Why does INTERNIC require two nameservers?

[Jim Reid]

>>>>> "Mark" == Mark Drummond <wizard at clandrummond.com> writes:

    Mark> "Olivier M." wrote:
    >>  Yes that is right. But some registrar (.at or .de, I don't
    >> remember) requires nameservers on two _different_ C classes.

    Mark> This is absolutely silly ...

It's actually rather sensible. What if there was a routing/peering
problem that caused some zone's name servers - on only one /24 - to be
unreachable?

And have you had any operational experience running a ccTLD registry
dealing with clueless admins who blame you for their inability to get
the basics of DNS configuration right? [I've not, but I know people
who do and the very real problems they get because some registrants
don't have even the basic notion of DNS. Most of these bozos are only
interested in owning a cute name in the likes of .com or .nl or .de.]
It looks to me like the registries that operate these rules are doing
a number of Very Good Things. They're keeping out the clueless or at
least insisting they get a clue. They're helping to ensure that their
customers have decent DNS infrastructure. And they also deter
domain-name speculators who can't be bothered with the hassle of
setting up name servers for the ccTLD name pollution they create.

    Mark> so on the premise that the second one can take over if the
    Mark> first is dead. But why not just have 2 nameservers on your
    Mark> own net? If one goes down, the other takes over anyway
    Mark> regardless of it's physical location. And if you network
    Mark> goes down, well even an external nameserver is not going to
    Mark> help much.

That depends on what's in the DNS data. Who says the RRs in some zone
have to point at IP addresses in the same nets as the zone's name
servers. There are other good reasons for having DNS data available
even if the addresses for those are unavailable. Read RFC2182.