The PROPOSAL then is that, after NAT-PT has been deprecated to
experimental, some work on v6-to-v4 translation MAYcontinue.
My question is "what is hard about that".
Any translator has all the issues of NAT, and is *not* going to try to
insert units of 12 bytes into application data, or making IPSEC work.
Therefore, this doesn't begin to work for any application that has a
hard time traversing a v4-v4 NAT.
Having limited ourselves to those applications for which it works
without changing application data, there are four requirements:
- that DNS be made to work (the far device is only advertising an A or
MX record; we need to set up relevant state and advertise a AAAA record
referencing the translator)
- that the assigned IPv6 address correlate with the IPv4 address+port
in question (let me think hard, I'll bet the IPv4 address+port get
embedded in the IPv6 address)
- that the router not be required to re-assemble datagrams (use Path
MTU, MSS modification, etc to effect this)
- TCP/UDP/SCTP works (adjust the transport checksum appropriately)
I'm not opposed to necessary work being done, but I'm not so sure what
work needs to *be* done. We seem to have a lot of transition efforts
around, but there must be a way to bring them together into a single
simple solution. Bring us that one simple path, and then we'll talk.