Tim Shoppa wrote: > Peter Roozemaal <[EMAIL PROTECTED]> wrote: >> There won't be a convergence to a stable set of 4 servers if you >> throw out one of them on a daily basis. You are not even guaranteed >> to get a single stable server. Consider having 5 "optimal" peers, >> all at the same nearby provider; each of those servers alternately >> spend 1 day in the global pool and gets swamped by ntpdate bursts >> during the day. > > Does that really happen - "good" servers getting swamped and thrown > out regularly?
If the one server in the global pool is 0.5 ms worse than the three other servers in the auto-selecting client, it will get thrown out. > I'm not saying that there aren't clients that have stupid patterns > for asking for time, but they only hurt other similarly stupid > clients so I don't see that as fundamental. Stupid clients may interfere with the upstream requests of their servers. IP packets get buffered in routers and at the server. More buffering means more jitter, less accuracy. I have offsets and jitters in the order of a milisecond over my ADSL link. It also takes about a milisecond to put a single NTP packet out on the ADSL link. Peter. _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
