[ntp:questions] NTP Performance on WINNT
Richard B. Gilbert
rgilbert88 at comcast.net
Mon Jul 20 15:49:34 UTC 2009
> On Jul 20, 10:36 pm, "David J Taylor" <david-tay... at blueyonder.not-
> this-part.nor-this.co.uk.invalid> wrote:
>> paul wrote:
>>> Hi All, I'm new to NTP, glad to meet you here.
>>> I did some experiments to test NTP performance on WINNT. In an
>>> isolated network, two machines are inter connnected with a switcher.
>>> Machine A is configure as a stratum 12 NTP server, using lcl as
>>> reference clock; machine B is sychronized to machine A. Then, 'ntpq -
>>> p' is performed on machine B every 16 seconds. The output of ntpq is
>>> cooked by a python script and offset field is recorded to log file.
>>> After about 2 hours, I check the log file and found the offset value
>>> is 4 ~ 8 ms. Is this normal? I mean if things are properly done, how
>>> small the offset I can expect? Thanks.
>>> The OS is Windows XP Pro SP3 and NTP software is 4.2.4-p7 from
>> For comparison, I have the performance of a mixture of Windows systems
>> running a mixture of NTP versions plotter here:
>> The Windows XP system synced to a local stratum-1 clock is Narvik, showing
>> an offset of within 1.5ms, but that is with a poll interval clamped to
>> 64s. The best is the system Feenix with a local GPS serial and PPS
>> reference clock (but no temperature control) where the offset is within
>> about 0.2ms, and the worst the Windows Vista system Gemini which has a
>> DVB/USB data reception system capturing about 20GB/day, but clobbering the
>> offset with transient peaks of at least 50ms.
>> "How long is a piece of string...." comes to mind!
> Thank you, David.
> In my situation, no GPS is availiable. So can I expect better
> performance when GPS is used as reference clock, or when a stratum-1
> NTP server is added to the network?
Assuming a GPS *TIMING* receiver, and that it can be installed with the
antenna having a good view of MOST of the sky, you can expect results
that are *almost* as good as the atomic clocks on board the satellites!
You do need to run a "site survey" to establish your location as exactly
as possible! Once your location is known, the calculation of the "speed
of light delay from your selected satellite to you can be done quickly
> And, I noticed the offset for Narvik is more stable than that for
> Hydra, is it due the PPS feeded to Narvik? But Bacchus is doing good
> too without PPS, Why?
Luck of the draw? Some of those cheap quartz crystals installed in PC
clocks are "right on". If you have such a crystal and maintain it at a
constant temperature you can get very good timing.
More information about the questions