[ntp:questions] NTP client with 4 servers lost sync
nuno.pereira at g9telecom.pt
Fri Jul 3 10:39:57 UTC 2015
> De: questions [mailto:questions-
> bounces+nuno.pereira=g9telecom.pt at lists.ntp.org] Em nome de Brian Inglis
> On 2015-07-02 13:40, Nuno Pereira wrote:
> > But how can you explain that a client with just one source was better than
> > client with 4 sources?
> Okay as long as that single source is up and working and accurate - if it is
> inaccurate, clinets would not know.
A time island, which is acceptable, but without backup in case the source
fails, which isn't optimal.
> > It's not just from this leap second situation: it's from some months where
> > have some clients with just one source that are having less problems that
> > 4 sources configuration.
> > 4 is not odd, I know, but in that case I have to go for a 5 sources
> > configuration, as a 3 source configuration can fall into a 2 available
> > configuration if one of them fails?
> > And as we prefer to use local sources, in order to have all of our clients
> > with better accuracy between them, in that case we would need to have 5
> > servers of NTP! It's a little insane, in our opinion.
> Three local servers, plus two good external servers, plus a few good pool
> should allow your clients to get reliable time. Let ntpd do its own thing to
> accurate time. Collect loopstats and peerstats and compare consistency.
That sounds good, except for one thing: our local servers get the minority,
they may a better source, but some problem in the pool servers (as they happen
a lot) or in the network,
or some less accurate time in our local servers will bring us to a situation
where our local clients
have inconsistent time between them: some still use the local sources, some
already went to the external ones.
In other words, we prefer a time island than to have more accurate times.
Isn't a peer source good in order to keep consistency between local servers?
>From what I've tested, peer servers are discarded by the ntp algorithm (and
marked with a "-" signal)
and aren't used as time sources.
> Take care. Thanks, Brian Inglis
> questions mailing list
> questions at lists.ntp.org
More information about the questions