[ntp:questions] ntpd stays synced after loosing gps

Chris Albertson albertson.chris at gmail.com
Wed May 11 16:28:09 UTC 2011

On Wed, May 11, 2011 at 8:29 AM, Nickolay Orekhov <nowhere at mail.ru> wrote:

> And what if it has no "other source" ?
> May be you could give me an advice on my system. I have one device synced
> from GPS and other devices synced from it by NTP. Often there will be only
> one NTP server.
> When sync server looses GPS I set some variable to "unsync" ( according to
> ntpd system status ).
> Other devices should also have status "unsync" and stop syncing but they
> will continue to sync from unsynced source with stratum 1 infinitely.

If you have a requirement for a robust system then you need redundancy.
Just to handle failures and normal system maintanance you need two of
everything (at least) and if you want to be able to detect subtle errors
where the GPS is not working perfetly you need at least three.

A cheaper way to go is to use a set of (say) five pool NTP servers.  Set up
two NTP servers and connect each of these to the five pool servers and then
point allof your clients at both of your local NTP servers.   If you need
more accuracy then add a GPS receiver to each of the two servers.  If you
want to be able to detect problems with GPS you need three DIFFERENT GPS
receivers made by three DIFFERENT companies.  But you can also use the set
of five pool servers as a check on the GPS.

There are just a few simple "rules".  The first comes from an old joke that
just happnes to be true
1) A man with one watch kows what time it is, A mand with two watches is
never sure.  So clearly you need more than two if you want ot be sure.
Three UNRELATED clocks are the minimum.  If you care about being sure of the
time.  Pool servers, GPSes in any combination
2) There should be no single point of failure.  Any service that fails, any
GPS antenna that fails and so on should not bring the whole system to a
halt.  This means two of everything that is important.

Chris Albertson
Redondo Beach, California

More information about the questions mailing list