[ntp:questions] Consider the Undisciplined Local Clock synchronized at start-up

Serge Bets serge.bets at NOSPAM.laposte.invalid
Sat Apr 7 23:43:27 UTC 2007


 On Thursday, April 5, 2007 at 16:38:13 +0200, Spoon wrote:

> In this two-system setting, what is the advantage of orphan mode over
> the master-slave configuration?

Faster sync. And response to temporary shutdown of either one of the
machines is smoother. But you're never sure if your peers follow the
best clock, or which clock to set so both follow.

OTOH the master-slave gives better control to the carefull operator. You
designate the machine with the best clock to be the master, set it
precisely with eyeball and wristwatch, calibrate its driftfile, and can
be quite sure both machines will follow. Unless you shutdown the master.
Then the slave alone will continue quite well on its last frequency,
wandering only a little. But at restart the master will brutaly impose
its time again, perhaps bumping the slave frequency for some time, or
even forcing the slave to step.

This does not seem to happen so often with orphans, probably because the
just restarted system will follow the long-running one. Time steps are
acceptable at startup, but are always annoying later, aren't they?


>> Apply the iburst for refclocks patch, and use this keyword on your
>> server's "server 127.127.1.1 iburst" line. Ntpd will then sync on the
>> local clock in only 5 seconds.
> Do you have a link to the patch?

It's short enough to be pasted here:

= = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
NTP iburst for refclocks patch by Nathan Hintz
in http://ntp.isc.org/bugs/show_bug.cgi?id=294

--- ntp-4.2.4.orig/ntpd/ntp_refclock.c	Tue Jun  6 22:16:43 2006
+++ ntp-4.2.4/ntpd/ntp_refclock.c	Thu Jan 25 01:21:30 2007
@@ -373,6 +373,8 @@ refclock_transmit(
 				report_event(EVNT_UNREACH, peer);
 				peer->timereachable = current_time;
 			}
+			if (peer->flags & FLAG_IBURST)
+				peer->burst = NSTAGE;
 		} else {
 			if (!(oreach & 0x07)) {
 				clock_filter(peer, 0., 0., MAXDISPERSE);
= = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =


Serge.
-- 
Serge point Bets arobase laposte point net




More information about the questions mailing list