On Wed, 2008-12-03 at 11:09 -0500, Steve Kostecke wrote: > Cal Webster said: > > >Please explain what a "peer loop" is or point me to the doc page that > >explains it. I don't see the disadvantage of having common peers. > > "peers" unfortunately has multiple meanings. The remote time servers > that an ntpd is synced to are often refered to as peers. But these ntpds > could be in a peer (bi-directional) or server (uni-directional) > association. > > Systems which are in a peer association exchange their notion of time > with each other. > > In a server association the "client" requests the time from the > "server". > > >So, the new behavior is to select no candidates if they have common > >peers? > > A peer association will be discarded if both peers are have selected the > same "sys_peer" (i.e. they are both synced to the same time source).
Ahh! Okay, now I get it. As is often the case, context is everything. > >This would seem counter-intuitive from a reliability/redundancy > >standpoint. What will this ver 4.2.4 server do when it can no longer > >reach the master server? My guess is that it will either stop serving > >time or die since it has no other reference (local undisciplined clocks > >not configured). If true, this is an undesirable condition, and seems > >especially wasteful when there are other time sources available even if > >they may not be as precise as the master. > > An ntpd keeps running and continues to discipline the system clock > with the last known frequency correction when all time sources become > unreachable. So, no, it won't fall over. But what will happen is this > ntpd will no longer be able to serve time to others unless it is > configured with (a) the Undisciplined Local Clock or (b) Orphan Mode. Alright then... that makes more sense. Since all servers must be capable of Orphan mode in order to participate, my configuration will have to be based on undisciplined local clocks as backup references. I'm scheduling OS upgrades for our local servers throughout January 2009. We'll move to the Orphan mode configuration once we are able to host compatible versions. I also hope to have a GPS time source rolled out in the first quarter too. > One thing you need to keep in mind is that the associations are not > static. ntpd evaluates them at each poll and will select (as sys_peer or > candidate) or discard associations as conditions change. Thank you. I have observed this behavior too. > In your case the remaining servers will select the reachable server with > the lowest stratum Undisciplined Local Clock when the master (server > A) is unreachable. Assuming that you restarted ntpd with the revised > configuration files. Good. You have alleviated my concerns. Thank you for your detailed explanations and your time. ./Cal _______________________________________________ questions mailing list [email protected] https://lists.ntp.org/mailman/listinfo/questions
