Re: [BEHAVE] DNS64 detecting dual stack clients

[Brian E Carpenter <[e-mail]>]

Andrew,

I should have started with a problem statement. The administrator
has to deal with a network with some dual stack hosts and some
IPv6-only hosts, only has the blunt instruments of DHCPv6 or RFC 6106
to tell the hosts about DNS servers, and doesn't know which
hosts are which.

Yes, it's a kludge on a kludge. It might just be preferable to the one
I've been arguing with Dan about (draft-wing-behave-dhcpv6-reconfigure),
but it's still a kludge.

   Brian

On 2011-11-09 16:30, Andrew Sullivan wrote:
> This is not to pick on Brian.  But he provides the occasion.
> 
> On Wed, Nov 09, 2011 at 03:13:52PM +1300, Brian E Carpenter wrote:
>> Maybe this has been discussed before, but it seems to me that a DNS64 can
>> figure out which clients are dual stack and act accordingly.
> 
> What is the "act accordingly" that is the right thing here?
> 
> <rant> 
> 
> I am sick unto death of suggestions to make DNS64+NAT64 work more
> reliably.  It is a nasty, 80% solution hack.  I cannot speak for
> anyone else involved with it, but to me it was _always_ supposed to be
> a nasty hack -- useful as a kind of field bandage for the most
> egregious problems caused by the fact that there was no real IPv4-IPv6
> upgrade path.  We live in a world with no more IPv4 network addresses
> and little IPv6 penetration; and also one where "the Internet", for
> many people, means "what I get in my browser."  I have taken as my
> motto a sort of Homer Simpson one: "Aim low."  If we can mostly fix
> things most of the time for the most common case, it's useful.  It's
> still a nasty hack, but it's useful for now.
> 
> </rant>
> 
> In my opinion, "act accordingly" is to get out one of those
> muzzle-loading-cannon-on-scissors that is so popular in Warner Bros
> cartoons, and shoot it at the network administrator who configured
> some poor slob's computer with a dual stack and a NAT64.
> 
> Any other case -- like solving the case where I somehow accidentally
> stumble onto a NAT64 network with an unmolested IPv4 connection that
> nobody knows about -- sounds to me like an optimization that shouldn't
> be considered.  If the complaint is, "Yes, but then people will use
> NAT when they don't have to," I express my doubts.  An infinitesimally
> small number of people use unmolested, NAT-free IPv4 today.  Those
> people already know how to cope with IPv6 and how to avoid sludge like
> NAT64.  Everybody else just has a crappy choice of NATs to make.
> 
> If the argument is instead that we can make protocols with embedded
> IPv4 literals in them work marginally better this way, then I suggest
> the additional complications are not worth making for protocols that
> are wrong in their conception to begin with.
> 
> Please, let us not fix something that is broken exactly as much as it
> ought to be.  The pig is wearing enough lipstick.  We should put away
> the cosmetics now.
> 
> A