> > no, all that is necessary is that there be a single query for each > > of the "popular" TLDs at each resolver for every time the cached NS > > record for that TLD at that resolver goes away. (that's what I mean > > by "popular"). so an increase in TLDs can cause the load on the root > > servers to increase drastically without the number of end-system > > queries increasing at all - all that is needed is for those queries > > to exhibit less locality of reference than before. > > If the overall number of lookups has not increased, there will be fewer > overall lookups for *existing* TLDs as well. not from the resolvers to the roots. they'll each contribute one hit per NS record that has timed out of their cache for the root zone, regardless of how many queries they handle (>= 1 per zone). > This means that the number of > stale cache hits for existing TLDs goes down at the same time as the > number of stale cache hits for new TLDs goes up (although not necessarily > the same rate). In order to significantly threaten the root load, you > would have to significantly increase the number of queries. that's simply incorrect. > My point is that ICANN only needs to design policies which allow for the > creation of TLDs such as .auto and .car, but they do not need to decide > that specifically .auto gets in while .car does not. Sure, but such decisions still have to be made due to various considerations including but not limited to root server load, and moving that decision away from ICANN to another controversial organization isn't likely to solve any problems that I can see. Ketih