At 01:41 AM 2/19/2009, kERPLUNK wrote: >Hi, i have a serious problem with my players ping/lag. >I think, that comes from my server-side fps fluctuation. >Is there some way to make it really really stable?
No. >I tried a lot of things, like: > >- Changing sys_ticrate to 0, 100, 120, 333, 500, 1000, etc. > >- Recompile kernel with/without: >Dynamic Ticks This is bad and increases interrupt latency. >100/250/300/1000 HZ 1000hz increases the precision of things like sleep, but the error vs realtime increases. It also updates the interrupt frequency so interrupts fire more often. >Preemtible/No Forced Preemtion (Desktop/Server) >Machine Check Exeption Premption should also be set to "Low Latency" >Microcode Patch This just patches in microcode for <x,y,z> errata on cpus. >NUMA Numa isn't supported on anything other than x64 + Opteron systems. >High Resolution Timer Needed for high resolution nanosleep timers. >Multi-core scheduler >hpet The first just adds a few more ns of scheduler latency. HPET is nice to have and use. >real time clock >dma Doesn't affect anything. >(i think the problem comes from here.. kernel settings..) > >- Another kernel with rt patch (following this guide: >http://wiki.fragaholics.de/index.php/EN:Linux_Kernel_Optimization) RT Patches don't really eliminate problems with power management stuff. It's supposed to completely elminate scheduler latency and guarantee nanosleep/gettimeofday latency. >- Allocating process to specific core (taskset) Good to use. Prevents errant hardware interrupt latency >- Using nice/renice/chrt Nice really doesn't help out as much as chrt. >- Booting kernel with noapic noapm Turning off APIC is not recommended. Try disabling power management settings in your bios. It sounds like SMI latency, or speedstep. _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
