Ryan Malayter wrote:

I suppose I'm just bothered - in an intellectual, obsessive-compulsive
way - by all the DSL-based pool servers out there advertising
stratum-2 but serving up time with multi-millisecond phase errors.
Especially when those errors are easy to calculate and correct.
The stratum-2 servers should not be the main problem, assuming they
are synchronized over the same link that they use to advertise their service.

What is becoming clear is that the situation varies very much per country.
We had slow and very a-symmetric ADSL here when service started a
couple of years ago, but fierce competition with the cable market has
driven up speeds towards the theoretical limits, which means more symmetric.

For example, I have an ADSL connection that I subscribed to when service
started, and was 512/64 at that time.  It was called "Basic", and a "Fast"
version of 1024/256 was available as well.
Today, my "Basic" connection is 6000/768 and "Fast" is now 8000/1024.
(or 12000/1024 when you are connected to an ADSL2-upgraded CO and have
an ADSL2 modem)

However, I am running a stratum-1 server.  I have three stratum-0
references, one longwave receiver and two GPS receivers.  This is mainly
because I use it as a test environment for the GPS daemon (gpsd), but
it enhances my time service.  What I observe is that other GPS referenced
systems are usually within a millisecond and often within a quarter of a
millisecond from my time.  That should indicate that any error on my
service is only caused by queueing delay on my link (which of course is
used for typical home Internet usage)

In my experience, it is very hard to tell ntpd what to do. Tweaking settings
using fudge usually accomplishes little, except maybe for tuning a fixed
time offset for a longwave receiver.  Most ntpd instances, given a couple
of external references, seem to pick one at random, even when you know
that it is not the optimal one (e.g. when given multiple sources at the same
stratum, it has little tendency to pick the one with the lowest delay).

Rob
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to