[ppml] The myth of IPv6-IPv4 interoperation, was: Re: Legacy /24s

[Iljitsch van Beijnum]

On 3-sep-2007, at 1:32, Paul Vixie wrote:

> <http://playground.sun.com/ipv6/INET-IPng-Paper.html>.
> i especially like section 11, "IPng Transition Mechanisms", which gets
> funnier every year, even without considering RFC 4966.  consider  
> this gem:

> 	The IPng transition mechanisms ensures that IPv6 hosts can
> 	interoperate with IPv4 hosts anywhere in the Internet up until the
> 	time when IPv4 addresses run out, and allows IPv6 and IPv4 hosts
> 	within a limited scope to interoperate indefinitely after that. This
> 	feature protects the huge investment users have made in IPv4 and
> 	ensures that IPv6 does not render IPv4 obsolete. Hosts that need only
> 	a limited connectivity range (e.g., printers) need never be upgraded
> 	to IPv6.

> that was the face that launched a thousand ships, yes.

It looks like you suffer from the (fairly common) misconception that  
goes along these lines:

     "If only the IETF had made IPv6 interoperable with IPv4,
     we wouldn't have any transition issues."

(See http://cr.yp.to/djbdns/ipv6mess.html )

The assumption here of course is that it would have been POSSIBLE to  
build IPv6 in such a way that an IPv6 host can talk to an IPv4 host.  
Unfortunately, it isn't. Or rather, the only way to do that is for  
the IPv6 host to stick to 32-bit addresses (by setting the other 96  
bits to 0 or some such and having something in the middle do the IPv6- 
IPv4 translation), because that's the only thing an IPv4 host  
understands.

This would work well as long as IPv6 hosts only use 32-bit addresses,  
but obviously, the idea behind IPv6 is that at some point, someone  
starts using an address that uses more than 32 bits. But this would  
be impossible, as there is no way to know whether a destination is a  
real IPv6 host that can handle the extra bits, or a fake IPv6 host  
that really does IPv4 behind the scenes which won't handle the longer  
addresses.

Today, the situation is much cleaner: do IPv4 and be limited to all  
that that entails, do IPv6 and the IPv4 limits go away. The presence  
or absense of AAAA records in the DNS tell you if someone you want to  
talk to can do IPv6 or not.

That is not to say that things couldn't have been better: rather than  
revisit all protocols that touch addresses and replace all the  
instances of 32-bit addresses with 128-bit ones, the IETF could have  
gone from 32-bit addresses to variable length, higher level  
identifiers, i.e., FQDNs, so the protocols became address length  
agnostic.