[ntp:questions] NTP Bug 2328 - Vista/Win7 time keeping inaccurate and erratic

Martin Burnicki martin.burnicki at meinberg.de
Wed Nov 27 17:48:10 UTC 2013


David Taylor wrote:
> Thanks, Martin, that's most interesting.  I can get nothing like that
> performance from my Win-7 systems, although they are all connected via
> Wi-Fi which doesn't help.  I've just changed the order in my MRTG
> presentation to show XP, versus  Vista, Win-7, and Win-8.  As you'll
> see, there's nothing like the smooth graph you got from Win-7, although
> these are offsets reported by NTP rather than measurements against a PCI
> reference like yours. However, the offset scale is different, as I have
> a range of +/- 3 milliseconds, and your is +/- 30 milliseconds, so
> perhaps your noise would be the same as mine once the graph was scaled.

OK, here's another graph from the same data, with +/- 3 ms time offset 
range, but a longer time:
http://support.ntp.org/people/burnicki/windows/bug2328_workaround_fine.pdf

My setup is using a Linux NTP server on a local LAN, connected via 
cable, so I'd expect this yields a better accuracy than a WiFi connection.

Also, the time offset is not the offset reported by ntpq -p, but the 
offset of the Windows system time (disciplined by ntpd) compared to the 
time from a built-in GPS PCI card.

> Maybe my Win-7 systems are the unaffected ones?

Possibly.

We have also seen setups (under Windows) where time synchronization is 
very smooth versus systems where time synchronization steps back and 
forth by 1 ms or so. My *assumption* is that a possible reason could be 
the sign and/or the absolute mean frequency offset, which may cause 
rounding errors or so in the Windows kernel when converting 1 ms ticks 
to nominal tick adjustment count which is 15.6 ms, or vice versa.

> Did the one you tested have any NTP environment variables set, to
> enable/disable interpolation etc. etc.?

No, I just used the default installation, and replaced the binaries later.

Martin
-- 
Martin Burnicki

Meinberg Funkuhren
Bad Pyrmont
Germany



More information about the questions mailing list