James, I have no doubt, you and many other on this list have far more linux knowledge than me. Myself, I learned the little I know by trial and error, and using "history" and man pages. If its a comparision of knowledge of linux os and its kernels, I am not competent enough to dispute findings with technical backup.
However, I have shown in the 5yrs of running my business, that I find or notice problems long before most techs do. I am even really good at figuring out what is wrong, though sometimes I am personally unable to fix it due to limited knowledge. As far as this kernel topic goes, I am confident that something is definitely wrong with hlds binaries. I ran the beta linux since the day it was out, and I am religious about watching cpu usage, fps, ping and in game performance. Almost fanatical about it. I do not believe a kernel that is a couple years out of date, with a VM that no longer exists in any kernels since 2.4.9 is the solution to the usage issues with steam hlds. As I posted, we tried 2.4.9 at 4 different hz variables. 100 default, and then 500,700,1000. I do not feel the cpu usage stats from 100hz was any more reliable than at 500/700 or 1000hz. I know that in game performance on 2.4.9 was no where near the performance of my 2.6.0-test-mm3 kernel. Not only by the pings, fps, but the smoothness of play. I also know from testing a gcc3 compiled Intel binary that there can be improvements to the hlds engine itself. Lower load, higher fps and lower usage. Nothing "great", but a marked difference. If I believed believed the information from cpustat/top/stats, I would have no fear to load up another three 20 player servers on the test box. But, the fact remains that regardless of kernel or operating system, a cpu can only handle so much usage. Prior to 1.6, I could run a 20 player server smoothly on a single 1ghz processor under my 2.6.0 kernel. On 1.6 hlds binary, I can't run a 12 player server on a 1ghz server. Now, the only difference between the now and then, is the binaries, and the changes to the game/maps. I would be more likely to believe an improvment in stats using a new kernel versus switching to an outdated one with a VM that no longer exists. 1% cpu I could never believe, even if the pope himself swore by the results. I will undertake one more test. Since testing on dual xeon's gave great results from stats point of view, but decreased performance from an in game point of view....I will take one of my dual p3 1ghz boxes with hardware that is 3yrs old. The same box that ran my two 14 player 24/7 awp map servers. In 1.5 and previous I ran both with low ping and smooth play..with max cpu usage on each processor at about 70%..while 1.6 usings over 95% for the same single 12 player awp server on 1ghz cpu. If I can run 2 14 player servers with the same low ping, smooth play and much reduced cpu...then I will consider your results. But if load on the server is the same, and in game play is as bad with the 2 servers on, and cpustat/top/stats show low cpu, then for me it is proven beyond reasonable doubt that cpu reporting for the 2.4.9 kernel by ANY program is faulty. I personally think that is why it was probably ripped out of the kernel suddenly. They had all these beautiful stats on cpu, which in the end turned out to be incorrect. The stats could quite easily be reading the proc information correctly, but that doesn't mean the kernel VM is showing or reporting that information correctly. I will let you know. dan ----- Original Message ----- From: "James Couzens" <[EMAIL PROTECTED]> To: <[EMAIL PROTECTED]> Sent: Thursday, September 18, 2003 10:38 AM Subject: Re: [hlds_linux] Crazy Usage > Daniel, > > Statistics information, IE "cpu jitters" are constantly updated in a single > place in linux, and that is proc. Thats where the HLDS "stats" function > gets it from, and thats where ps, top, and any other utilitiy in linux that > reports CPU % will also obtain it from this directory. This is where > CpuStat gets it from also. > > Provided you do not bugger with the kernel, the information found in this > directory its EXTREMELY accurate, and I can only think that the sheer > disbelief I am getting from most of you on this list is due to a lack of > knowledge regarding how the linux operating system functions. > > Numbers, in this case, do not lie. If you start to bugger with HZ then yes > you are going to get some bizzare and unbelievable results. Provided there > is no interference from some modification such as this the information in > /proc will be true. > > To prove my point to you all, I now