On Sun, 30 Sep 2007, Adrian von Bidder wrote: > On Friday 28 September 2007 21.25:37 Todd wrote: > > I strongly suspect that the new tickless timer system has not improved > matters. But note that this is pure speculation - anyonoe did any testing? > (OTOH I have the impression that tickless has definitiely improved desktop > snappiness here, and I'm eagerly awaiting CFS - and to me, this is more > important than sub-microsecond timekeeping :-)
I run Linux 2.6.20.3 #1 SMP PREEMPT on my main ntp server and my kernel time stays between +- 5 ms of the servers I connect to to get time. I found that the quality of the hardware clock seems to be the most important thing to consider. I have machines that needs to poll 64 my main ntp server via the LAN in order for them to keep accurate time ;-) An interesting thing to consider is that my main ntp server also runs 5 instance of vmware-server and typically runs with more than 250 processes, not counting the within vmware virtual processes. So there is quite a bit of context switching going on on that machine. I also run about 2-5 instances of the Xvnc graphical interface within the vmware servers. Like you said some work have been done lately on linux kernel timer. I compiled 2.6.22.3 #4 SMP PREEMPT (latest is 2.6.22.9) on one machine and it seems to work fine as well as long as you have got a good hardware clock in the computer. There was a discussion on slashdot about CFS lately, you might want to review the comments. Basically, the complexity of the first draft is questioned, even by the author of CFS. http://linux.slashdot.org/article.pl?sid=07/09/01/1853228 NOTE the poll values ! remote refid st t when poll reach delay offset jitter ============================================================================== +128.4.1.1 .PPS. 1 u 52m 137m 177 24.523 2.000 1.163 *132.246.168.9 .PPS. 1 u 52m 68m 377 15.902 1.980 0.288 -64.230.242.45 64.230.242.33 4 u 329 1024 377 24.796 3.221 0.069 +69.156.254.2 132.246.168.164 3 u 350 1024 377 27.337 2.277 1.214 -69.156.254.38 64.26.173.192 4 u 1322 68m 377 20.622 2.760 1.212 +64.230.159.74 .GPS. 1 u 308 1024 377 29.891 -3.194 5.222 top - 20:44:15 up 17:07, 1 user, load average: 0.87, 0.76, 0.62 Tasks: 265 total, 2 running, 257 sleeping, 6 stopped, 0 zombie Cpu(s): 10.1% us, 27.3% sy, 0.0% ni, 54.2% id, 4.0% wa, 1.5% hi, 2.8% si Mem: 4131064k total, 3989456k used, 141608k free, 1560k buffers Swap: 1999800k total, 0k used, 1999800k free, 3096380k cached > > cheers > -- vbi > > -- > "It blows my mind that you can't get 3.5 ounces of toothpaste on a > plane," he said, "yet somebody can sneak on a plane and take a nap." > -- http://www.newsobserver.com/102/story/523482.html > Louis http://blogtech.oc9.com _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
