Custom Kernels for the most part are complete snake oil (imo). I've had locking issues with CK's patchset, BFS does not function that well with high load systems (4, 8, 24 thread systems). BFS seems to work fine on my FileServer with BFQ though... Your best bet with SRCDS is just go vanilla and be smart with your options. Micro Kernels will not help you with performance. Apparently the grass is greener on Windows (Guys pulling off 50+ player servers with a Q9550), so that's always something to try if you're unhappy with Linux (If it is better, do let us know).
Kyle. On Fri, Jul 15, 2011 at 11:56 PM, Nephyrin Zey <[email protected]> wrote: > Kernels can affect performance in other ways, i personally use a vanilla > 2.6.39 kernel with high resolution timers and get great performance out of > it. > > On 07/15/2011 11:40 PM, William Balkcom wrote: >> >> With that being said would having a custom kernel no longer be required to >> achieve the best performance? >> >> William Balkcom >> >> On Jul 16, 2011, at 2:29 AM, Nephyrin Zey<[email protected]> wrote: >> >>> A 'frame' in the orangebox engine does this, server side: >>> >>> while (sleep(1ms)) { while (ShouldRunATick()) { RunATick() } } >>> >>> You can refer to the disassembly or even the old HL2 leak if you don't >>> believe me. Everyone who claims otherwise is just wrong. >>> >>> So the only advantage FPS gives is how often the idle engine wakes up to >>> check for the next tick. It's worth noting that sleep(1ms) could sleep 1ms, >>> 5ms, or even 15ms depending on the OS, kernel timer method, whether high >>> resolution sleeps are supported, etc.. >>> >>> So a higher FPS can give greater accuracy in tick *timing*, as each tick >>> would be starting closer to its ideal (at 66tick, a tick would happen >>> exactly every 15ms). If your FPS is 100, this means wakeups can happen +/- >>> 10ms, which could be almost an entire tick late. You'd still get roughly >>> 66tick, but with lower tick accuracy, some might argue you get worse hit >>> registration as you're less in sync with clients. However, most people play >>> at pings of 50ms or higher. Is 9ms really that significant? >>> >>> You can pull up net_graph 4 and watch the 'var: ' value, which I'm pretty >>> sure is the variation in wakeup times in milliseconds, hence the max lag in >>> milliseconds that could be alleviated by higher FPS. >>> >>> TL;DR: If you're getting solid 66 updates/second and your var is under >>> 10, nobody with less than 60ms ping should be complaining about "hit >>> registration". >>> >>> - Neph >>> >>> On 07/15/2011 11:04 PM, William Balkcom wrote: >>>> >>>> With valve now limiting FPS in orangebox to 500FPS, is it recommended >>>> just to run the server at the default now? Does this same concept apply for >>>> the HL1 engine and non orangebox games? Are you truly gaining anything by >>>> running at 500FPS? > > > _______________________________________________ > 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