have 4 16 player Steam CS 1.6 servers > running on this machine. I shall post results later today. > > James > > ----- Original Message ----- > From: "Daniel Stroven" <[EMAIL PROTECTED]> > To: <[EMAIL PROTECTED]> > Sent: Thursday, September 18, 2003 7:19 AM > Subject: Re: [hlds_linux] Crazy Usage > > > > James, > > > > Can you give us some "rcon stats" results on these games? When you the > > servers are close to full? > > > > I have cpustat installed when I tested out the 2.4.9 kernel. Pretty neat, > > but I would like to see some info from rcon stats if possible from you. > > > > Also if you can show some load stats from top for these servers. thanks > > > > dan > > ----- Original Message ----- > > From: "James Couzens" <[EMAIL PROTECTED]> > > To: <[EMAIL PROTECTED]> > > Sent: Thursday, September 18, 2003 3:59 AM > > Subject: Re: [hlds_linux] Crazy Usage > > > > > > > Daniel, > > > > > > Believe me, it can, and it does. Taking the server to larger maps does > > > yield cpu above 1%, please note that I was testing the smallest map, > > > de_dust. Previously posted on this list were some screenshots: > > > > > > http://prodigy.redphive.org/images/de_airstrip.jpg > > > http://prodigy.redphive.org/images/de_aztec.jpg > > > http://prodigy.redphive.org/images/de_aztec2.jpg > > > http://prodigy.redphive.org/images/de_dust.jpg > > > http://prodigy.redphive.org/images/de_dust2.jpg > > > > > > The cpu use is very REAL, and even for YES 30 slots. I have seen > > identical > > > results in windows. As Eric kindly pointed out previously, 30 slots is > a > > > relative term. A 30 slot server, with all slots bound to the same > server > > > would use an exponentially larger amount of cpu than two servers running > > 15. > > > > > > Cheers, > > > > > > James > > > > > > ----- Original Message ----- > > > From: "Daniel Stroven" <[EMAIL PROTECTED]> > > > To: <[EMAIL PROTECTED]> > > > Sent: Thursday, September 18, 2003 12:34 AM > > > Subject: Re: [hlds_linux] Crazy Usage > > > > > > > > > > Server FPS I have seen over 500. 100 is what i would ideally want. > 50 > > > fps > > > > or lower is not what I want. > > > > > > > > To me its simple. Regardless of kernel or regardless of OS, a cpu can > > > only > > > > handle so much. A cpu is not going to just use 1 percent for near 30 > > > slots, > > > > unless its some super duper chip. There is way to much going on for > it > > to > > > > use 1%, that is not even realistic to think so. > > > > > > > > Put that same chip on windows servers which seem to utilize cpu better > > > than > > > > linux currently, and you will not see 1% usage. In fact I would say > its > > > > down right physically impossible with the current chips..even 3ghz > chip. > > > > > > > > If it works for you great, for me the performance simply wasn't there. > > > > ----- Original Message ----- > > > > From: "James Couzens" <[EMAIL PROTECTED]> > > > > To: <[EMAIL PROTECTED]> > > > > Sent: Thursday, September 18, 2003 2:15 AM > > > > Subject: [hlds_linux] Crazy Usage > > > > > > > > > > > > > Jeremy, > > > > > > > > > > I assure you, it is not incorrect. Its very correct. Kernel > timings > > in > > > > > kernel-2.4.21 which I have also tested against, use identical kernel > > > > > sleeping habbits: > > > > > > > > > > LINUX 2.4.21 : > > > > > > > > > > [EMAIL PROTECTED] asm-i386 # cat > > > /usr/src/linux-2.4.21/include/asm-i386/param.h > > > > > #ifndef _ASMi386_PARAM_H > > > > > #define _ASMi386_PARAM_H > > > > > > > > > > #ifndef HZ > > > > > #define HZ 100 > > > > > #endif > > > > > > > > > > #define EXEC_PAGESIZE 4096 > > > > > > > > > > #ifndef NGROUPS > > > > > #define NGROUPS 32 > > > > > #endif > > > > > > > > > > #ifndef NOGROUP > > > > > #define NOGROUP (-1) > > > > > #endif > > > > > > > > > > #define MAXHOSTNAMELEN 64 /* max length of hostname */ > > > > > > > > > > #ifdef __KERNEL__ > > > > > # define CLOCKS_PER_SEC 100 /* frequency at which times() counts > > */ > > > > > #endif > > > > > > > > > > #endif > > > > > > > > > > LINUX 2.4.9: > > > > > > > > > > [EMAIL PROTECTED] asm-i386 # cat > > > /usr/src/linux-2.4.9/include/asm-i386/param.h > > > > > #ifndef _ASMi386_PARAM_H > > > > > #define _ASMi386_PARAM_H > > > > > > > > > > #ifndef HZ > > > > > #define HZ 100 > > > > > #ifdef __KERNEL__ > > > > > #if HZ == 100 > > > > > /* X86 is defined to provide userspace with a world where HZ=100 > > > > > We have to do this, (x*const)/const2 isnt optimised out because > its > > > not > > > > > a null operation as it might overflow.. */ > > > > > #define hz_to_std(a) (a) > > > > > #else > > > > > #define