Michael Thomas wrote: > Randy Bush writes: > > > Maybe if you and Randy stopped playing elliptical > > > word games we could have some communication > > > here. If you have a point, make it. > > > > keith did. it's just that you didn't like it. > > > > the reality is that today, identifiers are addresses are > routing. > we may or may not like this, but it is fact. > > Fine. The reality is NAT and RFC1918 too. That > doesn't mean that we should just give up. And > unlike the intractable set of problems that drive > NAT's, it's trivial to djinn up collision > resistant names which can be used in lieu of the > abdication of thought implicit in recvfrom() as > the Bearer of Names. Diffie Hellman is your > friend. What many are missing here is that this is not about 1918 style addressing. This is about the fact that addresses do not have the same visibility and accessibility throughout the network. This operational reality causes the affect we have labeled scoping. Many on this list 'don't like' to acknowledge that scoping exists in the network, and some claim that the network has failed because it doesn't support a global flat routing space. We should not be focused on likes & dislikes (as some recent voting has done), we are here to do engineering. Shooting the messenger does not solve the problem. The network has always had the operational capability to limit where any particular address is accessible from. This coupled with the application shortcut that an interface topology locator equals an endpoint identifier has been the reason that supporting multihomed nodes has been difficult. With the rollout of IPv6, every node becomes multihomed (in this context that means the node actually has multiple topology locators simultaneously). The debate over the last few months on the IPv6 list has been about trying to restrict any given node to having addresses of a single scope. This attempt to force an operational model that doesn't exist in the network today, simply to preserve an invalid (granted longstanding, but still invalid) assumption that any address can be used equally throughout the network, has met with resistance from the operators of the edge networks. Our task is to look at the overall system the way that network managers really run (or want to run) it, then figure out what it will take to make that happen. The first thing one will realize is that any process that passes an address outside its scope of relevance as defined by the local network manager is not working to achieve that goal. One class of processes that do this today are name to address resolution services, another class are multi-party apps that pass a topology locator rather than an label that can be turned into a topology appropriate locator. At the same time we are working on making the system work the way network managers want to run it, we need to find ways to keep the applications as simple as possible. The combination of those will likely result in a more complex infrastructure than we have today, but no more complex than it should have been for the last 15 years. I don't know what a solution looks like, but I do know that continuing to bury our collective head in a dark place will not make the problem that needs solving go away. Tony