[ntp:questions] ntpdate refusing ntp.nmi.nl

Terje Mathisen terje.mathisen at hda.hydro.com
Sat Jan 10 09:12:56 UTC 2009


Folkert van Heusden wrote:
>>> Does a valid NTP source need to set the reftime to something valid?
>>> Does the ntp spec say so?
>> I can't tell you that, Folkert, but, in your particular environment, would
>> /you/ trust a NTP source which was last synched 3 days ago?
> 
> Well, they (NMi) are the Dutch national organisation for standards. This
> time ought to be no more than micro (nano?) seconds from UTC.

Absolute worst case should be something like 50 ns offset from the UTC 
assemble clock, more probably somewhere in the 1-10 ns range.

(Do note that the real UTC clock doesn't exist, except as a paper number 
generated about a month after the fact: After comparing and weighing all 
the clocks that go into the official UTC assemble, each contributing 
clock (or national lab assemble of clocks) get back a number stating how 
far they were off.)

> What the guy from NMi tells me is that they synced the time a couple of
> days ago and then hooked it up to their 4 cesium clocks. So in theory
> there shouldn't be any difference with UTC.

See above.
> 
> I'm bothering you people for this as I would like to get them to fix it.
> But for that I need to be able to give a explanation.

Assuming the clock itself is OK, then I believe the problem might be 
with their dispersion calculation, i.e. the dispersion value will be 
assumed to increase linearly from the time of the last sync to a better 
(lower stratum) reference clock.

For a standards lab Cs clock, that dispersion value should be in the low 
ns range even after multiple days, and never get up to a single us, afaik.

Terje

-- 
- <Terje.Mathisen at hda.hydro.com>
"almost all programming can be viewed as an exercise in caching"




More information about the questions mailing list