...then I updated the kernel to kernel 2.6.10-1.14_FC2smp & the result is... <!--QuoteEnd--> </td></tr></table><div class='postcolor'> <!--QuoteEEnd--> Do you have the metamod and AMX on just one server or both, I'm not sure I understand. If only 1 server, does the server without the plugin stuff do better?
When you upgraded to the 2.6.10 kernel, did you use RPMs or did you compile it yourself?
...then I updated the kernel to kernel 2.6.10-1.14_FC2smp & the result is... <!--QuoteEnd--></td></tr></table><div class='postcolor'><!--QuoteEEnd--> Do you have the metamod and AMX on just one server or both, I'm not sure I understand. If only 1 server, does the server without the plugin stuff do better?
When you upgraded to the 2.6.10 kernel, did you use RPMs or did you compile it yourself? <!--QuoteEnd--></td></tr></table><div class='postcolor'><!--QuoteEEnd--> First. Both servers have metamod with amx0.99 & amxmodx 3 of them together. (I have sum reason cuz I have to run amx & amxmodx together. cuz some plugin problems make me do that.) Second: I just use yum to update. I didn't use RPMs or compile it myself. (that should be RPMs. right??)
i provide a Natural Selection 3.01 Server on a Debian 3.1 with a Sempron 2200 und 512 RAM.
Ive build my own kernel and setted some things like fps_max and sys_ticrate, but the server needs too mutch cpu power.... with 14 Players on the Server i get 70-90 % cpu but memory looks nice ... about the 15% ....
So, what can i do??
One machine for one NS Server is ill <!--emo&:)--><img src='http://www.unknownworlds.com/forums/html/emoticons/smile-fix.gif' border='0' style='vertical-align:middle' alt='smile-fix.gif' /><!--endemo-->
<!--QuoteBegin-xantrox+Mar 27 2005, 11:28 AM--></div><table border='0' align='center' width='95%' cellpadding='3' cellspacing='1'><tr><td><b>QUOTE</b> (xantrox @ Mar 27 2005, 11:28 AM)</td></tr><tr><td id='QUOTE'><!--QuoteEBegin--> Ive build my own kernel and setted some things like fps_max and sys_ticrate, but the server needs too mutch cpu power.... <!--QuoteEnd--> </td></tr></table><div class='postcolor'> <!--QuoteEEnd--> Try reducing the Hz setting.
Whoa. Amazing. I didnt think ns would take so much cpu :o I tried installing it on my server just to see how it would run - and - yeah, ~80%cpu at 15 player (1800mhz)
And by reading the other posts here it seems windows runs the server much better. So this seems like a serious server issue. Im not going to run a server that takes up most my cpu power when im also using it to NAT and other stuff.
Has anybody even had luch with the Hz thing? Lowering this would prolly mean less cpu utilization but also less hlds fps - which just suck.
Maybe one of you should contact the devteam and make sure they know about this issue and make them look into it asap.
<!--QuoteBegin-jasonmog+May 5 2005, 09:19 AM--></div><table border='0' align='center' width='95%' cellpadding='3' cellspacing='1'><tr><td><b>QUOTE</b> (jasonmog @ May 5 2005, 09:19 AM)</td></tr><tr><td id='QUOTE'><!--QuoteEBegin--> it takes up so much cpu because linux is a poop operating system. it doesn't optimize nearly as much as windows. <!--QuoteEnd--> </td></tr></table><div class='postcolor'> <!--QuoteEEnd--> dont post ignorant comments about what you are unfamiliar with.
linux is not a bad os, and neither is windows. some of us just prefer linux for one reason or another.
the nice thing about linux is that you can change it. Im using windows right now, but when I start up my server im switching to linux strait up. In any old distro you'll get better multitasking and resource management just from the kernel, from there you can turn off unnecissairy features and change quite a bit to recompile the kernel with your custom settings, for further optimization, stability, and hack prevention. On top of that most distros are free, you can't go wrong with free.
The only thing I really have against windows is that its like a dell computer, it comes loaded with stuff I'd never care to use or want to have installed. It takes up space and power and can be sometimes quite hard to get rid of.
