> it's rather odd to argue that a technology that partitions the network > into separate realms, such that the hosts in those realms > can't talk to > each other except by coincidence, is allowing us to maintain global > connectivity.
I entirely agree. One goal of network engineering should be wide support of applications. (i.e., network engineers should try to not make assumptions about what the applications will want, but should design an 'open' architecture that will support any application model, etc.) The network engineer shouldn't have to re-design for every new application that comes along. If he does, then he did a poor job to begin with. That is why ubiquitous IP host to IP host communication is 'better than' an application gateway or NAT approach. From the network perspective, it is a solution that provides open connectivity. (Further, instituting policy can still be done, but the network itself shouldn't be the limiting factor--the policy control point (firewall, etc.) should be.) The overhead incurred in managing NAT is acceptable as a work-around (i.e., a hack) until a real solution is available. In my opinion IPv6 is a real solution. -Benson