[ntp:questions] configuring server is not as expected

David Woolley david at ex.djwhome.demon.invalid
Tue Nov 23 22:07:05 UTC 2010


ntp at proinbox.com wrote:
> Hello and thank you for your reply.
> Unfortunately, the solution does not appear to be so simple, and I could
> still use some assistance troubleshooting this.
> 
> I've made the appropriate adjustments to the config file, replacing the
> server's address on each network with each network's broadcast address.

This is still weird.  The network level servers should be pulling, not 
having time pushed.

> Also I've added the ntp servers that are to be queried in such an order
> so that in the case that their shared primary time source becomes

I don't believe that there is any guarantee of how the order of the 
servers in the config file will affect the way ntpd handles them.

> unavailable, each can sync from the other, by means of either of the
> networks connecting them.

You need orphan mode for this to work at all sensibly.
> 
> I remain unable to make sense of this behavior.
> 
> Until now I've always relied on the internet-provided sources.
> Occasionally there's drift that I hope to eliminate by shifting queries
> to the most central of all servers.

If you must have multiple servers with local clocks configured, they 
must form a chain, otherwise, they will separate into two cliques which 
may stop you every synchronising to real time again.  Servers with local 
clock drivers report spuriously low error bounds.  (Very few people 
should be using the local clock driver.)

It looks as though you may not have local clock drivers configured, in 
which case, without orphan mode, the servers will simply fail towards 
stratum 16 and be ignored.

> Am I understanding correctly that by adding the broadcast addresses to
> the list, that these 2 servers can be queried just as the stratum 1 time
> source server out on the internet is?

No.  Broadcast will cause at most one poll, after which the clients will 
just listen.  Broadcast will not work if the stratum as  failed to 16.
> 
> For what it's worth, my intention is for all these networks to sync from
> server A 'Bascule', with source redundancy provided by server B
> 'Planck'.
> Server A's time will be defined by either the external stratum 1 time
> source first, or from server B as a fallback.

There is no evidence that this network is big enough to justify a public 
stratum 1 server.

> Server B's time will be defined by either the external stratum 1 time
> source first, or from server A as a fallback.

This is only achievable with orphan mode.

> Both server A & B will have a cron job to update their hardware clock
> every other month, interleaved.

If this is Linux, the hardware clock will be updated every 11 minutes!

You need to explain a lot more as to why you cannot do things normally.




More information about the questions mailing list