[ntp:questions] Allan deviation survey

David L. Mills mills at udel.edu
Fri Sep 10 22:10:08 UTC 2010


All my measurements were in temperature-controlled environments, such as 
a campus lab or home office, and the data were collected over one week. 
The temperature varied less than a degree C. However, I have data from 
Poul-Henning Kamp for a similar experiment done in summertime Denmark 
where the environment was not controlled. As expected it looked 
something less than shown by the graph attached to my previous message. 
The Allan characteristic tends to fizzle out at lags greater than about 
one-fourth the span of the measurements. Thus, if you collect samples 
for only one day the maximum lag could not be more than a few hours and 
the diurnal effects would not be apparent.

A previous message implied that, once the Allan characteristic was 
determined, it would show chrony to be better than ntpd. Be advised the 
default time constant (at 64 s poll interval) was specifically chosen to 
match trace 1 on the graph mentioned above. In other words, it is in 
fact optimum for that characteristic and chrony can do no better.

Having said that, modern machines are faster and with less phase noise, 
although with the same rotten clock oscillator. Thus, I would expect a 
modern machine to behave something like trace 3 on my plot, where the 
intercept is more like 200 s than 2000 s. From anecdotal evidence, 16 s 
is about right, but 8 s is too vulnerable to jitter in the Ethernet NICs 
and switches. NIC jitter can vary widely, even using the same Ethernet 
chip, specifically the PCNET chips that do scatter-gather on-fly and 
coalesce interrupts. I have measured jitter components from 150 ns with 
a i386 running FreeBSD  to over 1 ms with a Sun Ultra running Solaris 
10. So, any performance comparisons must take these differences in account.


unruh wrote:

>On 2010-09-10, Miroslav Lichvar <mlichvar at redhat.com> wrote:
>>I'm trying to find out how a typical computer clock oscillator
>>performs in normal conditions without temperature stabilization or a
>>stable CPU load and how far it is from the ideal case which includes
>>only a random-walk frequency noise.
>>A very useful statistics is the Allan deviation. It can be used to
>>compare performance of oscillators, to make a guess of the optimal
>>polling interval, whether enabling ntpd daemon loop to use FLL will
>>help, how much better chrony will be than ntpd, etc.
>>If you have a PPS device and would be willing to run the machine
>>unsynchronized for a day, I'd like to ask you to measure the Allan
>>deviation and send it to me.
>>I wrote a small ncurses program that can be used with LinuxPPS to
>>capture the PPS samples and create an Allan deviation plot. An
>>overview is displayed and continuously updated while samples are
>>collected. Data which can be used to make an accurate graph (e.g. in
>>gnuplot) are written to the file specified by -p option when the
>>program is ended or when the 'w' key is pressed.
>>Available at:
>>Obligatory screenshot :-)
>>           Allan deviation plot (span 11:09:55, skew +0.0)
>>     ???	      
>>     +
>>     ???     
>>     ???  + +
>>1e-06???      +     
>>     ???       +++
>>     ???          ++ 
>>     ???            +++ 
>>     ???               ++  
>>1e-07???                 +++
>>     ???                    ++ 
>>     ???                      +++
>>     ???                         ++                ++++++
>>     ???                           +++         ++++      
>>1e-08???                              +++++++++         
>>     ???                                              
>>     ???                                             
>>     ???
>>     ???
>>   1e+00       1e+01	   1e+02       1e+03	   1e+04       1e+05
>>w:Write   q:Quit   r:Reset   1:Skew 0.0   2:Skew +1.0   3:Skew -0.5
>>To make a good plot:
>>1. disable everything that could make system clock adjustments
>>2. start ./ppsallan -p adev.plot /sys/devices/virtual/pps/pps0/assert
>>   (change the sys file as appropriate)
>>3. let it collect the PPS samples for at least one day
>>4. hit q and send me the adev.plot file
>Except yo uknow that the typical computer clock is driven mostly by
>temperature variations, and those are time of day dependent. People tend
>to work during the day thus their computer works during the day, and
>does not at night. Ie, there is a very strong daily cycle in the temp of
>the computer. That is NOT within the Allan model, and the Allan
>variation and minimum are really irrelevant with this highly
>non-stochastic noise model.
>questions mailing list
>questions at lists.ntp.org

More information about the questions mailing list