[ntp:questions] Isolated Network Drift Problem

David Woolley david at ex.djwhome.demon.co.uk.invalid
Mon Nov 24 07:57:13 UTC 2008


Cal Webster wrote:

> 
> "adjtimex" is not available on our systems. They're all Red Hat/Fedora
> or derived from them. I do have "hwclock" but I don't think it will do
> what was suggested.

adjtimex(8) might be called ntptime, but adjtimex(2) exists on all Linux 
systems that support the kernel time discipline.  If it doesn't exist, 
you would certainly get failures when trying to use the kernel discipline!

> 
> If I set a "bias" value in the drift file won't NTP change it anyway?
> After I zeroed it out, it has been changing over time.

The bias will only change on machines that have a source other than 
their own local clock.  The local clock always has a zero offset, by 
definition, and never results in frequency adjustments.

> 
> I'd like to stay away from a hierarchy with a single point of failure.

Then, if you also don't want random drifting, you need real, UTC 
traceable, hardware clocks.

Even if pure orphan mode works on your system, machines dropping in and 
out will result in machines changing from absolute master to slave, and 
therefore becoming at risk of having their drift file value change. 
timed, which would be the traditional appropriate solution under those 
conditions, would also wander somewhat randomly as the timing source 
changed.




More information about the questions mailing list