On 26-10-07 19:38, Ryan Malayter wrote:
> On 10/26/07, Evandro Menezes <[EMAIL PROTECTED]> wrote:
>> is a lousy server, often being discarded as a false-ticker.
>> But because of its high declared bandwidth, it often makes
>> to the first server in a DNS query.
> I do think that we should weight DNS rotation frequency by the square
> or cube root of the bandwidth, just to be sure that one server doesn't
> soak up the majortiy of local pool traffic.
I guess a good server accepting a lot of traffic (configurable by the server
admin) should get a lot of traffic. There's no need to take the square or
cube root of bandwidth.
Currently the score acts like a switch: you're in or out of the pool. As a
next step the score info might be taken into account. Perhaps we can reduce
the DNS visibility for bad performers, for example
effective_bandwidth = bandwidth / (21 - score)
This way a server just entering the pool at score 5 gets 1/16th of the
traffic it will get at score 20.
Arnold
--
Keeping the load on time servers acceptable (providing time at all) has more
priority than providing time with best accuracy possible.
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
