"Brian Garrett" wrote in message
news:O%2Ka.70027$Dr3.13243@fed1read02...
> BTW, www.gpsclock.com/check.html lets you check the status of NTP servers
on
> the net (the ones that don't filter out this kind of intensive polling,
> anyway.) Unfortunately it judges the accuracy of the server's time by its
> own system clock, which has been off by several hundred ms in the past.
> (_Hundreds_ of millisecs! Somebody call the cops!) Pretty cool site
> though.
It only compares the timeserver to its local clock as one check. It also
has a fairly large fudge factor built in to this comparison. This isn't to
compensate for an inaccurate clock, it was written to run on a machine right
next to a stratum one timeserver, it's because asymmetric network lag could
create the illusion of an inaccurate server.
It's more looking at the internal sanity of the timeserver. What stratum
is it running at? What stratum are its peers running at? What is it
reporting its stability as? How reachable are its peers in its opinion? Does
it have enough reachable peers at an equal or lower stratum that it
considers sane? Do all the servers its relying on themselves rely on the
same server? And so on.
DS
|