Server question CPU usage
AssassinY2K
Join Date: 2002-11-09 Member: 7931Members, Reinforced - Supporter
With no players on the server it is using 5~6% of the CPU, as soon as 1 person joins it jumps to 100% CPU usage.
I set player limit to 12 on summit map
I am using a Dell PowerEdge 1950
2 x Dual-Core XEON 5160 3.0Ghz Processors
16Gb Ram
2 x 73Gb SAS 10k 2.5" HDD's
Windows Server 2008 x64
What is causing it to use so much? Am I missing a setting somewhere?
When I joined, tick rate was a stable 30, but when I filled the 12 slots with bots, it went down to 20~24 tick rate, thats with about 20 sentry guns and several buildings
I set player limit to 12 on summit map
I am using a Dell PowerEdge 1950
2 x Dual-Core XEON 5160 3.0Ghz Processors
16Gb Ram
2 x 73Gb SAS 10k 2.5" HDD's
Windows Server 2008 x64
What is causing it to use so much? Am I missing a setting somewhere?
When I joined, tick rate was a stable 30, but when I filled the 12 slots with bots, it went down to 20~24 tick rate, thats with about 20 sentry guns and several buildings
Comments
I set player limit to 12 on summit map
I am using a Dell PowerEdge 1950
2 x Dual-Core XEON 5160 3.0Ghz Processors
16Gb Ram
2 x 73Gb SAS 10k 2.5" HDD's
Windows Server 2008 x64
What is causing it to use so much? Am I missing a setting somewhere?
When I joined, tick rate was a stable 30, but when I filled the 12 slots with bots, it went down to 20~24 tick rate, thats with about 20 sentry guns and several buildings<!--QuoteEnd--></div><!--QuoteEEnd-->
a dual core seems kind of weak for a server, that probably explains why.
Ns2 server.exe only uses 1 core...
I would have to agree, I set CPU affinity to the 2nd CPU (no programs) and got the same result
I'm having a problem with starting a dedicated server. Once up, you can see it in the server list and also connect, but after 1-2 minutes it disappears. Port is forwarded correctly (27014 - 27016 to keep sure), I even deactivated the routers and pcs fire walls to make sure that this is not the problem. (searching for the problem got me no useful results)
Also, after the server disappeared from the list i think you should be still able to connect "manually" by typing connect in the console.
Anybody knows what could be wrong?
just wait
I would sacrifice a small child to get 20-24 tick rate under those circumstances... seriously. I'm running an HP ML110 G5 and once my server starts filling up, 5-8 tick rate.
Even the HBz and Frelge servers don't maintain a steady 30 tick rate in the heat of the moment.
Lets see what happens....
Just to confirm, what ports on a dedicated server must be opened to the public? <b>27015 </b>for steam/hlds and <b>27016 </b>for ns2 server on UDP ?
Lets see what happens....
Just to confirm, what ports on a dedicated server must be opened to the public? <b>27015 </b>for steam/hlds and <b>27016 </b>for ns2 server on UDP ?<!--QuoteEnd--></div><!--QuoteEEnd-->
27015 is the connect port
27016 is the query port
Both are UDP only.
So u say, Server.exe use 1 core. Thats right btw. and tell us ur going to upgrade to 24 cores ?
What should happen in your opinion ?
I tell you: 23 cores do nothing. The server.exe need 500mb ram. So you have 63gb left.
27016 is the query port
Both are UDP only.<!--QuoteEnd--></div><!--QuoteEEnd-->
27016 also provides Steamworks, which I believe Spark currently does via UDP, but is also available via TCP, so if TCP & UDP is possible you might want to do that.
Lets see what happens....<!--QuoteEnd--></div><!--QuoteEEnd-->
Since, right now the single-core speed is the biggest determining factor, I question whether you'll see any improvement doing this. That's why the guys running servers on home gamer PCs (Win7, i5/i7) have better quality servers than ones doing it on dedicated server systems.
Also, unless you plan on hosting a lot of other game/VoIP/servers on this system (or you're running this for free), this might be a big waste of money.
So if my i7 970 Hexacore is overclocked to 4.2 Ghz, I am going to trounce a 2.4 Ghz Xeon without question.
So if my i7 970 Hexacore is overclocked to 4.2 Ghz, I am going to trounce a 2.4 Ghz Xeon without question.<!--QuoteEnd--></div><!--QuoteEEnd-->
Early tests on this system indicate the same problem, luckily it was a work server so I didn't pay the £8000 server cost :)
I won't be attempting another setup until I can change the port and run a dedicated server on a dedicated server :)
tl;dr: sleeping is (usually) imprecise, using a full core gives better timing.
The fix was to have the server essentially sleep when there's noone connected. The server using 100% cpu when there's any players connected is a "feature", not a bug. I doubt this is really an easy thing to fix, and it's probably going to wait until the engine gets some more/different optimizations.
My question is, can server ticks be affected in a similar way, with micro-tick-stuttering? And if so, is there a way to track it? I imagine CPU usage 100% would correspond to a server micro-stutter.
There are bottlenecks, hopefully they will be found and addressed soon.