[ntp:questions] serialPPS+gpsd+ntpd large offset & jitter

Mike S mikes at flatsurface.com
Fri Jan 14 01:38:09 UTC 2011


At 08:08 PM 1/13/2011, Chris Albertson wrote...
>On Thu, Jan 13, 2011 at 3:59 PM, Mike S <mikes at flatsurface.com> wrote:
>de-facto' timer hardware for IA-PCs."
> >
> > If trying to improve the system time stability of a PC, the first 
> thing to
> > look for is a 14.31818 MHz crystal
>
>I just now did a parts search at Digikey for the best posable 14.31818
>MHz crystal to put on a PC motherboard.

One note, having just gone through this: if using Linux with NTP on a 
modern PC motherboard, it's likely that the kernel will want to use the 
HPET for system timing. But, to do so it calibrates the HPET during 
boot (against the RTC or maybe the 8259, I didn't dig deep enough). It 
does it in a hurry, and the calibration is NOT consistent. I saw NTP 
reporting corrections (drift) varying more than 60 PPM between boots. 
That behavior would not change regardless of the source of 14.31818 
MHz, although you would have better stability as long as you didn't 
reboot.

I was able to force the kernel to use the 8259 by adding 
"clocksource=acpi_pm"  to the boot options. The available clock sources 
can be found with "cat 
/sys/devices/system/clocksource/clocksource0/available_clocksource". 
Now, the NTP PPM correction is consistent between boots.





More information about the questions mailing list