[ntp:questions] Beginner's questions to NTP configuration option "peer"

David Woolley david at ex.djwhome.demon.co.uk.invalid
Tue Sep 2 21:19:51 UTC 2008

Nottorf, Stefan wrote:

> synchronisation via NTP in a small to medium size network. Reading
> through several web pages and parts of the book of Mr. Mills I still

Being able to read the book probably disqualifies you as a beginner.

> have some questions, which I will post after briefly describing a part
> of the network.
> located in. The stratum 2 blades then use each other as peers. In case
> of a failing clock, these servers pretend to be stratum 5 (via fudge).

This can be dangerous.  You risk a mutual appreciation society in which 
they all confirm each other's idea of the wrong time, and as the local 
clock driver appears to have a rather low root distance, a single real 
clock doesn't have to differ much to be rejected.

I believe orphan mode may avoid this.  I think preferring the real time 
server may also do so.

On other approach is to fudge each clock to a different value with steps 
of two between them.

> |-------------|
> | Lantime V4  |
> | Stratum 1   |
> |-------------|
>       |
>       X
> |-------------|              |-------------|
> | Enclosure 1 |<--peer------>| Enclosure 2 |
> | Blade 1     |              | Blade 1     |
> | Stratum ?   |              | Stratum 2   |
> |-------------|              |-------------|
>       |          ___________/
>       |         /
>       V        /
> |-------------|
> | Enclosure 1 |
> | Blade 2     |
> | Stratum 3   |
> |-------------|

This will only happen if both enclosure 1 and enclosure 2's blade 1's 
are specified as servers of enclosure 1 blade 2.  In this case ? is 3.
Note.  If they are both servers, the no fault case may actually have 
enclosure 2 blade 1 as the official reference source, although both will 
be used to discipline the time.

> In the documentation it is written that "Should one of the peers lose
> all reference sources or simply cease operation, the other peers will
> automatically reconfigure so that time and related values can flow from
> the surviving peers to all hosts in the subnet". Does the above
> ASCII-art describe this behaviour correctly?

Probably not, because it probably assumes that rearrangement can add 
links implied by transitive relationships, when all it can do is reverse 
peer relations and choose a new spanning sub-tree.

> 2) If a stratum 2 host has peer connections to other stratum 2 hosts AND
> it's network connection to the stratum 1 server(s) fail... Does the host
> in question
>    a) drop to stratum 3 (because it's next time source is the stratum 2
> peer) like it would if both peer sides used each other as time server
> via the "server" keyword? 

Drop to stratum 3.

>    b) drop to it`s fudged stratum level (stratum 5 in this case)?

Note that you will probably need multiple networks for you to be able to 
have a failure of connectivity from just one of the peers to the 
Meinberg without it also losing contact with the other peers and its 

More information about the questions mailing list