On 17-okt-2007, at 22:49, Brian E Carpenter wrote:
In my opinion we have to be very systematic about this. The only way that we can avoid the intrinsic problems of stateless translation seems to be to have awareness at the IPv6 end that translation is happening.
Note that my draft goes in the opposite direction and imposes new requirements on the IPv4 side rather than the IPv6 side. I do agree that having an unmodified IPv4 host talk to an unmodified IPv6 host is problematic if you want to avoid NAT.
Being able to talk to unmodified IPv4 hosts is attractive in the short term because there are so many of them. But being able to keep IPv6 clean is attractive in the long term.
The effort to update either would be approximately the same, in my opinion, because all current OSes support both IP versions. The big difference would be in the appliances such as firewalls and load balancers.
The way I see it, the large value isn't in making a random IPv6 host talk to a random IPv4 host. Rather, it's allowing a random IPv6 host to talk to the infrastructures of large content networks that exist in IPv4 which seem to be having a hard time moving to IPv6.
As far as having the IPv6 host know about the translation: a very simple way to do this is with proxies. Today, applications need to know about proxies but I don't think it would be too hard to make the OS support this. I.e., when an IPv6-only host wants to set up a TCP session towards an IPv4-only host, the stack intercepts the request, connects to a proxy and asks the proxy to set up a session towards the IPv4 address in question. When the connection is there, the TCP session is given to the application.