<!--QuoteBegin-jasonmog+May 5 2005, 06:19 AM--></div><table border='0' align='center' width='95%' cellpadding='3' cellspacing='1'><tr><td><b>QUOTE</b> (jasonmog @ May 5 2005, 06:19 AM)</td></tr><tr><td id='QUOTE'><!--QuoteEBegin--> it takes up so much cpu because linux is a poop operating system. it doesn't optimize nearly as much as windows. <!--QuoteEnd--> </td></tr></table><div class='postcolor'> <!--QuoteEEnd--> Actually, this is valves issue here not linux. Most other games(bf1942, ut, etc etc) have very solid linux server implementations that work better under linux than windows. This is because valve has a single person working on the linux port and his skills aren't up to par with the other games out there.
If valve would put the same amt of work into their linux code as the rest of the game vendors who care about linux, hlds would run like something other than crap.
I do agree with you tho, right now the wintel dedicated server runs better. I don't agree that this is linux fault, the fault is directly in valve's court.
<a href='http://kerneltrap.org/node.php?id=464' target='_blank'>In this KernelTrap article</a> they talk about how previously the 2.4 kernel (the one in my 7.3 box) is typically set at HZ=100. On Redhat 8 and 9 Kernels this is set to HZ=512. Nearly all 2.5/2.6 Kernels are set at HZ=1000. Could this one change affect the perfomance of Linux systems so dramatically? I think I'm going to see what happens if I recompile the FC2 kernel with HZ=100 setting...this might take a while so don't hold your breath. <!--QuoteEnd--> </td></tr></table><div class='postcolor'> <!--QuoteEEnd--> What's happened with this testing...
Vadakill did this testing a while ago and it showed that 100 hz lowered CPU usage. Most new kernels have the option to change the kernel interrupt frequency in menuconfig now, so change it to 100 hz for server usage (as recommended). Also, removing desktop-friendly features such as preemption can also improve performance, but I personally haven't seen how much they lower performance as I always have preemption off.
puzlThe Old FirmJoin Date: 2003-02-26Member: 14029Retired Developer, NS1 Playtester, Forum Moderators, Constellation
2.6 kernels tend to allow you to change the user hz dynamically. Simply change sys_ticrate in your hlds and your server cpu occupancy will increase producing more server side fps and consuming more cpu. Finding the right balance of cpu/server fps is something each serverop is going to have to do for themselves.
Let me know if anyone has more detailed test results, as I'd be interested in seeing them.
The NSA servers run with pretty bad CPU even though they are windows boxes. They usually run somewhere from 30% to 35% cpu each. It's probly because of their size because they are much bigger than 16 man but when we ran a cs server of similer size it took almost 1% cpu full. <img src="style_emoticons/<#EMO_DIR#>/nerd-fix.gif" style="vertical-align:middle" emoid="::nerdy::" border="0" alt="nerd-fix.gif" />
<!--quoteo(post=1563465:date=Jul 19 2006, 10:21 AM:name=Urd)--><div class='quotetop'>QUOTE(Urd @ Jul 19 2006, 10:21 AM) [snapback]1563465[/snapback]</div><div class='quotemain'><!--quotec--> The NSA servers run with pretty bad CPU even though they are windows boxes. They usually run somewhere from 30% to 35% cpu each. It's probly because of their size because they are much bigger than 16 man but when we ran a cs server of similer size it took almost 1% cpu full. <img src="style_emoticons/<#EMO_DIR#>/nerd-fix.gif" style="vertical-align:middle" emoid="::nerdy::" border="0" alt="nerd-fix.gif" /> <!--QuoteEnd--></div><!--QuoteEEnd-->
That's because CS doesn't have structues like offense chambers or turrets. 30-35% still isn't horrible, especially if it's a huge server.
ok, something gotta be wrong with the linux hlds ns server!
Check this:
I have a box running linux with a pretty weak cpu. Anyways, when I run hlds natively on linux (Debian etch), I get 30% idle cpu usage with 6 bots.
<!--c1--><div class='codetop'>CODE</div><div class='codemain'><!--ec1-->CPU In Out Uptime Users FPS Players 33.25 3.16 2.89 1238 451 60.14<!--c2--></div><!--ec2-->
This is after I set the sys_ticrate to 60 for comparison (on windows, fps dont go higher than 60). Without this, the fps would be around 90-100 but the cpu usage is the same.
