[ntp:questions] NTP.conf using Dave Hart code.

unruh unruh at wormhole.physics.ubc.ca
Wed Feb 10 02:19:41 UTC 2010

On 2010-02-10, Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
> David Lord wrote:
>> BK wrote:
>>> Details:
>>> First I installed NTP4.2.4p8
>>> Then I updated the binaries with your updates labeled 4.2.7p8
>>> Then I installed the serial PPS driver 20091228
>>> Although for NTP purposes, you would like only one output string, I
>>> have to output two NMEA strings because there is another device
>>> looking at the serial stream also.  I am outputting GGA and RMC
>>> messages.  According to the GPS manufacturer (I am using a Garmin
>>> GPS15H) the PPS signal is applied just before outputting the NMEA
>>> sentences that would be for that time period.  I have the PPS signal
>>> set to 80ms width.  One oddity about my configuration is that the NTP
>>> server will not be up 24x7.  The machine will be booted, and I would
>>> like the ntpd to discipline the local clock to a reasonable (+-10ms)
>>> accuracy within 10 minutes. I have another machine that I will then
>>> synchronize to the computer with the GPS.
>> 10 minutes might be difficult from a cold start. How long from
>> bootup to ntpd starting? How far out will clock be after bootup?
>> I've been using "ntpd -q" before starting ntpd. That takes
>> around 5 minutes before time is set to usually well within 10ms.
>> After that ntpd is started and it's another few minutes before
>> it's serving time from nmea and another few minutes before it's
>> using pps to condition the clock.
> Ten minutes is not quite "asking for the impossible" but it comes very 
> close.  NTPD needs about thirty minutes to get a reasonable 
> approximation of the correct time and ten to twelve hours to achieve the 
> accuracy of which it is capable!
> If you want the accuracy that the system is capable of, you run it 24x7!
> With a GPS timing receiver and a computer running 24x7 you can stay 
> within 100 microseconds or less!  The GPS receiver is accurate to about 
> 50 nanoseconds; the difficulty is getting that time into a computer 
> while preserving accuracy.

Actually if the temp does not vary in the room and the computer does not
have large spikes of heating ( due to wirking) you can get 1-2usec. 
I used to be able to get that easily. Ssomething has happened recently
that I get a 10usec oscillation. Either periodic heating or something is
going wrong with the system.
If you were not on Windows, I would advise chrony, which has much faster
convergence, but that does not help with Windows. 

More information about the questions mailing list