KGL ... q6LPY/edit#


The server is NOT multithreaded.

I capped at 13600, because faster netspeeds aren't going to do much except put more strain on the server.
https://www.utzone.de/forum/showthread.php?t=4326

And any fps over 205 isn't really necessary, you can run those locally, but we don't need you swamping the server for frames you probably can't even display -- monitor refresh matters if you at 144hz, your monitor will show max 144 fps.

We run the tick rate at 28. Higher Tick rates really drive up CPU. Every actor (player, vehicle, etc) usually has "tick" function that runs every tick. Vehicles can be pretty tick heavy vs other things.

What we wanted to test out one night either on Roger's or Anon's server was if we had dedicated fast CPU, does it run similarly to NFO, meaning we had an engine limitation, vs CPU limitation.

poof

NFO sucks. You hit that guy with triple rox to the face and nothing happened.

omni, hosted at NFO in chicago


hoc, hosted at NFO in chicago


Look at those big lag spikes. And that's at 4am. It's not really NFO's fault though. Their ISP sucks. The average latency is OK but consistency is not very good.

Those routes are different: beNNNN.ccrNN are all different machines, that could account for different latency.
I guess Cogent isn't good? https://cogentco.com/en/

And none of that routing mess would be mitigated by a private server, you'd still have the same issue.

But we are back to the point where we need to test the server on dedicated, faster hardware, I don't think we are going to be able to do much with ISP routing.

Those graphs are from pingplotter.

Maybe some others can try from their locations and share pics?

Apart from the third hop (probably throttles icmp) it looks ok. In-game ping settled at 48 with an empty server just now. Think I get ~60 with more players.