Now, guess what happens when I run the hlds within a Windows 2003 Server VMWare environment on the same box? (yes! virtualized environment!) ->
<!--c1--><div class='codetop'>CODE</div><div class='codemain'><!--ec1-->CPU In Out Uptime Users FPS Players 1.55 10.71 10.52 16 17 64.42<!--c2--></div><!--ec2-->
Though I just checked the vmware cpu usage, and it is close to the other hlds process (that runs natively).
So, in conclusion: I think its only some sort of bug in the display of the actual cpu usage.
Well first of all, check your uptimes and user counts. They should be around the same to have an eligible comparison. HLDS tends to get more resource hungry after it has been running for some time. No doubt linux version has it worse.
Also, are you running plugins (metamod, amxx etc.) ?
In windows, you can get fps higher using -pingboost and I think you need to start Windows media player or something to enable higher kernel tickrate.
Btw, You'll get a better idea of the whole CPU usage using uptime to get average system load.
No, even a fresh start of hlds on linux has 30-50% cpu with 6 whichbots. Fresh start on windows with 6 whichbots has 0% cpu in windows but the vmware reports to have a usage of 30-50% as well. (repeat: Cpu usage inside the vmware in windows is 0-5%, vmware cpu usage in linux is 30% -> the win vm gets 30% cpu but does not really use it)
Thats why I think there is some sort of bug that causes linux to think that hlds is using a lot more cpu than it is actually using.
<!--quoteo(post=1700148:date=Feb 12 2009, 11:52 AM:name=asmodee)--><div class='quotetop'>QUOTE(asmodee @ Feb 12 2009, 11:52 AM) <a href="index.php?act=findpost&pid=1700148"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->While I'm not sure, I think hlds on linux reports in a similar manner to that of uptime, meaning it reports load, not actual cpu usage.<!--QuoteEnd--></div><!--QuoteEEnd-->
Nonsense, a load of 33, would mean 3300% cpu usage, or 33x time the load the cpu could handle, on my centOS box, the load on a ns server with running roughly 500-1000fps is around 3%, loaded i guess 20%. (16 slots).
On windows this is a bit less, but almost the same.
Comments
...then I updated the kernel to kernel 2.6.10-1.14_FC2smp & the result is... <!--QuoteEnd--> </td></tr></table><div class='postcolor'> <!--QuoteEEnd-->
Do you have the metamod and AMX on just one server or both, I'm not sure I understand. If only 1 server, does the server without the plugin stuff do better?
When you upgraded to the 2.6.10 kernel, did you use RPMs or did you compile it yourself?
...then I updated the kernel to kernel 2.6.10-1.14_FC2smp & the result is... <!--QuoteEnd--></td></tr></table><div class='postcolor'><!--QuoteEEnd-->
Do you have the metamod and AMX on just one server or both, I'm not sure I understand. If only 1 server, does the server without the plugin stuff do better?
When you upgraded to the 2.6.10 kernel, did you use RPMs or did you compile it yourself? <!--QuoteEnd--></td></tr></table><div class='postcolor'><!--QuoteEEnd-->
First. Both servers have metamod with amx0.99 & amxmodx 3 of them together. (I have sum reason cuz I have to run amx & amxmodx together. cuz some plugin problems make me do that.)
Second: I just use yum to update. I didn't use RPMs or compile it myself. (that should be RPMs. right??)
i provide a Natural Selection 3.01 Server on a Debian 3.1 with a Sempron 2200 und 512 RAM.
Ive build my own kernel and setted some things like fps_max and sys_ticrate, but the server needs too mutch cpu power.... with 14 Players on the Server i get 70-90 % cpu but memory looks nice ... about the 15% ....
So, what can i do??
One machine for one NS Server is ill <!--emo&:)--><img src='http://www.unknownworlds.com/forums/html/emoticons/smile-fix.gif' border='0' style='vertical-align:middle' alt='smile-fix.gif' /><!--endemo-->
thx,
Xantrox
Try reducing the Hz setting.
And by reading the other posts here it seems windows runs the server much better. So this seems like a serious server issue. Im not going to run a server that takes up most my cpu power when im also using it to NAT and other stuff.
Has anybody even had luch with the Hz thing? Lowering this would prolly mean less cpu utilization but also less hlds fps - which just suck.
Maybe one of you should contact the devteam and make sure they know about this issue and make them look into it asap.
dont post ignorant comments about what you are unfamiliar with.
linux is not a bad os, and neither is windows. some of us just prefer linux for one reason or another.
