Chuck Swiger wrote:
David J Taylor wrote:
Chuck Swiger wrote:
David J Taylor wrote:
[]
Doesn't iburst violate the "more than 5 per minute" rule?
Yes. But with the default NTP settings, it shouldn't violate the
old 20 requests/hour guideline. iburst should be reserved for hosts
providing NTP services to other machines, and not used by end-client
workstations.
Time to start a different thread, perhaps.
My understanding of iburst is that it is for /any/ NTP client,
whether providing downstream services or not. Indeed, it is
particular appropriate for end-client workstations which may be up
and down more than servers, and therefore have more need of the
quick synchronisation which iburst can provide.
If you are using your clients against your NTP servers on a LAN,
iburst is fine.
However, it's not needed by clients by definition, because nobody
else is asking them for time, so whether ntpd is synced is has vey
little bearing on the accuracy of local time.
(Unless your ToD/BIOS hardware clock is seriously busted-- if you've
ever been able to sync to NTP, ntp.drift will be calibrated to
compensate. [1])
You probably should not use iburst against someone else's server
without asking them for permission or their access policy permits
rapid updates.
Note that Apple recently changed their defaults to "minpoll 12
maxpoll 17" for time.apple.com. That's 1+ hour (4096 seconds) and
36-hours, respectively. Clients should poll remote NTP servers at
once-an-hour internals or less frequently by default; polling quickly
to start up is reasonable and useful for servers, but it just doesn't
matter for pure clients.
OK, Chuck, I see the sense in your argument.
I'm not sure that message comes across in the official documentation,
though!
David
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
