[ntp:questions] Ginormous offset and slow convergance

unruh unruh at invalid.ca
Wed Nov 30 02:21:57 UTC 2011


On 2011-11-29, Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
> On 11/29/2011 1:42 PM, Pete Ashdown wrote:
>> Running with an Oncore GPS&  a TAPR TAC.  If I "ntpdate -b" a nearby
>> synchronized server before I start ntpd, the offsets initially look pretty
>> good:
>>
>>       remote           refid      st t when poll reach   delay   offset  jitter
>> ==============================================================================
>>   GPS_ONCORE(0)   .GPS.            0 l    -   16    0    0.000    0.000   0.000
>>   time-C.timefreq .ACTS.           1 u   39   64    1   37.534  -11.785   0.001
>>   ntp-nasa.arc.na .GPS.            1 u   38   64    1   19.868   -0.948   0.001
>>
>> As time goes on, things start to get crazier:
>>
>>       remote           refid      st t when poll reach   delay   offset  jitter
>> ==============================================================================
>> xGPS_ONCORE(0)   .GPS.            0 l    6   16  377    0.000  -999.97   0.003
>>   time-C.timefreq .ACTS.           1 u   50   64    3   37.370  -16011.   0.080
>>   ntp-nasa.arc.na .GPS.            1 u   52   64    3   19.823  -16000.   0.046
>>
>> More time with 100% reachability:
>>
>>       remote           refid      st t when poll reach   delay   offset  jitter
>> ==============================================================================
>> xGPS_ONCORE(0)   .GPS.            0 l    5   16  377    0.000  -15999.   0.002
>> +time-C.timefreq .ACTS.           1 u   19   64  377   37.887  -16011.   0.122
>> *ntp-nasa.arc.na .GPS.            1 u   24   64  377   19.760  -16000.   0.027
>>
>> Finally after 30 minutes, a lock and convergence, but still big offsets:
>>
>>       remote           refid      st t when poll reach   delay   offset  jitter
>> ==============================================================================
>> oGPS_ONCORE(0)   .GPS.            0 l    4   16  377    0.000  172.425  34.702
>> +time-C.timefreq .ACTS.           1 u   29   64   77   37.579  151.544 104.339
>> *ntp-nasa.arc.na .GPS.            1 u   22   64   77   19.779  165.811 106.581
>>
>>
>> Is there anything I can do to decrease the convergence time?
>
> Little or nothing!  NTPD can, and sometimes does, take ten hours to 
> reach "steady state".  It needs about thirty minutes to find a 
> reasonable facsimile of the correct time.  For the next nine hours and 
> thirty minutes, it will refine that value until it's as good as it's 
> going to get.
>
> If you maintain a constant temperature in the room in which you keep 
> your computers, NTPD will maintain an accurate and stable clock as long 
> as you maintain that stable temperature.
>
> If you can't maintain that steady state you need to either fix things 
> until you can or lower your expectations.
>
> There is another "product" That will synchronize quickly.  I don't 

Possibly chrony. Except that the "price" of faster convergence is vastly
improved accuracy and just as stable as NTPD. 
But it only runs on Linux or bsd, not windows.

(If time spans, rather than time is most important to you, then chrony
might be slightly worse than ntpd, because of the algorithm used to
correct the clock-- a fairly rapid slew to correct the offset, and a
frequency change to correct the drift. )

Note that if you get rid of ntpd's drift file, and use the burst option,
you can get faster convergence from ntpd apparently. 




> recall its name. Supposedly there is a price in accuracy and stability
> for that quick start.  Consult the Control Systems Theory people about
> the NTP algorithms and how they differ from algorithms giving fast 
> convergence.
>
>
>



More information about the questions mailing list