Nelson Minar wrote:
>
>> It's surprising ntp doesn't take into account network delays in
>> some way...
>
> NTP does try to measure and account for network delay; otherwise our
> clocks would be 70ms or more off! But my understanding is it assumes
> that network delay is symmetric: ie delay from A to B is the same as
> delay from B to A. That hasn't been true on the Internet for 10+
> years, but it's really not true if your NTP server is on the end of
> an ADSL link. Particularly one where the slow upstream pipe gets
> full.
I am on an ADSL link; the round-trip time to the nearest Internet
Exchange is about 10 ms. That means that the asymmetry error in my time
must be under 5 ms. (I do take care to avoid saturation of my ADSL link)
> What I don't know is how much error we get from the reality of
> asymmetric network routing. Has there been anything written about
> that? I was a little surprised ADSL servers are in the pool at all;
> I'd naively assumed their time would be bad enough to not be worth
> it. But better to have some uncertainty than no server at all.
I am pretty sure to serve the correct second ;-). "The pool" gives you
synchronisation in the 100ms range, which is good enough for most uses.
If you want microseconds, buy a GPS with PPS output.
Greetings,
Peter.
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
