That is a bug; the initial sleep is undershooting due to an integer rounding error. Will be addressed in another update.
-henry -----Original Message----- From: hlds-boun...@list.valvesoftware.com [mailto:hlds-boun...@list.valvesoftware.com] On Behalf Of John Sent: Tuesday, October 11, 2011 11:21 AM To: hlds@list.valvesoftware.com Subject: Re: [hlds] Mandatory TF2 update coming Wednesday On 10/11/2011 2:25 AM, AnAkIn . wrote: > Can anyone from Valve comment about host_timer_spin_ms? > > This seem to make the FPS more consistent when enabled, it always stay > at 66. When it's disabled, it's mostly between 60-70. It's described in the server as "Use CPU busy-loop for improved timer precision (dedicated only)". Without it, it looks like the server runs nanosleep to wait 13000000 ns or 14000000 ns (at least, in an empty server -- with a busy one, I'd expect this to be less), then submits one or more 1000000 ns sleeps in order to align with its desired next tick time. With it enabled, it appears that the server just submits one nanosleep and then spins the rest of the time for its alignment. Modern Linux kernels are quite precise when it comes to timers, at least when slack is disabled (using chrt or other means). Are there plans to take advantage of this with more precise sleeps from the get-go? -John _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds
