[ntp:questions] ATOM falseticker with flag3 enabled
snews at lordynet.org
Mon Apr 2 10:26:13 UTC 2012
A C wrote:
> On 3/31/2012 20:19, David Lord wrote:
>> A C wrote:
>>>> I'm not sure which value you are referring to. From the
>>>> "ntpq -p" billboard over a day, I see values of jitter
>>>> of the GPS as low as 0.002
>>> Yes, it's that jitter however there's a minimum computed value below
>>> which it never goes (varies per system). On my system it used to be
>>> computed at 0.061 (and all the configured clocks would report that as
>>> an initial jitter until some data came in). On the same system with
>>> 4.2.7 the value is now computed to be 0.122.
>>> In any case apparently according to David Hart the code has indeed
>>> changed so the computations are different.
>> Even after warm restarts I don't concern myself with the
>> level of jitter until ntpd has had time to stabilise. After
>> an update to ntpd or kernel, and without a pps source, it
>> can be a day or more until a stable value for drift has
>> been established.
>> Has using 4.2.7 solved the problem with the type 20 and 22
>> drivers both being marked as falsetickers?
>> Were you able to determine a reasonable fudge time value
>> for the type 20 driver by using noselect for the type 20
>> driver and having several good internet sources?
> Here's one to break everyone's brain:
> Using flag3 on ATOM with no prefer peer still disciplines the clock via
> the PPS signal but marks ATOM as a falseticker. Setting a prefer peer
> causes ATOM to be marked as a good clock ('o' tally code).
> The kicker: the clock runs better with the first configuration (no
> prefer peer but still disciplined) than with the second configuration.
> The offset on my system stays around +/- 5 us normally (unless the
> furnace turns on) when ATOM is marked a false ticker. But when it's
> configured with the prefer peer, it hovers around +/- 30 us.
> I'd rather have the better clock AND not be marked as a false ticker.
I was using the type-22 driver without problem before I
updated from ntp-4.2.6p3 to 4.2.6p5. I started using PPS
early in 2009 and the type-22 driver was the only option
back then. I'd have to search through the logs to find
what versions of NetBSD and ntpd I progressed through.
Still with the single type-20 driver I reconfigured to
disable kernel PPS (flag3 0) and changed polls from 64s
to 16s. After 100 minutes offset=0.003, jitter=0.004.
More information about the questions