[ntp:questions] Strange jumps in PPM

unruh unruh at invalid.ca
Thu Aug 22 18:06:06 UTC 2013


On 2013-08-22, A C <agcarver+ntp at acarver.net> wrote:
> On 8/21/2013 23:57, David Woolley wrote:
>> On 21/08/13 21:24, A C wrote:
>>
>>>
>>> So the idea isn't that ntpd should figure out on its own without user
>>> intervention.  Instead there would be some kind of
>>> configuration/fudge/tinker options available so that the user can supply
>>> baseline information that can help ntpd identify a rogue source versus
>>> normal system behaviors.
>>
>> Actually, one of the main complaints about ntpd is that it favours the
>> theory that the source is the problem, when the crystal temperature is
>> the real problem.  (The exception to this is people who manually yank
>> the local clock and then complain that ntpd doesn't immediately reset it.)
>
> In most cases of steady drift or cyclical drift that may be true and I 
> certainly do see temperature effects on my system over the course of a 
> day.  However in this particular case I can pretty much guarantee that 
> it's the source that caused the problem.  I certainly didn't have a 
> temperature swing large enough to change the clock rate from -76 to -34 
> PPM in 16 seconds and then have it go back to -76 after 128 seconds (8 * 
> 16) especially considering I was sitting in the room at the time it 
> happened staring at a thermometer that looks at the internal temperature 
> of the machine.  The nominal swing caused by temperature on my system is 
> about 2-3 PPM (-76 to -73 or so) for an internal temperature swing of 
> about 10 degrees Fahrenheit.

What I said was that it is obvioius to you, after the fact, and with the
knowledge you have of your system, that the source is the problem. But
ntpd is an extremely stupid program, in that it has virtually no memory
and must base its decisions on the latest measurement of the offset
only. Thus a sudden change in the is taken to mean either a change in
the source, or a change in the measurement of the source. In case it is
the latter, it changes the computer time only slowly, assuming that
measurement errors will average out if you make the change slowly
enough.



More information about the questions mailing list