[ntp:questions] Re: correct NTP-configuration for combined Microsoft ADS- and UNIX/Linux-environment

Ry malayter at gmail.com
Fri Apr 7 03:45:32 UTC 2006


Richard B. Gilbert wrote:
> It may also be worth noting that Windows does not use NTP software;
> Microsoft supplies a broken implementation of SNTP called W32TIME.  NTP
> software is available for Windows if you wish to download and install it

Windows 2003 Standard Server with Service Pack 1 seems to have a much
better "almost real" NTP service. It's not just SNTP anymore. I've been
testing it compared with NTPv4 for a few weeks now, but I haven't had a
chance to do a full write up yet.

Here's a quick outline of the results, when Win2003sp1 or R2 is
compared with NTPv4 running against the same set of Stratum-1s from the
same network:
   - it keeps reasonable average offsets with NTPv4 at stratum 2 for at
least a week (my test length)
   - it selects its reference clocks roughly the same as NTPv4, usually
hopping to a new reference server a few minutes ahead or behind NTPv4
as network conditions warrant
   - it polls servers at similar intervals to NTPv4
   - it advertises stratum and reference ID correctly (tested with
stratum 1 & 2 servers)

The big caveat: Windows 2003 SP1 time service only advertises -6
precision in its packets, and unfortunately, this claimed precision
seems about right. On the graphs I've made so far, it looks like it
steps the clock in ~10 ms increments rather infrequently. The result is
a far higher jitter than you would see with NTPv4, although the offsets
are almost always within 20 ms of UTC, and a moving average is within a
couple of ms of NTPv4's offset moving average.

Basically, I'd say that w32time in Windows 2003 SP1 and R2 is
definitely more RFC-compliant than the SNTP service in previous
Microsoft offerings. It is usable for many applications, but still of
course not nearly as good as the full NTPv4.

Regards,
  Ryan




More information about the questions mailing list