My research group, as well as everyone who currently uses PlanetLab (and presumably the future GENI platform, if it comes to pass) faces a different deployment scenario than what the operational folks are used to. Setting up anycast might be possible, but is operationally very very difficult for lots of (partly non-technical, partly technical) reasons. Hence the genuine question on where the limit stemmed from. > >A quick question: Right now, we'd like to have a domain delegated to a > >large number (say 100+) of nameservers. The registrars we have gone > > The terminology used here indicates a need for a deeper understanding of DNS. My terminology is correct, and your message is a simple ad hominem. You should point out the error if there is one, and demonstrate how it translates to a lack of understanding. Earlier in the thread, I addressed people who claim to understand distributed systems, who could not properly define the term "Byzantine," and we did not stoop to this level. Someone with a spine would offer an explanation or an apology, but I'm not holding my breath on either count :-). SB>I suspect that he is deliberately trolling, in order to prove a point SB>(that DNS is too limited to handle domains who need a lot of SB>reliability). Stephane, I say enough controversial things as it is that you can (perhaps?) argue over what I actually said, instead of making up random and wildly inaccurate stuff. I don't believe "DNS" is too limited. The wire format is very flexible. > Your parenthetical comment is contrary to one of the most important > principles in the DNS, coherency. Especially high up in the > hierarchy. Did you make this principle up or is it enunciated somewhere? There are lots of places where it is violated. > Of course, the 13 name limit does not limit you to 13 name servers. > With anycast, the number can be unbounded subject to the concerns > with routing. I asked because global anycast is not always feasible, esp. when the hosts have already assigned IPs, and it is not easy to assign new IP addresses to them. > And if you add in load balancers you can have even > more servers. This spreads the load within a point of presence, but won't improve service by expanding the bottleneck prior to the balancer. Gun. _______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf