On Jul 23, 2008, at 10:27 PM, Pekka Savola wrote:
Why is using the mapped addresses on the wire such a holy grail of
v6-only operation?
As a destination address, it might result in IPv6 DFZ being polluted
with the corresponding v4 routes except if you only use it in very
restricted environments (e.g. default route only).
I really don't think that this is a serious problem in practice. As
long as we need to exchange packets between the V4 and V6 networks, we
will have gateways at strategic points that do this. I was just
observing the other day that I simply never use a routing table with
anything other than a default route in practice anymore.
I'm going to make an extremely naive observation, because I haven't
been here through all the big battles over IPv6 addressing. When I
read Alain's draft I concluded quickly that it didn't meet my needs -
it's more something that he needs. But I do see a need for 6-to-4
NATting nevertheless, and I am naive as to the reasons why people
don't like 6-to-4 NATting (other than the ones that apply to NATting
in general).
The observation is that an IPv6-native node could, when an application
opens a socket and connects to an IPv4 address (or sends a datagram)
form an IPv6 packet with an IPv4-mapped destination address. This
would require no change to the application - in many cases it would
Just Work, and in those cases where it wouldn't, the application
really can't be done without being IPv6-aware anyway.
This IPv6 packet would be routed by the native IPv6 routing
infrastructure to some kind of 6-to-4 NAT, which would repackage the
packet as an IPv4 packet and send it on its way, spoofing the source
address as NATs unfortunately must do.
I wouldn't say this is the holy grail or anything, but right now I
simply can't use IPv6-only nodes. It's not practical. So even an
ugly bandaid that makes it possible for me to deploy an IPv6-only node
that has all the functionality of a NATted IPv4 node, plus end-to-end
IPv6, is a huge win over what I would be doing instead - an IPv4 NAT
with net 10 or 192.168 IP addresses.
So while I would describe this as an ugly bandaid, it's a lot less
ugly than what I'm using now. But as I say, I'm probably being naive.
_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area
