Christian Huitema writes:
As we transition to IPv6, we will see three kinds of nodes: a large
part of the unreconstructed installed base will be IPv4 only, many
hosts will be dual stack, and many others will only have IPv6
connectivity.
The current reality is that people who ``only have IPv6 connectivity''
are not on the Internet. An ISP that tried to sell an IPv6 address as if
it were an Internet address would be laughed out of the marketplace. No
sane company would set up www.company.com reachable only through IPv6.
IPv6 addresses are vastly less useful than IPv4 addresses.
The magic moment in the IPv6 transition will be the moment when this
changes: the moment when typical Internet computers can talk to IPv6
addresses.
How do we make that happen? There are several big deployment problems:
upgrading network stacks; upgrading applications; upgrading routers. Do
we also have to suffer the administrative hassle of setting up an IPv6
address for every IPv4 machine?
Under the RFC 2893 transition plan, the existing IPv4 address space is
included in the IPv6 address space. If this is done properly, it can
entirely eliminate the administrative hassle.
If, on the other hand, the IPv4 address space is _not_ included in the
IPv6 address space, then the magic moment won't happen until the entire
IPv4 universe has acquired IPv6 addresses.
People who characterize IPv6-over-IPv4 tunneling as a local decision are
massively confused. The big benefit appears if _everyone_ includes IPv4
addresses in the IPv6 address space.
This is one of the most basic questions about the IPv6 architecture. It
should have been settled years ago.
---Dan