[ntp:questions] ntp & system without a rtc

David Lord snews at lordynet.org
Wed May 15 19:18:25 UTC 2013


unruh wrote:
> On 2013-05-14, David Woolley <david at ex.djwhome.demon.invalid> wrote:
>> Richard B. Gilbert wrote:
>>
>>> NTPD is NOT designed for fast convergence.  From start up to get the 
>>> minutes correct, NTPD will need about thirty minutes!  To get the best 
>>> time you are going to get, you will need to wait for about ten hours!
>>>
>> Convergence to within 128ms should take a lot less than that.  With 
>> iburst, I believe it should be less than one minute.
> 
> Since ntpd steps ( changes the rate by an infinite slope) if the time is
> out by 128 ms, yes, convergence IS faster than that. ( however it has to
> make sure that the clock really is out by that much and that takes a few
> cycles of clock querrying).
> 
> For a ms accuracy your could expect a couple of hours. for 10
> microsecond, about 10 hr. 
>>>

Hi

I posted a reply on May 11, but did not give the ntpq stats
as they'd already been archived away.

After three reboots to replace kernel and userland taking
around 10-15 minutes 'ntpq -p' with polling at 6 minutes gave
the following:

    min  reach   offset   jitter

    -18    377   -0.001    0.004
    -12    377    0.007    0.007
     -6    377   -0.001    0.006
      0      0    0.000    0.000
      6     17   -0.036  275.881
     12    377    0.126    0.125
     18    377    0.173    0.049
     24    377    0.047    0.092
     30    377    0.007    0.028
     36    377   -0.001    0.007
     42    377   -0.001    0.004

There was a well established ntp.drift file and my rc.conf starts
ntpd with flags="-g -N -p /var/run/ntpd.pid"

NetBSD/6.1_RC3
ntpd 4.2.6p5-o
PPS from Sure demo board


David



More information about the questions mailing list