Andrew Robinson wrote:
Out of curiosity, would it work to "master" every system to its internal clock and "peer" it to every other system on the network?
If you made the internal clocks stratum 1, then peered them, it would be the wrong thing to do. Each machine would think *it* knew what time it was, offering it to others. Like a roomfull of prima-donas. They wouldn't come to consensus.
I've had the best success with a simple chain, having two or more machines slave to an external server, then all the other machines consider these couple of machines servers. That way I minize the load on primary servers. You can use ntptrace to see your timing chain:
% ntptrace localhost.localdomain: stratum 4, offset 0.000008, synch distance 0.32141 ahab.rinconres.com: stratum 3, offset 0.062624, synch distance 0.26173 seismo.cmr.gov: stratum 2, offset 0.124024, synch distance 0.12637 ntp2.usno.navy.mil: stratum 1, offset 0.112434, synch distance 0.00072, refid 'USNO'
I've never really understood the benefit of "peer" for most applications.
-Bob Arendt
-- Shrike-list mailing list Shrike-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/shrike-list
