On Oct 7, 2010, at 1:32 PM, David Conrad wrote: > Keith, > > On Oct 7, 2010, at 4:32 AM, Keith Moore wrote: >> As currently defined, IP assumes a global address space that is used consistently throughout the network, > > I actually think it's a bit worse than that. As currently defined, IP assumes a global address space in which each individual address has the potential for being topologically significant. Topological aggregation to permit scaling was an afterthought that doesn't fit particularly well into that architecture. Fair enough. But very few applications out there make assumptions about topology based on address assignments. If IP address assignment suddenly became non topologically significant, hardly any applications would break. Routing would have to change, but assuming you had a way to do it, such a change would be far less disruptive than say IPv4->IPv6 transition. >> Who is to say whose prejudices are right? > > If it doesn't work (for some value of the variable 'work'), it's fairly clear it's wrong. >From that point of view, everything in IPv4 is wrong. Using NATs is wrong because it doesn't work for some apps; not using NATs is wrong because it doesn't provide enough address space. In other words, the problem of making IPv4 continue to be viable is (probably inherently) overconstrained. >> Given people's reliance on the Internet, the idea that we can throw out the existing (non-)architecture and replace it wholesale with something new is mere fantasy. Even back with IPng was being chosen, the assumption that this would be possible was probably a core mistake. No disagreement there. Clearly you can't replace the internet architecture wholesale, but it might be possible to evolve it. What doesn't work well is to have everyone decide for himself how to change the architecture. Keith _______________________________________________ Ietf mailing list Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf