I think we should hereby distinguish between data plane and routing plane
and ask
what does incremental deployment mean in either case.
Associated is the question how backward compatibility is enabled, again in
either case.
All of this needs an answer, and yes, also a good solution, but at the right
time.
It should not be the number one issue at the beginning. At first comes the
architectural design, then come the details which include "what needs to be
transmitted when and how".
Wrt data plane:
According to the resulting model, the details of a new IPvN header have to
be specified (whatever that will be). We should not go after an architecture
which assumes that IPv4 resp. IPv6 as defined in the last millenium are the
best ever and must therefore be kept unchanged.
Obviously a new IPvN header will be identified by a new protocol type. What
ever its other details will be, for reasons of backward compatability, a "new"
router must be able to convert it back to some IPv4 header, if the
forwarding-to neighbor is not as knowledgeable - at least for a certain subset
of
destinations.
Wrt. routing plane:
A new routing model could be installed completely apart from the existing
ones (BGP,OSPF).
It wouldn't happen for the first time: Sailors know to use the stars, the
sun, the compass , the GPS for orientation. Not always all possibilities are
available.
I am pretty sure, that after the model has been determined, acceptable
enhancement as well as acceptable compromises will be found for solving the
backward compatibility issue.
Heiner
