• Please make sure you are familiar with the forum rules. You can find them here: https://forums.tripwireinteractive.com/index.php?threads/forum-rules.2334636/

Server Server Performance Configuration Tweaks?

Devourer said:
I would just like to post for the record. I did read the thread & never was I really replying to you "Ninjaboy". You stated in your 1st post. Even though that doesn't really make sense as right before you stated which totally sounds like a lack of bandwidth to me...... In any case. I was replying to post #2 "[5thW]Heide" who said & #6 "Zetsumei" which both most certainly sound like a lack of bandwidth. Although given what Zetsumei said I believe he has pretty limited bandwidth.

bandwith of the server isn't the problem here. even if we set it at 25000 the problem won't occur earlier. Thats what i wanted to point out setting the internet rate to something higher or lower won't change the fact at at around 28+ players the pings go up.

While if changing the tickrate (tickrate eats cpu performance alive) if you set it too high for the server to handle you'll encounter lag but not the pings going up.

Its really looking like a bandwith problem but bandwith is the last thing that should be a problem here ^^. Its most likely the cpu, but its just weird that changing the tickrate has no effect on when the pings start to act weird.
 
Upvote 0
[RO]Ramm-Jaeger said:
The dedicated server for the Unreal 2.5 Engine is quite CPU hungry. I highly recommend a dedicated processor of Intel 3.0 GHZ and up or an AMD 3000+ and up for a 32 player server. The ram usage however is nowhere near 2 gigs. The ram usage for the server should never go over 512 mb (if not even much lower than that). While BF2 can do 64 player servers, there is also an issue of fidelity and the types of levels the engine can do. BF2 can't do highly complex geometry (for things like building interiors) like Unreal can. We're working on some CPU improvements for the servers for future patches, but we can't really rewrite the engine or anything of that nature. Unreal will always be Unreal, and it is just CPU heavy.

Something I thought of ... what this game (and any FPS) really needs is more destructible environments. Meaning if your tank shoots a building it goes down. Soldner's ability to do this was incredible but unfortunately the game suffered from numerous other problems. I'm hoping the next Unreal engine supports this ability as it would make the realism aspect incredible.
 
Upvote 0
Testing with rates higher than 10000 to see any difference is nonsense as all clients have a maxrate set to 10000 by default. So no client even tries to use your avised 25000 or what ever. If you want to check your bandwith you need to have 28 players with a clientrate higher than 10000. Just as a hint for your testing.

And tickrate DOES have an impact on performance. But you have to test it on tankmaps, Arad for example. If will recognize that it reaches the critical limit way earlier if you have a tickrate of 35. On the other hand you can squeeze 2 slots more if you set it to 20 but that would have a negative effect on hit calculation, as a noticabel amount of bullets and even shells seems to go through the target without effect.

I am testing on an AMD64 3000+ with 1GB only dedicated to running the RO server. To have a lag free game with standard settings, we can run 26 players max.
 
Upvote 0
GS_Schimpf said:
Testing with rates higher than 10000 to see any difference is nonsense as all clients have a maxrate set to 10000 by default. So no client even tries to use your avised 25000 or what ever. If you want to check your bandwith you need to have 28 players with a clientrate higher than 10000. Just as a hint for your testing.

And tickrate DOES have an impact on performance. But you have to test it on tankmaps, Arad for example. If will recognize that it reaches the critical limit way earlier if you have a tickrate of 35. On the other hand you can squeeze 2 slots more if you set it to 20 but that would have a negative effect on hit calculation, as a noticabel amount of bullets and even shells seems to go through the target without effect.

I am testing on an AMD64 3000+ with 1GB only dedicated to running the RO server. To have a lag free game with standard settings, we can run 26 players max.

I believe if you set your connection in game to LAN/T1 it will increase the Maxclient rate of your game to 15000 (which is the default maxclient rate of LAN in the config). You could also type this in your console once you're inside the game:

set IpDrv.TcpNetDriver MaxInternetClientRate 10000 (or whatever you wish)

That should change it on the fly but you can hit F6 to test it. I noticed that on some of the vehicle maps when moving you hit that 10000 cap. Most of the time although you do not. Might be interesting to test the server with 15000 and set yoru connection to LAN. Most broadband clients can support that speed easily but just make sure your upload rate on the server can of course.
 
