7706 is wrong headed on a number of levels, but its worst offense is to
think that the root zone is special. we need dns to have leases on its
delegation chain including glue and dnssec metadata. everything you
might need to use in order to reach a zone authority and trust its
results should be kept warm. the owner of the data you've leased must
have the ability to reach out and invalidate it in a trusted way.
having .CN's delegation data resident because of 7706 doesn't help you
reach your own EXAMPLE.CN name servers if the network outage you were
concerned about is inside china rather than between china and the rest
of the world.
NFS gave an example of how to solve this, 25 years ago, with NQLEASE. i
am not asking for new computer science, only application of what's
already well understood outside the DNS community, as to keeping the hot
side hot.
unbound has pioneered a bit of this by automatically refetching data
that's near its expiration point. we should work from there outward, by
standardizing it, prioritizing delegation and dnssec metadata, and
exploring ways that the .CN servers could invalidate old NS RRset or DS
RRset data (or indeed DNSKEY and RRSIG) if it was willing to do the work
of remembering who it had handed the now-invalid data to and what trust
markers would be needed to get an RDNS to accept some new form of NOTIFY
to purge its cache.
_that_ would be complexity worth its cost. 7706 was not. HAMMER is not.
indeed nothing which treats the root zone as special is worth pursuing,
since many other things besides the root zone are also needed for
correct operation during network partition events.
the fact that i have to hotwire my RDNS cache with local zone glue in
order to reach my own servers when my comcast circuit is down or i can't
currently reach the .SU authorities to learn where VIX.SU is, should not
only concern, but also embarrass, all of us.
vixie
_______________________________________________
DNSOP mailing list
DNSOP@ietf.org
https://www.ietf.org/mailman/listinfo/dnsop
