[ntp:questions] Re: Quasi-On_topic: System (kernel) time jumps 3600 seconds at random times. Stumped.

Allen McIntosh nospam at mouse-potato.com
Thu May 19 03:09:32 UTC 2005

> As for the script we run to set the time, it's a legacy app. to say the
> least. However, I can't just start NTP on machines or configure my own
> stratum 1 as I'm with a large company where the ROI for every action I
> propose has to be carefully scrutinized.
How much time is being wasted now with support folks having to reset 
clocks?  How much of your time will be needed to track the problem down?
How much of your frame relay link gets eaten when ~350 machines, all 
roughly in sync, start banging on the NTP server at 3:15 a.m.?
> I'd love to run NTP on every machine we have or at least frequent
> ntpdates, but our network network group would scream at me if I
> suddenly started 2500 machines banging away at 4 time servers every 60
> seconds over frame relay connections(too much jitter already to get the
> poll times to increase).
As Brad suggested, use broadcast NTP.  (Even ancient versions of xntpd 
will have it).  I don't think a decent turnkey GPS box costs all that 
much compared to a day of your (fully loaded) salary.  I will admit that 
getting the right 25 :-) people up on the roof to mount the antenna may 
be a pain though.
> With 2500 machines, I can't run around starting processes on them on a
> whim; there must remain a reasonably strict level of standardization or
> they will become unmanagable.
> So far, running a cron to log time changes is probably my best bet, but
> it's somewhat daunting because of the extremely random nature of the
> time errors I'm seeing and the huge quantity of data I'm likely to
> gather.

So use the "daunting" part on management, assuming they care about 
fixing this.

I work for a large company, so I understand your problem, but there are 
probably some fairly good arguments you can make in favor of a gradual 
switch to NTP.

More information about the questions mailing list