[ntp:questions] ATOM falseticker with flag3 enabled
A C
agcarver+ntp at acarver.net
Thu Mar 29 05:47:51 UTC 2012
On 3/28/2012 22:02, Dave Hart wrote:
> On Thu, Mar 29, 2012 at 02:44, A C<agcarver+ntp at acarver.net> wrote:
>> Without flag3 it appears that ntpd does the heavy lifting but is unable to
>> use the PPS unless a prefer peer is set. With flag3 enabled, PPS is running
>> without a prefer peer.
>
> Being unable to use the PPS unless either a prefer peer is set or
> another knob is adjusted (see what I quoted earlier) is operating by
> design. The change in operation with flag3 1 seems likely to me to be
> an ntpd bug, which is why I'd like to see a bug report filed. Unless
> I'm missing some design intention, it seems to me PPS should not
> become reachable with no prefer peer regardless of flag3.
>
> Cheers,
> Dave Hart
>
I see that it appears to be by design which brought up my earlier
comment (and a response from Harlan Stenn) about the need for a prefer
peer. If a multitude of peers are available and all of them survive the
clock selection, why can't any of those be used as an enumeration peer
without the need for a preferred peer?
In my case I actually do want to use PPS but allow the other clocks to
be selected by the selection algorithm. The one flaw with this bug is
that the ATOM driver is constantly being marked as a false ticker when
flag3 is set so my system stratum and reference ID is incorrect as far
as the source of time is concerned (downstream servers see this GPS
system as just an Internet only clock without a PPS reference clock).
I can set the minsane to 0 if that will stop the ATOM driver from being
marked a false ticker (I know it's working correctly but ntpd doesn't
believe it). But my motivation is to avoid a preferred peer simply
because I've seen the system misbehave when the preferred peer has a
problem. It won't select one of the other peers and the clock goes
haywire. If I let the clock algorithm work normally then it manages to
ignore any flaws with the peers and gracefully shifts from one peer to
another.
More information about the questions
mailing list