On Sep 27, 2007, at 8:35, Nelson Minar wrote: Thank you for the stats! It's interesting and helpful.
> I've been playing with my server's "net speed" setting over the > past few > days to see what effect it had on my traffic. The result seems to be, > not much. Am I right? You might be right. The "random logic" is pretty simplistic and when I made it I did consider this, but as far as I could figure then if anything low-bandwidth hosts would get "too little" traffic, which really isn't much of a problem. > My server is 72.36.170.170, I have usage graphs here: > http://www.somebits.com/~nelson/tmp/ntp/one%20week.html. It had > been set > at a net speed of 1.5Mbit for months. I upped it to 10Mbit about on > 2007-09-24 16:30Z , and again to 100Mbit at 2007-09-25 21:30Z. The change should take effect within about 20 minutes (FWIW). [...] > Some of the discussion here suggested going 10 -> 100 would result > in a > 10x increase in traffic. That alarmed me and I'm glad to see it's not > that drastic. Are you sure you weren't getting 10x as much "new" traffic. Remember it should never come in peaks anymore... I suspect most long-running servers end up getting most of their traffic from slowly building up ntpd clients. If you set your server to be deleted within the next ~12 days then you should stop getting new clients altogether - that might be a fun experiment. > But exactly how does the new DNS treat the bandwidth setting? Every 15 minutes or so the pool system updates the DNS configuration data. The big zones are randomly split into 5 (pool, 0.pool, 1.pool, 2.pool, 3.pool, 4.pool). The smaller zones just get the same data in the 5 sub-zones (currently "smaller" is "less than 20-30 servers" if I remember right). When a request comes in the DNS picks a zone (typically just what was requested; if {,0,1,2,3,4}.pool.ntp.org is requested then it tries to pick a country or continent zone instead. When it has a list of zones, the server then picks a random 5 servers to return, weighted by the netspeed setting. Here are the stats on how many servers are in each netspeed (just from all servers, not taking the zones into account): > select count(*), netspeed from servers s where s.score_raw > 5 and deletion_on is null group by netspeed; +----------+----------+ | count(*) | netspeed | +----------+----------+ | 152 | 256 | | 130 | 512 | | 73 | 768 | | 344 | 1000 | | 73 | 1500 | | 65 | 3000 | | 170 | 10000 | | 34 | 25000 | | 32 | 50000 | | 318 | 100000 | | 87 | 1000000 | +----------+----------+ 11 rows in set (0.01 sec) [....] > relatively fewer are sending more than twice a minute. Not a very > interesting result, honestly. I'll check again in a year and see how > overall our client population is doing. Hopefully, just hopefully the clients gets better rather than increasingly broken. :-) - ask -- http://develooper.com/ - http://askask.com/ _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
