> Unless the asymmetry is on a `critical' link that *all* my NTP packets
> travel along (such as my link to my ISP) I don't have a problem,
> right? Furthermore, X would have to be different to Y by a
> non-negligible amount, no? So even if the link *was* asymmetric, a
> very fast link would still be very accurate.

There are (at least) two sources of asymmetry.  One is link speed - examples 
are ADSL and cable modems.

The other is queuing delays.  You can get some neat graphs if you plot offset 
vs delay.  The typical pattern is a wedge pointing left.  The point of the 
wedge is the no-delay case.  To the right of that, above the horizontal is 
more delay in one direction.  Below is more delay in the other direction.  
There are examples in various papers.  I don't have URLs handy.  Or you can 
make your own.


> I know I'm pushing the issue, but I always thought that one of the
> goals of NTP was to make timekeeping as accurate as possible *without*
> requiring the kind of devices you're talking about, and that it does
> this by using servers `democratically'. 

What sort of accuracy were you expecting for "as accurate as possible"?  Why?

Over the net, 1 second is easy.  1 ms is hard.  I'd expect generally within a 
few 10s of ms.  Depends on your network and which servers you are using.

It's generally easy to get better than a ms with a local GPS clock.  I'd 
expect 10s of microseconds.

In general, ntpd is a good definition for "as accurate as possible".  It's a 
hard problem.  You could probably do a bit better by tweaking various 
parameters and/or heuristics if you know something about your setup that 
makes it different from the average/typical setup.


-- 
The suespammers.org mail server is located in California.  So are all my
other mailboxes.  Please do not send unsolicited bulk e-mail or unsolicited
commercial e-mail to my suespammers.org address or any of my other addresses.
These are my opinions, not necessarily my employer's.  I hate spam.



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

Reply via email to