On Sep 14, 5:42 pm, Unruh <unruh-s...@physics.ubc.ca> wrote: > "rotor...@yahoo.com" <rotord...@gmail.com> writes: > > a) Get one or more Garmin 18xLVC gps receivers and set them up on a few > of your nodes. their time will then be within a few usec of UTC. Use > those nodes as the servers to the rest of your network. > You do need a view of the sky.
A HW time reference isn't feasible, nor should it really be necessary given our modest accuracy requirement. > b) Set up one node to be the master and have it sync to the outside > world. One node can never disagree with itself. On the other had, one > node could die (eg due to someone tripping over it and pulling the > plug). The problem with picking one master node is how to handle node failures. If necessary, it is technically feasible to monitor all the nodes, and reconfigure a new master if the original one failed. But given that the reference ntp daemon has the concept of peer associations, I expected that there was a built-in mechanism for this type of redundancy. Orphan mode sounds like it might be, but I was looking for confirmation of that. > c) Have an external program keep track of the unity of your servers, and > send you a warning if they disagree by more than a few ms with respect > to each other. You mean the external NTP servers? Even monitoring them isn't sufficient, as their could be a communication breakdown between them and the cluster. That should actually be ok, as it doesn't really matter if the cluster's internal time drifts a few seconds or even minutes during any lack of contact with the external server, as long as it can resynchronize when communication is reestablished. But what is of paramount importance is that the nodes agree with each other, with a tolerance of around one second. thanks for your response. Tim _______________________________________________ questions mailing list questions@lists.ntp.org https://lists.ntp.org/mailman/listinfo/questions
