Joel Reicher wrote: > 'scuse my naivete, but don't you get very good accuracy if you have > a consistent hardware clock, a drift file, and enough time (months?) > of using many servers to build up a good value in the file?
No. All it takes is one asymetric link between you and your timeserver to through your time off. NTP works by sending a query to a time server. That query is placed in a UDP packet, and generally sent over the Internet. >From your machine to the timeserver, it takes X milliseconds. From the timeserver to your machine, it takes Y milliseconds. Your machine (or any machine) can only measure the sum X+Y, and it then assumes X = Y. With asymetric links (i.e., upload speed != download speed), X != Y. Your time will be reliably skewed either a few ms ahead of or behind of true time. This is because an asymetric link consistently transmits packets faster in one direction that the other. Then there are other effects that can randomly push the time askew like jitter. These won't give a consistent offet, but add a random component to your offset from true time. The only way to get 1ms accuracy, really, is to have a local time source. GPS is nice. Time radio will work too, but remember that you *must* calibrate for speed-of-light delay (if you fail to do so, being ~140mi from the transmitter will already put you 1ms off). _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