Upvote 0
GS_Schimpf said:
Testing with rates higher than 10000 to see any difference is nonsense as all clients have a maxrate set to 10000 by default. So no client even tries to use your avised 25000 or what ever. If you want to check your bandwith you need to have 28 players with a clientrate higher than 10000. Just as a hint for your testing.

And tickrate DOES have an impact on performance. But you have to test it on tankmaps, Arad for example. If will recognize that it reaches the critical limit way earlier if you have a tickrate of 35. On the other hand you can squeeze 2 slots more if you set it to 20 but that would have a negative effect on hit calculation, as a noticabel amount of bullets and even shells seems to go through the target without effect.

I am testing on an AMD64 3000+ with 1GB only dedicated to running the RO server. To have a lag free game with standard settings, we can run 26 players max.

most people actually did it when i asked them to change it i asked every individual ^^ and checked personally if they did change it. and if its a socalled bandwith problem why doesn't it stop ocurring at a rate of 5000 ^^.
 
Upvote 0
As I said: bandwith doesn't appear to be the problem. Rather try some experiments wit the rate settings. If set up correctly you will clearly notice the impact on performance. Although it is not recommended to go lower than default with the tickrate.
Trust me, I have been testing with this stuff since the early mod days as we first set up a server for RO. Back then, performance did increase largely with introduction of 3369 patch. Too bad it doesn't seem to have any impact for the retail release...
 
Upvote 0
The strange thing is that ucc.exe does appear to be running 4 threads on a hyperthreaded server.
Now that's not to say that all 4 threads are being hyperthreaded but something is being shared out somewhere it would seem?

I am looking at my servers task manager right now and the Cpu usage for ucc.exe is showing 24% (on a tank map with full 32 players) in the process section but clicking on the performance tab it clearly shows that both real Cpu's are running about 13% each appx.

And i am seeing 100+ pings for 80% of the players on the server?

This seems very strange for a game that's not supposed to support hyper threading.
So it would appear to me that whilst I do agree Cpu usage will always impact player pings if it goes beyond 80% but the UT2 engine itself is in fact playing some part in these highish pings on a full server running a tank map.
My Cpu load would be high if it was seen on a single HT cpu but I am not seeing this at all on a single cpu?
Ergo the high pings are in part at least being caused by I think a mix of netcode, the engine and the players personal set up.
My server has unlimited bandwidth so I have discounted this as an issue but with a cap on maxrate of 10000 could this be forcing the server cpu to work harder?
Would a bigger cap make it easier for the serverside to handle the information or would it just cause more problems?

When you consider that most players on virtually all the servers I have seen appear to be running at 100+ pings on a full 32 player server whilst playing on a tank map.
The quality of PC, quality of connection and how they have set up the game all seem to play their part in impacting on a players ping whilst playing RO.

I have also noticed that the netcode appears to compensate for players having either lower end PC's and or poor connections and still makes the game very playable for them and I wonder if this is a trade off with the server taking up the slack?

You can see this especially on Infantry maps where the average player with a decent PC and connection will see his ping drop by around 50/60% to around 60 or so whilst some not so fortunate will still be playing at around 100.

It would be most welcome if the Devs could publish some results and reccomendations regarding ideal server set ups for both Windows and Linux.
All of us Admins want to run the best and most efficient servers we can so if they have some hints or tips on how to get the best from this engine with regards to running a server that would be very helpful.
 
Last edited:
Upvote 0
ucc is only capable of using one core. So even HT is not used. no chance to get it to use.

Your load % are just not right, as it is only a calculated percentage on average of all 4 or 2 + HT cores.

32 players on a tank map will definately use up one core to its maximum capacity. Only good thing for multicore systems is that you can run multiple instances of RO that don't interfere with each other on one machine. ;)
 
Upvote 0
I don't dispute what should be happening Schimpf.

UT2 should be maxing out a single HT core....I can't work out why I am not seeing that in the task manager?

ucc.exe should be using 1 cpu core and if that is the case I should be seeing 25% = 100% on that core on a HT enabled server.
The process reads as using 24% (with 4 threads running) but the HT screens show that it appears to be spread over 2 cpu cores?
Very strange....maybe some lesser parts of the engine are being shunted across?