The only thing I really have against windows is that its like a dell computer, it comes loaded with stuff I'd never care to use or want to have installed. It takes up space and power and can be sometimes quite hard to get rid of.
Actually, this is valves issue here not linux. Most other games(bf1942, ut, etc etc) have very solid linux server implementations that work better under linux than windows. This is because valve has a single person working on the linux port and his skills aren't up to par with the other games out there.
If valve would put the same amt of work into their linux code as the rest of the game vendors who care about linux, hlds would run like something other than crap.
I do agree with you tho, right now the wintel dedicated server runs better. I don't agree that this is linux fault, the fault is directly in valve's court.
<a href='http://kerneltrap.org/node.php?id=464' target='_blank'>In this KernelTrap article</a> they talk about how previously the 2.4 kernel (the one in my 7.3 box) is typically set at HZ=100. On Redhat 8 and 9 Kernels this is set to HZ=512. Nearly all 2.5/2.6 Kernels are set at HZ=1000. Could this one change affect the perfomance of Linux systems so dramatically? I think I'm going to see what happens if I recompile the FC2 kernel with HZ=100 setting...this might take a while so don't hold your breath. <!--QuoteEnd--> </td></tr></table><div class='postcolor'> <!--QuoteEEnd-->
What's happened with this testing...
Did it work? not work?
Vadakill did this testing a while ago and it showed that 100 hz lowered CPU usage. Most new kernels have the option to change the kernel interrupt frequency in menuconfig now, so change it to 100 hz for server usage (as recommended). Also, removing desktop-friendly features such as preemption can also improve performance, but I personally haven't seen how much they lower performance as I always have preemption off.
Let me know if anyone has more detailed test results, as I'd be interested in seeing them.
The NSA servers run with pretty bad CPU even though they are windows boxes. They usually run somewhere from 30% to 35% cpu each. It's probly because of their size because they are much bigger than 16 man but when we ran a cs server of similer size it took almost 1% cpu full. <img src="style_emoticons/<#EMO_DIR#>/nerd-fix.gif" style="vertical-align:middle" emoid="::nerdy::" border="0" alt="nerd-fix.gif" />
<!--QuoteEnd--></div><!--QuoteEEnd-->
That's because CS doesn't have structues like offense chambers or turrets. 30-35% still isn't horrible, especially if it's a huge server.
Check this:
I have a box running linux with a pretty weak cpu. Anyways, when I run hlds natively on linux (Debian etch), I get 30% idle cpu usage with 6 bots.
<!--c1--><div class='codetop'>CODE</div><div class='codemain'><!--ec1-->CPU In Out Uptime Users FPS Players
33.25 3.16 2.89 1238 451 60.14<!--c2--></div><!--ec2-->
This is after I set the sys_ticrate to 60 for comparison (on windows, fps dont go higher than 60). Without this, the fps would be around 90-100 but the cpu usage is the same.
Now, guess what happens when I run the hlds within a Windows 2003 Server VMWare environment on the same box?
(yes! virtualized environment!)
->
<!--c1--><div class='codetop'>CODE</div><div class='codemain'><!--ec1-->CPU In Out Uptime Users FPS Players
1.55 10.71 10.52 16 17 64.42<!--c2--></div><!--ec2-->
Though I just checked the vmware cpu usage, and it is close to the other hlds process (that runs natively).
So, in conclusion: I think its only some sort of bug in the display of the actual cpu usage.
Though I still have a strange feeling about it..
Also, are you running plugins (metamod, amxx etc.) ?
In windows, you can get fps higher using -pingboost and I think you need to start Windows media player or something to enable higher kernel tickrate.
Btw, You'll get a better idea of the whole CPU usage using uptime to get average system load.
Fresh start on windows with 6 whichbots has 0% cpu in windows but the vmware reports to have a usage of 30-50% as well.
(repeat: Cpu usage inside the vmware in windows is 0-5%, vmware cpu usage in linux is 30% -> the win vm gets 30% cpu but does not really use it)
Thats why I think there is some sort of bug that causes linux to think that hlds is using a lot more cpu than it is actually using.
Nonsense, a load of 33, would mean 3300% cpu usage, or 33x time the load the cpu could handle, on my centOS box, the load on a ns server with running roughly 500-1000fps is around 3%, loaded i guess 20%. (16 slots).
On windows this is a bit less, but almost the same.