[ntp:questions] refclock on Windows graphs

Unruh unruh-spam at physics.ubc.ca
Fri Mar 20 16:50:23 UTC 2009

"David J Taylor" <david-taylor at blueyonder.neither-this-bit.nor-this.co.uk> writes:

>Unruh wrote:
>> "David J Taylor"
>>> I put the temperature graph next to the timekeeping graph here:
>>>  http://www.satsignal.eu/mrtg/performance_feenix.php
>>> but I think that ambient temperature would be much more useful than
>>> HDD temperature.
>> Unfortunately the scale on the graph is a bit useless (the variation
>> is
>> squeezed up far too much).

>Yes, and no.  It's MRTG, so it's going to plot from 0, and if the 
>variation isn't as much as even 5 degrees then I think I need to look at a 
>different measurement.  As a quick fix, we have to make do with what we 

Ah, so I guess it should be plotted with an offset of -25 or something like
that. But as you say, since the resolution is 1C, that may not be terribly

>> Also It is not at all clear what the
>> "Timekeeping" graph is a graph of. Is that the offset, the drift rate
>> correction? And what is 5001's ? (the quoted offset) It is the drift
>> rate
>> that should correlate with temp.

>Timekeeping is the offset.  The original graph had "offset + 500us" where 
>the "u" was the micro symbol, but while that has displayed correctly in 
>most of the place, it has failed on the graph Y-axis label.  I'll replace 
>"micro" with "u".

To see the effect of temp, it would be better to plot the drift. Mind you,
ntp has such a long time constant, it is not clear how well that will work.
I recall one of your graphs had a drift plot, and the correlation with an
expanded temp plot might be revealing. 

>> It is certainly true that something like disk temp is a best a poor
>> proxy
>> for the timing crystal temp. esp since the disk is probably not the
>> main
>> heat source. The CPU temp would rpobably be a better proxy, but
>> obviously
>> even that is not very good. "motherboard temp" might be better.
>> Anyway, by plotting both the drift rate correction and the temp of
>> these various
>> locations you could get a better idea of which correlated best with
>> the
>> timing crystal rates.

>Thanks for your input, Bill.  If I can get a more accurate temperature, I 
>will look into plotting the offset on MRTG as well, or using a better 
>plotting program.  For me end use, though, it's the offset which matters.

Sure. Unfortunately in ntp, the offset is purely a function of the
drift,(integral of true drift minus ntp correction to drift). There is
little which causes the offset to change directly (except random
measurement errors which should be in the 1 usec range for a PPS, 10s of
usec for network, but should be roughly random). Temp causes drift
changes and only secondarily offset changes. 


More information about the questions mailing list