[ntp:questions] Re: Standards for time sync via sntp/ntp

David Woolley david at djwhome.demon.co.uk
Sun Jun 11 13:57:41 UTC 2006


In article <1150020014.481403.102760 at c74g2000cwc.googlegroups.com>,
Alexander Kugel <alexkugel at gmail.com> wrote:

> Unfortunately I can not find any conofrmation of this behavior in any
> standards I've checked (rfc 2030, NTP documentation, etc.).

That's the point of SNTP, it doesn't specify how time offsets are handled.
However, the rules you quote are, in my view, not at all satisfactory, 
and if w32time actually obeyed them it would make it even more unsuitable for
feeding real NTP clients than it already is.

In general, all time errors greater than the clock resolution should be 
corrected.  A slew is generally advisable for small errors, but note that
nptd only slews upto 0.128 seconds, after than it prefers to step.

Slewing at one part in 60 is not something that would be supported by 
normal operating systems; one part in 2,000 is much more common

> Is this behavior standard/recommended? If yes, what are the standards
> that describe it in more details?

I'd say it is neither standard nor reccommended.




More information about the questions mailing list