Thanks for the explanation, Ask. My stats have settled down now. Changing my netspeed from 1.5Mbps to 100Mbps resulted in my traffic going from 15 reqs/second to 19 reqs/second and my unique IPs / second (in a 10 minute window) to go from about 3.8 to 4.2. Unless my analysis is wrong that suggests that increasing my speed 60x resulted in only about 1.3x more traffic after three days.
I've got two explanations for that.Maybe the weighting of netscores isn't affecting the DNS rotation like you intended. Or maybe hosts aren't looking up DNS entries very often and so haven't been exposed to my new weighting after three days. I kind of like that second explanation since it's consistent with what we know about how ntpd works. Do we know how often a typical pool user re-resolves pool.ntp.org? We know a few clients stick to an old IP address for months, but what do most of them do? I guess your proposed experiment of taking myself out of the DNS entirely would give us a graph of that. Does anyone already have a requests/second graph for some server they removed from the pool? If not I'll try it out. Ryan Malayter wrote: > Wouldn't it make sense to weight by a logarithmic or other sub-linear > function of the netspeed instead? We surely wouldn't want a pool > server with a gigabit connection receiving 667 times the traffic of a > pool server behind a T1. I'm incline to agree with that. _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
