Hal Murray wrote:
What you say assumes your clock is setup correctly and operating as expected. How do you know that everything is correct?
You can't... that way madness lies. Maybe one day I'll take a laptop to grenwich and try to sync off the atomic clock there... :P
You have to assume that the GPS device is operating within its design spec and isn't going to be wildly off. It'll be slightly out due to OS/Interrupt latency but nothing you're going to notice without some seriously expensive equipment.
You can use systems on the net as a sanity check. They may not be great, but they are available for not much cost/effort. If you plot offset vs rtt and
I posted the graph for a server *connected to the same switch* and the innacuracy was rarely less than 1ms. Connecting over the internet isn't going to get you anything worth measuring if you want the kind of accuracy we're talking about here. It's good enough to sync a network off, which is what it's there for...
the blob at the point of the wedge doesn't overlap an offset of 0 something interesting is going on. Aside from at offset in your local clock, it might be asymmetric network paths. (or maybe the remote clock is setup wrong)
A lot of internet paths are asymmetric in some way - the return paths vary greatly once you're more than half a dozen hops away (and it takes 10 for me to get out of my ISPs cloud.. 15 before I get anywhere useful). Anyone on ADSL or Cable has asymmetry builtin to the whole protocol (which NTP currently doesn't compensate for).
Plus you can't verify the remote clock is right at all.. plotting delay isn't going to help you one bit.. eg:
-veracity.mcc.ac 128.86.8.123 2 u 17 64 377 23.042 -0.494 0.621 +utserv.mcc.ac.u 128.86.8.123 2 u 12 64 377 22.897 0.453 0.310 +scarp.mc.man.ac 194.81.227.227 2 u 39 64 377 22.814 1.162 0.171
Those are all within 0.1ms delay of each other and also reading different times. Which one is more correct?? It's impossible to know.
I could say that since two of them are each side of zero, I'm likely to be about right... however I can't tell if the third one is actually more accurate... it requires an elusive '100% accurate reference clock' which very few people have access to.
Tony _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
