[ntp:questions] Re: NTP client on Windows platform provides lessaccurate results then on the UNIX or Linux. Why?

Karel Sandler sandler at ujf.cas.cz
Thu Mar 30 16:10:54 UTC 2006


"David J Taylor" <david-taylor at blueyonder.co.not-this-bit.nor-this-part.uk> 
wrote

> Karel Sandler wrote:
>> I'm using  ntp-4.2.0b at 1.1436mbg-xmas-o-win32 from Meinberg on a W2k3
>> system. Some stats an offset distribution for this system you can see
>> at http://lx.ujf.cas.cz/mrtg/ntp/. My early experience with ntpd on a
>> windows systems was rather bad but this version of ntpd I consider
>> good.
>> Karel Sandler
>
> Karel,
>
>  http://lx.ujf.cas.cz/ntp-lx/lx-off.html
>
> shows a series of one millisecond steps at about roughly 1 hour intervals. 
> Any idea why?
>
> (My systems may do this as well, but my plots are only one tenth of the 
> resolution of yours, I allow for +/- 100ms which was perhaps more 
> appropriate when I relied on Internet ntp sources).

I don't know the cause of this behaviour. The ntpd obviously uses a brute 
force approach to steer this Linux system. Maybe the hardware is not good 
enough. I have already shifted the original 155 ppm value to 55 ppm but 
without any impact. Another possible cause may be clock-hopping between two 
sets of peers but any subsequent change in the number of peers between four 
to seven has no effect as well and none of my peers allows me to simly 
prefer it. Hope the situation will improve after setting up one's own 
stratum 1. I think there are at least two other fields to play with namely 
ACPI and ntpd tos directive but it remains to be seen. There isn't much room 
to experiment with this public timeserver and our webserver. Any advice 
would be appreciated.

Karel





More information about the questions mailing list