Eugen COCA said the following on 04/07/2007 10:24 AM: > On Apr 7, 5:04 pm, "Richard B. gilbert" <[EMAIL PROTECTED]> > wrote: > >> I don't believe that more than one "prefer" keyword is useful. ISTR >> reading here that ONLY ONE source can be "preferred". > > Theoretically speaking you are right, but practically with only one > "prefer" the server loses synchronization after the GPS was powered > off. With the second prefer added in the 127.127.22.1 line the server > was locked on the PPS source for several days - with no significant > drift from the right time. I made this test several times, with the > same result. Anyone on this group could repeat the experiment and > confirm what I'm saying. I'm not a software specialist, but the ntp > developers could find out if this behavior was desired or is a useful > "bug".
There is, in my opinion, at least one very good reason to allow multiple prefer peers: If the refclock provides only a PPS signal and not a timestamp (e.g., an atomic clock), it's necessary to use an external server as the prefer peer to provide the time. Allowing multiple prefer peers greatly increases the robustness of such reference clocks by reducing their reliance on a single external server. It's OK if only one serves as prefer peer at a time, but being able to switch the prefer role to a second server if the first becomes unreachable or insane is a big win. John _______________________________________________ questions mailing list [email protected] https://lists.ntp.isc.org/mailman/listinfo/questions
