[ntp:questions] different Ntp servers...

Andy andyy1234 at gmail.com
Tue May 26 07:36:38 UTC 2009


On May 26, 7:26 am, Harlan Stenn <st... at ntp.org> wrote:
> >>> In article <JMydnbiIFOvm2IbXnZ2dnUVZ_gJi4... at giganews.com>, "Richard B. Gilbert" <rgilber... at comcast.net> writes:
> > Harlan Stenn wrote:
> >> If you connect directly to an S1 server and that server goes insane, you
> >> are screwed.  If you connect to a well-configured S2 server you have a
> >> lot better confidence that you'll be getting useful time.  I recommend
> >> you have as many of your machines as possible/reasonable peer with each
> >> other.
>
> Richard> Does this really help unless each peer has at least one unique
> Richard> source of time?
>
> I'm thinking of failure cases - it should leave a really big clique that
> will outvote an insane server.
>
> It would be interesting to document various failure cases and see if we can
> come up with a BCP document to describe these.
> --
> Harlan Stenn <st... at ntp.org>http://ntpforum.isc.org - be a member!

About 10 years ago we used to have 2 seperate GPS receivers feeding
into two Sparc servers via their serial port. I can't remember the
version of NTP we ran however these were our stratum one servers, One
of the Sparcs developed a fault on it's serial interface which delayed
the input meaning the two stratum one servers started showing
different time so all the statum 2 servers stopped trusting both
stratum 1 servers as they could not determine which server, if any,
was telling the  truth. Ever since then we have always used at least 3
stratum one servers with seperate stratum zero sources to form a
quorum so that startum 2 can differentiate the good from the bad. I
believe this is set down as best practise in the ntp.org FAQ - this is
a practical example of why this is wise.




More information about the questions mailing list