Re: (ngtrans) Joint DNSEXT & NGTRANS summary

[Johan Ihren <johani@autonomica.se>]

Jun-ichiro itojun Hagino <itojun@iijlab.net> writes:

> >If we look at DNS configuration for the v6 part of the world it is at
> >present based upon various kinds of ugliness to get around the fact
> >that there is almost no hierarchy deployed over v6 transport (starting
> >with the lack of roots accessible over v6 transport). 
> >
> >The most common is to use a forwarding configuration to escape over to
> >the v4 transport universe with full DNS connectivity.
> >
> >*All* this will have to change over the coming years if or when or
> >wherever we decide to start deploying roots accessible over v6
> >transport.
> 
> 	I don't get your argument at all.
> 
> 	Because first part of the IPv6 transition will be toward IPv4/v6
> 	dual stack, AAAA over IPv4 transport works just fine for us.
> 	The history proves that it works fine for us - due to the lack of
> 	IPv6 accessible root, IPv6 accessible ccTLD/gTLD, and lack of IPv6
> 	NS record registration support by many of the registries, we IPv6
> 	users are forced to rely upon IPv4 DNS infrastructure.  Luckily,
> 	it works quite well.
> 
> 	total transition to AAAA (or A6) over IPv6 transport is of course
> 	better, however, it needs a major upgrades, including:
> 	- IPv6-ready root
> 	- IPv6-ready ccTLD and gTLD
> 	- IPv6-ready random .com servers
> 	all of them needs to be done before the total transition, so it is
> 	unrealistic to talk about total transition this time.
> 
> 	So, what I'm saying is, AAAA deployment is already so wide enough
> 	(all BIND4/8/9 do support them okay), and for me it is way too late
> 	already to transition to anything else (including A6).  If we are
> 	to transition to something other than AAAA, the transition needs to
> 	be coordinated very delicately, like transition from IPv4 to IPv6
> 	and it will take multiple years to do so.  We cannot wait any more.
> 
> 	Think DNS transport issue (DNS over IPv6, or IPv4) and DNS payload
> 	issue (AAAA, A or A6) separately.  What we are talking about is
> 	mainly the payload issue, though, it affects the deployment of
> 	the IPv6 DNS transport (as NS records need to point to AAAA, or A6).
> 	I guess you are mixing up these two.

No, I am not mixing them up. 

I'm pointing out that due to the configuration change that will follow
from deployment of v6 roots it will be comparatively easy to reach all
the caching resolvers (i.e. name servers) on v6 transport to ensure
that they all do AAAA synthesis.

I agree that this will not by itself help the deployed base of
AAAA-aware, A6-ignorant caching resolvers that are v4 single stack.

Furthermore there are obvious problems with doing AAAA synthesis in
response to non-recursive AAAA queries, so we cannot shield those
servers easily by doing the synthesis in the authoritative end.

However, while agreeing that the deployed base of AAAA-aware caching
resolvers on the IPv4-only Internet is large it is less clear to me
how many of those actually care about v6 address records at all. Nor
is it clear how many of those that will start to care before having
upgraded their name server software (which contrary to popular belief
actually happens, although certainly not to all of them).

It is certainly a non-trivial number, but as Bill touched upon,
somewhere there is a price to evolution, and I much rather have it
there (among v4-only servers not caring about v6) than tying the hands
of people that care very much about IPv6.

Johan