[ntp:questions] Re: NTP sync on a standalone network (Windows 2k)

David Woolley david at djwhome.demon.co.uk
Sat Aug 19 22:49:53 UTC 2006


In article <18541.bryanh at giraffe-data.com>,
bryanh at giraffe-data.com (Bryan Henderson) wrote:


> for which there is a code assigned is described as, "uncalibrated
> local clock used as a primary reference for a subnet without external
> means of synchronization."

The thread was actually about NTP, which, while it still has this 
function, is clearly designed on the assumption that all reference
clocks provide true time.  RFC 1305 doesn't seem to have a local
clock concept, and I think its inclusion in later versions is
more pragmatic than an indication of desirability.

There are other protocols, which are designed to select amongst
free running clocks.

> The protocol also gives you a means of specifying precision so that an
> SNTP server could say, "It's UTC 2006-08-17 13:34:01 give or take 2048
> seconds" if it wanted.

Precision doesn't have that meaning.  It's a measure of how accurately
you can read the clock on the server, either because it is low resolution,
or because of the time to physically read a high resolution clock.  This
is consistent with the SNTP and NTP specifications.

The value in the packet that indicates the uncertainty in the time due to
drift is root dispersion.  root distance, in this case, in effect gives the
error due to the initial setting of the clock.

Note that the actual RFC 2030 document requires dispersion and delay 
to be set to zero and defines them in terms that don't make a special
case of an uncalibrated local clock.  However, I think my interpretation
is more in line with the intended meaning of these fields.  I presume the
intention of using zero is to indicate that the fields are invalid.

I have a feeling that the W32Time protocol reflects these unchanged, so,
for an unsynchronised client, the result tends to indicate poor
quality time.




More information about the questions mailing list