[ntp:questions] Re: synchronisation lost problem

Serge Tchesmeli zztchesmeli at echo.fr
Wed Dec 8 10:47:36 UTC 2004


Le Mercredi 08 Décembre 2004 10:37, vous avez écrit :
> Are you running a kernel that has HZ=1000 instead of 100?  That can cause
> lost interrupts, as I recall.

We do not have modified HZ variable. HZ=100

>
> The only other thought I have at the moment is to consider running the
> scripts/calc_tickadj script and then using that adjustment with the
> tickadj program.
>
> If you want to try this (this procedure is not generally recommended):
>
> - stop ntpd
> - run tickadj with no arguments to see your current "tick" value.

[root at m01 bin]# /etc/rc.d/init.d/ntpd stop
Shutting down ntpd:                                        [  OK  ]
[root at m01 bin]# tickadj
tick = 10000

> - run calc_tickadj, and:
perl calc_tickadj.in
40.385 (drift)
10002 usec; 10002385 nsec


> - - change the value in your driftfile to the new drift value
> - - run the tickadj script, providing it with the new tick value

root at m01 bin]# ./tickadj 10002
tick = 10002

> - restart ntpd

[root at m01 bin]# /etc/rc.d/init.d/ntpd start
Synchronizing with time server:                            [  OK  ]
Starting ntpd:                                             [  OK  ]

>
> Note that calc_tickadj assumes the "normal" tick value; if you have to
> run this more than once you must "accumulate" the tick deltas.
>
> Did I mention it is usually a bad idea to mess with the value of "tick"?
>
> If you are going to mess with "tick" it is best to do it in small amounts.
>
> After changing "tick", wait a little while and make sure things are
> getting better, not worse.
>
> H

Waiting to see what happening.



More information about the questions mailing list