Yahoo and IPv6

Sun May 15 19:33:31 UTC 2011

On 15 mei 2011, at 6:29, Matthew Kaufman wrote:

> And that would be the fault of NAT64, which for all of the  
> applications I mentioned (and more) made the seriously wrong  
> assumption that every IPv4 address is looked up in a DNS server.

This brings to mind the story of the physicist (but it could easily  
have be an IETF protocol engineer) who was scrambling around under a  
lamp post at night. A passer by asked what he was doing: looking for  
my keys. Are you sure you lost them here? No, but under the light is  
the only place I have a chance at finding them.

It's not that the people involved with NAT64 (full disclosure, I'm one  
of them) thought that every IPv4 address would have a working DNS  
name, but rather that using the DNS made it possible to have a  
transition mechanism that lets unmodified IPv6 hosts talk to  
unmodified IPv4 hosts.

However, all is not lost: you can easily set up sessions through a  
NAT64 if the application (or the system, but that will take longer to  
materialize) learns the other 96 bits and stuffs them in front of the  
IPv4 bits. If the NAT64 uses the well known prefix the 96 bits are  
easy to learn, if it does not you'll need another mechanism, which are  
now being discussed. But an application could easily roll its own by  
looking up a known IPv6-only A record and then taking the 96 bits from  
the resulting AAAA record.