On 25 jul 2008, at 17:11, Rémi Denis-Courmont wrote:
So, the question is: is a dual stack host a v6 host (i.e. changes are acceptable) or is it a v4 host (i.e. changes are not acceptable)?
Regardless of the current requirements, I would argue it is not acceptable that dual-stack nodes need to be modified, for the exact same reason as it is not acceptable that IPv4-only nodes need to be modified: deployment would behighly impractical.
Note that these issues aren't necessarily binary.If a dual stack host is presented with synthetic AAAA records along with A records, the host will try to use translated connectivity when it could have used native connectivity. The requirements require that this is avoided. One way to avoid this is to upgrade hosts so they automatically prefer real A records over synthetic AAAA records. This is rather inconvenient, bu there are only 10 - 20 operating systems in the world. Another way to avoid this situation is to make sure that dual stack hosts are never served by a recursive nameserver with DNS64 functionality. However, there are many thousands of ISPs and hundreds of thousands of organizations running significant networks in the world, which would all have to do this.
So yes, requiring host modifications is bad, but imposing operational restrictions has the potential to be much worse in time and money. On the other hand, everyone can change their own network, while few people can change their OS.