hz_to_std(a) (((a)*HZ)/100) > > > > > #endif > > > > > #endif > > > > > #endif > > > > > > > > > > #define EXEC_PAGESIZE 4096 > > > > > > > > > > #ifndef NGROUPS > > > > > #define NGROUPS 32 > > > > > #endif > > > > > > > > > > #ifndef NOGROUP > > > > > #define NOGROUP (-1) > > > > > #endif > > > > > > > > > > #define MAXHOSTNAMELEN 64 /* max length of hostname */ > > > > > > > > > > #ifdef __KERNEL__ > > > > > # define CLOCKS_PER_SEC 100 /* frequency at which times() counts > > */ > > > > > #endif > > > > > > > > > > #endif > > > > > > > > > > As you can clearly see, HZ is still 100. > > > > > > > > > > So, please explain to me how I am being lied to? I use 2.4.21 I see > > > > obscene > > > > > usage breaking to almost full capacity of my chip. In addition to > not > > > > only > > > > > SAYING its using that many cycles, it TRULY is, FPS drops, working > on > > > the > > > > > machine becomes sluggish etc... Yet when I use a 2.4.9 kernel, I > see > > a > > > > > remarkable difference, in addition to INCREDIBLY low cpu use, its > > > > accurate, > > > > > since the FPS is high, and the server is extremely responsive. > > > > > > > > > > I am not using top to discover cpu use either, and thats even a moot > > > point > > > > > since top agrees fully with the cpu use I arrive at, which I > manually > > > > > calculate by using the proc filesystem. > > > > > > > > > > Please explain to me the logic you are using to deduce that my > results > > > are > > > > > incorrect? FPS does not lie. If the server can do 50 FPS, which is > > its > > > > > MAX, then I think i'm on the right track here. > > > > > > > > > > I am telling you, and PROVING to you, that using this kernel, with > > > DEFAULT > > > > > kernel timings results in performance vastly superior to any kernel > > > > released > > > > > after. > > > > > > > > > > James > > > > > > > > > > ----- Original Message ----- > > > > > From: <[EMAIL PROTECTED]> > > > > > To: <[EMAIL PROTECTED]> > > > > > Sent: Wednesday, September 17, 2003 7:43 PM > > > > > Subject: RE: [hlds_linux] Crazy Usage > > > > > > > > > > > > > > > > >Here is an example: > > > > > > > > > > > > > >AMD XP2600 1GB PC2700 DDR333: > > > > > > >[7:19:pm] -r5-cs3- [CS3] Statistics: CPU: 1% FPS: 50 Players: > > 15/17 > > > > Map: > > > > > > >cs_assault Uptime: 757 m > > > > > > >[7:19:pm] -r5-cs2- [CS2] Statistics: CPU: 1% FPS: 50 Players: > > 14/17 > > > > Map: > > > > > > >de_chateau Uptime: 922 m > > > > > > > > > > > > > >Those two servers are on the same machine, and the cpu reported, > is > > > the > > > > > > >TOTAL CPU use for the ENTIRE server, so its 1% combined. So > there > > > are > > > > 29 > > > > > > >players effectively doing only 1% cpu. And as you can see, each > > > server > > > > > is> > > > > > > >still clearly getting 50FPS which is excellent also. > > > > > > > > > > > > > >VERY IMPRESSIVE. > > > > > > > > > > > > That's not impresssive, it's just incorrect. There is no possible > > way > > > > that > > > > > > your server has 1% usage with 30 active players. I'm sorry but top > > is > > > > VERY > > > > > > incorrect. > > > > > > > > > > > > Jeremy > > > > > > > > > > > > > > > > > > > > > > > > _______________________________________________ > > > > > > To unsubscribe, edit your list preferences, or view the list > > archives, > > > > > please visit: > > > > > > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > > > > > > > > > > > > > > > > > > > > > > > > > > _______________________________________________ > > > > > To unsubscribe, edit your list preferences, or view the list > archives, > > > > please visit: > > > > > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > > > > > > > > > > > > > > > > > > > > > _______________________________________________ > > > > To unsubscribe, edit your list preferences, or view the list archives, > > > please visit: > > > > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > > > > > > > > > > > > > > > > > > > > _______________________________________________ > > > To unsubscribe, edit your list preferences, or view the list archives, > > please visit: > > > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > > > > > > > > > > > _______________________________________________ > > To unsubscribe, edit your list preferences, or view the list archives, > please visit: > > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > > > > > > > > _______________________________________________ > To unsubscribe, edit your list preferences, or view the list archives, please visit: > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
