Time for Dedicated Server?

[captainsnarf]

UTComp adds its own PlayerReplicationInfo for the net code to work. There is no data in this object to replicate. The only thing replicated are the two function calls:

Code: Select all

replication
{
    reliable if(Role<Role_Authority)
        Ping;
    reliable if(Role == Role_Authority && bNetOwner)
        Pong;
}

The client sends a ping, gets a pong back from the server. That's all it does. It save the time it takes this to happen and uses that for aim compensation.

If this happens less often than the actors themselves replicate, then the timing will be off. If you are aiming at another player, but they get replicated before the pong happens, then the compensation doesn't match up anymore. It will be close, but not perfect. You will have to lead your targets. The point of UTComp net code is so you don't have to do that.

Since there isn't much data to replicate (two function calls) it really shouldn't bog down the server to have a higher frequency.

[pooty]

Ah, so 125 or 200 really shouldn't make any difference in network traffic, other than some small more frequent pings..

[captainsnarf]

Yeah, I added that comment at the end after you read this. It's hardly any data at all.

[captainsnarf]

The server was very laggy last night. I checked the usage and we were about 40% again. I made a donation.

[pooty]

Under 50% or so is usually fine. Previously when it spiked near 60% is when we saw issues. Last night night it looks like it never got above 30%
You can see on the NFO charts, CPU consumption is much lower than on our previous server.

I think we need to test it without some of the newer mutators, UTComp and Zounds. Those two were added, and the lag got much worse. I know UTComp is supposed to improve that and perhaps its just exposing the lag -- but I think we should test that out first, as its easy to do.

[Anonymous.]

I hate to say it but the way I see it, there's little point in buying a dedicated server unless people are willing to put forth a serious amount of money for however much longer we have. At our current rate we can economically survive for as long as the player base does. With NFO's dedicated offerings we won't get the performance we need unless we can stomach paying $220/mo for the Coffee Lake system (still not a worthwhile upgrade).

Unless NFO is willing to bend over for our needs and make a custom machine with only the hardware we need (extremely unlikely but I can ask via email, without mentioning Omni of course), that leaves us with having to self-host from someone's home. We don't need four nines uptime or insane bandwidth, just good enough to be able to handle the server+redirect during prime time (or offload the redirect somewhere else if bandwidth is the issue, quite annoying but wcyd).

[captainsnarf]

The E3-1270v5: A Skylake-based machine wouldn't be enough you think?

[Anonymous.]

Unless our lag originates from other servers on the machine (unlikely but VMs are shit for performance so I wouldn't put it past that), it would hardly be noticeable. There haven't been any meaningful IPC gains until Rocket Lake, which is 11th gen. for client CPUs - 11600K, 11700K, etc. Alder Lake (12th gen.) also had a sizable increase over Rocket Lake. Anything between Ivy Bridge and Skylake++++ was basically a clock/core increase. The 4 GHz clock limit also doesn't help since a lot of performance can be gained once you get near the 5 GHz region. I don't know what CPU we have now, but going off the old one, it's a 13% increase in single threaded perf which is lame.

https://www.cpubenchmark.net/compare/In ... 2154vs2651

[pooty]

Don't worry about the redirect, I have a website domain, that we used before for the redirect. It was just via bluehost, so it wasn't anything fancy but it got the job done.

I still think we need to check the software first before throwing hardware at it.

[Anonymous.]

Don't worry about the redirect, I have a website domain, that we used before for the redirect. It was just via bluehost, so it wasn't anything fancy but it got the job done.

I still think we need to check the software first before throwing hardware at it.

For sure, but even before UTComp performance wasn't the best, so it wouldn't be a waste even if we ended up disabling enhanced netcode.