Re: [hlds_linux] Fun with linux kernel configs
I looked into it: calling gettimeofday() in recent glibc versions uses the kernel's vgettimeofday() call, and thus isn't really a problem. See my other email about sleep() usage though. - Neph On Tue, May 20, 2008 at 1:20 AM, Thomas Hjorth [EMAIL PROTECTED] wrote: On Mon, 2008-05-19 at 19:02 -0700, Nephyrin Zey wrote: calling gettimeofday() is somewhat sickening - I might try and put together a plugin that re-implements it in userland via a timer and re-routes the call. That would be great, though I havent cracked open orangebox yet, but im guessing you would probably make it work for srcds as well? The server's play quality is still steady, i get ~100fps in heavy action at 66tic, which despite what some people may say is completely impossible to differentiate from 1000fps. Im guess you are talking about serverside FPS? 100 FPS seems low to me. All my servers are 500 max. and drops in steps like 333, 250, 200, 166, 125 and similar, but I still have perfect 100tickrate up until about 16 players, after that the engine takes a lot more hits and the tickrates degrades a lot at times, but the FPS is doing about the same FPS drops... especially with deathmatch plugins and stuff. Im working on some monitoringscripts to make it easier for me to spot where and when issues accrue. For the record I use a stripped Debian etch, 32bit, non-tickless, 1000hz, preemptive and have stripped just about everything from the kernel not need on the particular gameserver. Regards, Thomas @ DSRack.dk Though it's still pretty crap performance overall. I'll see what else I can find - Neph On Mon, May 19, 2008 at 12:33 PM, Kveri [EMAIL PROTECTED] wrote: Hello Have you noticed any quality changes? I never tested tickless system, but I'm using HPET and 1000Hz timer, but kernel 2.6.24-rc3 and srcds_i686 not i486. This can be good thing, I actually think about that, that about 90% of srcds calls are gettimeofday(). Good work and keep your posts comming. ___ 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
Re: [hlds_linux] Fun with linux kernel configs
On Mon, 2008-05-19 at 19:02 -0700, Nephyrin Zey wrote: calling gettimeofday() is somewhat sickening - I might try and put together a plugin that re-implements it in userland via a timer and re-routes the call. That would be great, though I havent cracked open orangebox yet, but im guessing you would probably make it work for srcds as well? The server's play quality is still steady, i get ~100fps in heavy action at 66tic, which despite what some people may say is completely impossible to differentiate from 1000fps. Im guess you are talking about serverside FPS? 100 FPS seems low to me. All my servers are 500 max. and drops in steps like 333, 250, 200, 166, 125 and similar, but I still have perfect 100tickrate up until about 16 players, after that the engine takes a lot more hits and the tickrates degrades a lot at times, but the FPS is doing about the same FPS drops... especially with deathmatch plugins and stuff. Im working on some monitoringscripts to make it easier for me to spot where and when issues accrue. For the record I use a stripped Debian etch, 32bit, non-tickless, 1000hz, preemptive and have stripped just about everything from the kernel not need on the particular gameserver. Regards, Thomas @ DSRack.dk Though it's still pretty crap performance overall. I'll see what else I can find - Neph On Mon, May 19, 2008 at 12:33 PM, Kveri [EMAIL PROTECTED] wrote: Hello Have you noticed any quality changes? I never tested tickless system, but I'm using HPET and 1000Hz timer, but kernel 2.6.24-rc3 and srcds_i686 not i486. This can be good thing, I actually think about that, that about 90% of srcds calls are gettimeofday(). Good work and keep your posts comming. ___ 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
Re: [hlds_linux] Fun with linux kernel configs
Hello Have you noticed any quality changes? I never tested tickless system, but I'm using HPET and 1000Hz timer, but kernel 2.6.24-rc3 and srcds_i686 not i486. This can be good thing, I actually think about that, that about 90% of srcds calls are gettimeofday(). Good work and keep your posts comming. [EMAIL PROTECTED] wrote / napĂsal(a): Send hlds_linux mailing list submissions to hlds_linux@list.valvesoftware.com To subscribe or unsubscribe via the World Wide Web, visit http://list.valvesoftware.com/mailman/listinfo/hlds_linux or, via email, send a message with subject or body 'help' to [EMAIL PROTECTED] You can reach the person managing the list at [EMAIL PROTECTED] When replying, please edit your Subject line so it is more specific than Re: Contents of hlds_linux digest... Today's Topics: 1. Fun with linux kernel configs (Nephyrin Zey) 2. Re: HLSW Banlist Broken (GZ Coldorak) 3. Re: HLSW Banlist Broken (tsuehpsyde) 4. Re: