Re: (ngtrans) Joint DNSEXT & NGTRANS summary

[Paul A Vixie <vixie@vix.com>]

> 	i have a major concern with AAAA synthesis - which is, it is unclear
> 	as to who needs to AAAA synthesis.  the concern is mentioned
> 	in my draft.
> 
> 	- you can't guarantee every first-hop DNS server to do AAAA synthesis.
> 	- if anyone does not, AAAA queries go into non-first-hop DNS servers
> 	  by recurse (imagine when pre-BIND9/non-BIND name server is used
> 	  as the first-hop name server).
> 
> 	therefore, AAAA synthesis basically asks everyone to run AAAA and A6
> 	in parallel, which raises a lot of concerns (query delays if you query
> 	both, database maintenance cost if you maintain both in zone, no-sign
> 	if you synthesize, and a lot of others).

i would much rather deal with those problems, since they are solvable, then
to deal with non-renumberable 128 bit addresses, since they simply lead to
NAT.

i cannot emphasize strongly enough that vast numbers of transit-consumers
who lack the various powers needed to own portable address space will NOT
deploy globally addressable IPv6 space if it simply means that they will
become even more captive of their ipv6-transit providers than they are of
their ipv4-transit providers today.

the gprs and other mobile data providers are mostly going to be able to
own portable address space so they don't care about this issue.

but for "the enterprise" this is a real problem.

if ipv6 is no easier to renumber than ipv4, then we're going to see large
scale NAT.  much larger scale NAT than we see today.

being a captive customer of some transit provider when you only have an
ipv4 /24 or even a /20 that you have to renumber if you change providers
has difficulty N.  being a captive customer of some transit provider when
you have an ipv6 /64 to renumber if you change providers has difficulty
N*(2^44).  if NAT is used then the difficulty in either case is a constant,
and for that matter, it's the same constant.

so, "let's standardize both and let the market decide."

but right now i'm hearing three camps trying to kill A6:

	implementors who consider bitstring, A6, and DNAME "too hard to do"

	transit providers who like the idea of captive ipv6 /64 customers

	transit customers who have enough power to own their own ipv6 blocks

who in this mix is representing the interests of the average MIS manager?
(and who do you think is going to drive the ipv6 economy, anyhow?)

this debate borders on "simply unbelievable."