[ntp:questions] Synchronizing to LOCAL(0): Startup time

David Woolley david at ex.djwhome.demon.co.uk.invalid
Mon Oct 15 06:34:47 UTC 2007


In article <47128E4A.2010409 at ntp.isc.org>,
mayer at ntp.isc.org (Danny Mayer) wrote:
> The normal stratum for the local clock is 10 and not 6 and iburst has no

That depends on whether or not it is externally disciplined.  Ideally if
it is only used with no source of true time, or only after the last
source fails, it should be 14 if there are only clients downstream, reducing 
by two for every layer of server with a local clock defined, and by one
for each layer when there is no fallback local clock.  In the strict
tree with local clock, rather than the orphan mode, solution to this 
case, I think that would mean it should be 12, to minimise the risk of false
time leaking, whilst still ensuring that all clients are synchronised to
the root server.

If it is externally disciplined, it needs to be penalised a bit, because
ntpd won't know when the discipline fails, but one is not trying to
minimise the spread of a completely bogus time, so 6 might be reasonable.
I suspect that is why commonly distributed sample files don't set it to
10.

> Be aware that Linux has been known to drop interrupts causing issues

Specifically timer interrupts.  Windows is also guilty of this, and it
wouldn't surprise me if multimedia timers make that worse, even though
permanently enabling them is now the reccommended option.

> with the proper operation of ntpd. This is even more likely to happen

They will also interfere with the proper operation of any real time 
multimedia that is relying on the clock, so a time synchronisation 
protocol will be beneficial, by preventing errors accumulating.




More information about the questions mailing list