In article <[EMAIL PROTECTED]>, Serge Bets <[EMAIL PROTECTED]> wrote:
> So you removed LOCAL()? Without sync machines will diverge. Perhaps This includes the machine with local clock as reference, so the result is all diverge in a similar way, relative to true time. > enough to take longer to sync during short online sessions. Or worse Firstly, NTP is not intended to be used with intermittent connectivity. In most cases, the delays will be very similar as the lowest stratum machine will still have to lock onto the valid server before the others can start locking onto it. > continue free wheeling but on a this anormal frequency. Finally a client > (re)booted will have false time until next dialup, if ntpd has not died > before on "no servers reachable". Not good. This is a bit of a problem, but is what hardware clocks are for. > My advice for you would be to set a 127.127.1.0 stratum 8 prefer on your > server, and another one stratum 14 (but not prefer) on a given client. Normal guideline is stratum 10, but local clock should always be a conscious decision. I don't believe it should be prefer here. If you include a local clock, you ensure you have enough real servers to outvote it in the intersection algorithm. Addition of a local clock here breaks a safety feature in NTP. If that safety feature weren't important, NTP would not drop to stratum 16 on losing its server. Local clock on a leaf node is never a good idea, as it serves no useful purpose, but can confuse the the intersection algorithm. A leaf node will coast at exactly the same frequency with and without a local clock configured. > This way you have eternal sync, only one sync source at a time, and you > can shutdown server *or* given client for maintenance. You can, and probably should have, multiple local servers, but, in any case, the client will coast through just the same with or without a local clock configured. > Mail-Copies-To: poster Please use a valid email address or remove this request from your headers. Normal netiquette is not to copy poster. _______________________________________________ questions mailing list [email protected] https://lists.ntp.isc.org/mailman/listinfo/questions
