[ntp:questions] ntpdate works, but ntpd doesn't (reach = 0)

Martin Burnicki martin.burnicki at meinberg.de
Tue Feb 10 09:03:22 UTC 2009


Harlan,

Harlan Stenn wrote:
>>>> In article <uck566-23p.ln1 at gateway.py.meinberg.de>, Martin Burnicki
>>>> <martin.burnicki at meinberg.de> writes:
> 
> Harlan> Because ntpd also gets restarted, and there is a strong belief
> that Harlan> -g is bad for a restart and restarts will happen more often
> than Harlan> boots.
> 
> Martin> Huh? I'm afraid I don't understand what you mean here.
> 
> The general case is that -g should only be used at startup, when time can
> be stepped.
> 
> If ntpd is restarted, it is Bad for the time to be stepped backwards for
> many people.  Since this could happen with -g, when restarting ntpd for
> this common case, one should not use -g.

Anyway, I don't understand where the problem with -g is.

If the system time has been synchronized better than 128 ms before, by ntpd,
ntpdate, sntp or whatever, then ntpd will not step the time when it is
(re-)started. 

If the system time *is* off by more than 128 ms then the system time *will*
be stepped by ntpd anyway (maybe unless the -x, no slew option is given),
if the time offset is below the sanity limit, even if -g is not given.

If stepping the time back is Bad in general then stepping it back by 1000
second with or without -g is as bad as stepping it back by more than 1000
seconds with -g.

IMHO if this happens then this is a configuration problem, not an NTP
problem. 

Maybe initial time steps should be allowed by default but inhibited if -x is
explicitely given.

> From what I can see, it might be an improvement if ntpd had a mode that
> said "Stepping forward is OK, but do not step backwards."

Agreed. 

> What I really think is needed is an in-depth study of the various cases,
> perhaps with some new timekeeping functions that better implement what is
> needed.

I think here is a good place to discuss this, maybe in a new thread.

Martin
-- 
Martin Burnicki

Meinberg Funkuhren
Bad Pyrmont
Germany




More information about the questions mailing list