HLSW Banlist Broken (GZ Coldorak) -- Message: 1 Date: Sun, 18 May 2008 22:10:22 -0700 From: Nephyrin Zey [EMAIL PROTECTED] Subject: [hlds_linux] Fun with linux kernel configs To: Half-Life dedicated Linux server mailing list hlds_linux@list.valvesoftware.com Message-ID: [EMAIL PROTECTED] Content-Type: text/plain; charset=UTF-8 I've been periodically switching up my linux kernel config in an attempt to figure out what our friendly srcds software enjoys, and recently ran into something fun with this config: 2.6.25.4-rt1 kernel (realtime patchset), full RT-mode with all the fun RCU options on, non-tickless, 1000hz and all fun RTC options on What happens is srcds_i486 uses 0-10% CPU when loaded up with 32 players. It was using 80-95% previously. The catch is 'sirq-timer/0' and 'sirq-timer/1' (for each cpu) bloats up to ~70% cpu usage. Net CPU usage is down a good 10-20% overall, as compared to 2.6.25.3 non-rt with the same settings. Has anyone else really messed with the RT/HPET/Etc. settings in depth on the more recent .23+ kernels? And has anyone here tried re-implementing gettimeofday() in a plugin/preload to remove the heavy overhead calling a (non-virtual) syscall a thousand times/second can have on srcds performance? Not that I don't trust valve's programming decisions... Kernel config in question: http://nemu.nephyrin.net/kconfig_20080501801.txt - Neph -- Message: 2 Date: Mon, 19 May 2008 07:56:03 +0200 From: GZ Coldorak [EMAIL PROTECTED] Subject: Re: [hlds_linux] HLSW Banlist Broken To: Half-Life dedicated Linux server mailing list hlds_linux@list.valvesoftware.com Message-ID: [EMAIL PROTECTED] Content-Type: text/plain; charset=ISO-8859-1 About bans, do timed bans really work? For example, if I want to ban someone for one week, how do I do that? I've tried banid 10080 steamid writeid but then the number never decreases Thanks Cold' On Sun, May 18, 2008 at 7:55 PM, Guy Watkins [EMAIL PROTECTED] wrote: Have you contacted the HLSW people? They would know what is wrong before anyone here. If there a HLSW forum? They may have a fix already. Or is that an unsupported piece of software? } -Original Message- } From: [EMAIL PROTECTED] [mailto:hlds_linux- } [EMAIL PROTECTED] On Behalf Of tsuehpsyde } Sent: Sunday, May 18, 2008 1:01 PM } To: Half-Life dedicated Linux server mailing list } Subject: Re: [hlds_linux] HLSW Banlist Broken } } My main concern is that the Linux SourceDS binaries now have a restriction } on how much data they can send back from an RCON request, as HLSW was } working fine before the update. Again, it's possible that HLSW is somehow } incompatible now, but it's been working for four years now, so I'm } guessing } it's due to the RCON *fix* making it more restrictive on the data it sends } via RCON. I could be wrong, but I'm just trying to figure it out. } } -tsuehpsyde } } On Sun, May 18, 2008 at 11:02 AM, Mark Chandler [EMAIL PROTECTED] } wrote: } } Maybe hlsw is having problems. Need to take it up with who ever wrote } it. } } -Original Message- } From: [EMAIL PROTECTED] } [mailto:[EMAIL PROTECTED] On Behalf Of } tsuehpsyde } Sent: Sunday, May 18, 2008 11:23 PM } To: Half-Life dedicated Linux server mailing list } Subject: Re: [hlds_linux] HLSW Banlist Broken } } From 1.1.6: } } 10:21:35 listid } 10:21:36 L 05/18/2008 - 11:20:56: rcon from 24.x.x.x:3014: command } listid } 10:21:36 HLSW NOTE: Disconnected } } It hangs for a moment, then reconnects and runs status/starts watching } the } users in the server. Maybe our banlist has become too large? Our current } list is 340K in size. } } -tsuehpsyde } } On Sun, May 18, 2008 at 5:41 AM, ics [EMAIL PROTECTED] wrote: } } HLSW1.1.6 works but you can always execute
[hlds_linux] Fun with linux kernel configs
I've been periodically switching up my linux kernel config in an attempt to figure out what our friendly srcds software enjoys, and recently ran into something fun with this config: 2.6.25.4-rt1 kernel (realtime patchset), full RT-mode with all the fun RCU options on, non-tickless, 1000hz and all fun RTC options on What happens is srcds_i486 uses 0-10% CPU when loaded up with 32 players. It was using 80-95% previously. The catch is 'sirq-timer/0' and 'sirq-timer/1' (for each cpu) bloats up to ~70% cpu usage. Net CPU usage is down a good 10-20% overall, as compared to 2.6.25.3 non-rt with the same settings. Has anyone else really messed with the RT/HPET/Etc. settings in depth on the more recent .23+ kernels? And has anyone here tried re-implementing gettimeofday() in a plugin/preload to remove the heavy overhead calling a (non-virtual) syscall a thousand times/second can have on srcds performance? Not that I don't trust valve's programming decisions... Kernel config in question: http://nemu.nephyrin.net/kconfig_20080501801.txt - Neph ___ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux