[ntp:questions] Re: SUSE OSS 10 NTPD Clock Drift

Richard B. Gilbert rgilbert88 at comcast.net
Fri Jan 20 00:58:08 UTC 2006


garskof wrote:

>I had the same clock drift when it was set up with WinDoze, so I doubt
>it is SUSE that is at fault.
>
>  
>
WinDoze exhibits the same problem with lost clock interrupts.  The 
difference is that there is nothing you can do about WinDoze except run 
a better O/S!  The problem results from interrupts being masked or 
disabled for two or more consecutive clock "ticks".   Only the first 
"tick" registers, the rest are lost.  This typically happens during 
heavy disk activity.  Runing the multi-media timers can also gum things 
up royally but there is a fix or workaround for that.  If your clock 
gains rather than loses time, you needn't worry about lost interrupts.  
If the machine drifts badly when idle, the problem is probably not lost 
interrupts.

>So ntp will only correct if the drift is less then some value? On
>startup it corrects no matter how off the clock is, but after that it
>never corrects again. I assumed there was a way to increase the number
>of times per day ntp checks with its servers. I guess a cron of ntpdate
>evey hour or so is my own choice. Sigh.
>
>  
>
If you start NTPD with the -g option it will set the clock on a one time 
basis.  If the clock is drifting at a rate greater than or equal to 500 
PPM,  there is nothing NTPD can do about it; the limitation is built 
into the O/S kernel I believe.

While computer clocks generally do not win any prizes for accuracy, very 
few are off by more than 500 PPM.  A small sampling in my household 
shows three Sun Ultra 10 workstations with frequency errors of less than 
10 PPM and a DEC Alphastation 200/233 with an error of less than 30 
PPM.  I think that with a sample large enough to be statistically 
significant, the majority would fall within +/- 100 PPM.  Your machine 
is way the hell and gone beyond 3 sigmas!




More information about the questions mailing list