Certainly the server engine is a real cpu pig either way and it does need some better optimization to get it to run at lower levels.
I just hope that the Devs will really try to improve this in the near future.
 
Upvote 0
As I said: bandwith doesn't appear to be the problem. Rather try some experiments wit the rate settings. If set up correctly you will clearly notice the impact on performance. Although it is not recommended to go lower than default with the tickrate.
Trust me, I have been testing with this stuff since the early mod days as we first set up a server for RO. Back then, performance did increase largely with introduction of 3369 patch. Too bad it doesn't seem to have any impact for the retail release...

I've been experimenting with stuff since the mod days aswell, ever played on the core 32 players server back then ....?

Its just that the comp can't handle ro anymore, changing to linux would help probably for the performance but thats still instable. (we ran the mod on linux) its just mean that the mod performed better on 32 players than ostfront does. On thesame comps.

We just need a new cpu for the comp probably Xeon 3.06 just doesn't cut it ^^.
 
Last edited:
Upvote 0
I will say this,I put up an ostfront server for a client on a machine that has a 3800 amd dual core and over 2 gigs of ram and a 100mb port and as soon as that server maxed at 16 the task manager showed ucc absolutely hogging all resources.Unreal tournament was never this bad and changing tickrates and client rates did absolutely nothing.People started reporting pings all over the map from 130 to 300+ This game is a complete resource hog and i will be damned if i will spend 220 dollars a month to run just one ostfront server.I sure hope something happens in this next patch otherwise I am stopping advertising osfront servers for rent.I love the unreal engine but as it pertains to this game something is horribly wrong.
 
Upvote 0
I must admit i have seen an increase in resources since the last patch and the devs must now seriously consider doing something about this before it becomes a real problem to run an RO server.
For instance I have run 32 player BF2 servers and fully loaded they have run at lower cpu levels than a full RO server and BF2 has boats and aircarft plus a big array of land vehicles!

Whilst I appreicate that the engine is having to work hard on creating and keeping tabs on everything it is my experience that games that are CPU hogs usually begin to become unpopular with server rental companies as they simply use way too much in the way of resources and they therefore begin to charge much more for renting these.
Eventually leading to a tail off of servers appearing online.

I was disappointed in the server side of the last patch because it offered no optimization at all.
If anything as I have stated it has worsened the situation.
Please Devs give some serious thought to getting the serverside to run with much lower cpu levels as I fear you may end up losing a great many servers if the present levels are maintained.
 
Last edited:
Upvote 0
Any suggestions Tripwire?

Any suggestions Tripwire?

To start, I have a 50 player Art of War Central server. I am getting absolutely frustrated with them, for they know nothing about RO. (I kid not). Tripwire! What is the recommended optimization for my server? I have yet to be able to host 50 players. You recommended them; please conference with them to help them understand the workings of RO. Because if they did understand, I would not be posting here. So again, please respond and let me know what I need to do in order to get my server running properly. Thank you. p.s - all I need is a checklist
 
Upvote 0
Options

Options

Thanks for reply. So currently from Art of War Central I have: Woodcrest dual core 2.0 GHz, 2000MB Ram, 160GB SATA.

My options are: Dual Xeon 3.0GHz, 4000MB Ram, 160GB SATA

or

Pentium 4 3.0GHz, 1000MB Ram, 120GB SATA

What should I do? Any info would be great, then I can go to AoWC and pummel them for pushing dual cores for RO; which cannot be utilized, TY.
 
Upvote 0
Thanks for reply. So currently from Art of War Central I have: Woodcrest dual core 2.0 GHz, 2000MB Ram, 160GB SATA.

My options are: Dual Xeon 3.0GHz, 4000MB Ram, 160GB SATA

or

Pentium 4 3.0GHz, 1000MB Ram, 120GB SATA

What should I do? Any info would be great, then I can go to AoWC and pummel them for pushing dual cores for RO; which cannot be utilized, TY.

dual xeon will not help you at all. take better a single xeon but with a newer arhitecture even if it has slightly lower GHz. The bigger cache (2X6) and shorter execution path makes a 2.5GHz Xeon faster than a 3GHz older xeon.
 
Upvote 0