[ntp:questions] Re: offset values from ntpq and ntptrace

David Woolley david at djwhome.demon.co.uk
Sat Apr 1 09:36:18 UTC 2006


In article <1143789434.602435.281440 at i40g2000cwc.googlegroups.com>,
Tanguy.ropitault at gmail.com wrote:

> I know that sometimes, it's boring to anwser to simple questions but

It's not a simple question, at least not without a lot of detail about
your particular case that I doubt you are allowed to tell us, and even
that would only make the answer simple.

> one more time, you don't answer to my question... juste make a sentence
> please for explaining. Estimated error is the estimated error compared
> to the server time, or compared to the reference clock? For me, it's

It is not the instantaneous error, but rather a statistical error bound to
some level of confidence; on most occasions, UTC will be somewhere between
local clock time - estimated error and local clock time + estimated error,
but on other occasions it will be outside that range.  If ntpd knew the
instantaneous error, it would corect it so as always to be zero!

It is intended to be relative to true UTC, not to the local time on any
particular piece of hardware, although, unless the reference clock is
giving out good error estimates itself, it might be rather closer to being
referred to the reference clock.  You shoulnd't, of course, have a network
that derives its time from a single reference clock.

>From the sounds of things, I think you should be telling your manager
the maximum error, as this one makes all the worst case assumptions,
except for lost clock interrupts and the application program time reading
precision, and is pretty close to the 100 percentile confidence limit.
If his system is using any Windows machines, make sure that you tell him
that Windows application programs can only read the time to a precision
of 10ms or worse. Also point out to him that time is generally only
useful in conjunction with external events, and all systems have an
indeterminacy in the delay between the external event and when the time
is read, that, without special hardware, and operating systems, is often
several milliseconds.

If you need to be concerned about 2ms error estimates, you, almost
certainly, need to be doing a white box analysis of the timing latencies
of the system and should be asking much more specific and technical
questions.

If you want us to tell you whether or not the timing is good enough for
you, you need to tell us the nature of the problem the computer system is
trying to solve.




More information about the questions mailing list