CircleID

Regarding the decoupling of 'address' and identity - it's already happening; where
'address' is the location of a device and identity is made up of the services a user has subscribed to.
A person's on-line identity is not a single attribute, but it's made up of the mailing lists and groups they are subscribed to, their Skype id, their avatar, the applications in their dock, toolbars and menus, their favourite web pages, etc.
All that's missing is the portability of the identity - just a small thing :-}

Hi Geoff,

Great article! However, I'm curious about your decision to highlight the "200 to 400 million uniquely addressed IP devices" that are currently supported by the IPv4 addressing regime, without mentioning the many hundreds of millions of dynamic user/access processes that are also currently supported by IPv4. My concern is for readers who might take your original number as a comprehensive reckoning of the total carrying capacity (or perhaps even the total required carrying capacity) for an internetworking addressing system, and in so doing completely miss the importance of "uniform equivalence" across both core or mid-path devices and edge or access devices and processes. As you rightly note, the material fact of this equivalence has been eroding for over a decade as a result of macro-level protocol changes (DHCP, NAT, etc.) as well as micro-level policy innovations (IP filter lists, selective port/service filter lists, etc.). Even so, I would argue that "uniform equivalence" remains an important normative architectural property, deserving (at least) of recognition along side your other essential properties. After all, it's this property that invests IP addressing systems with the essential qualities of end-to-end systems design; it's this property that permits (where it exists) end users to choose their own devices and applications—or even invent new devices and processes—independent of the commercial designs of their network service providers. Arguably, it was this property that accounts for the Internet's phenomenal growth rate over the past two decades, and its increasing centrality in so many aspects of life today—not to mention the general enthusiasm with which the Internet has been sought and embraced by (almost all of) the billion-and-counting users and would-be users around the world.

"Uniform equivalence" (UE) may be an endangered feature in current IP addressing, but I would argue that it is still embraced as the norm, and not merely by naive idealists like me ;-) Without some visceral attachment to UE, there would be no reason for anyone—e.g., any knowledgeable end user—to think that DHCP with public IP addresses is preferable to DCHP with NAT, or that the latter is preferable to DHCP with NAPT. Just because (even) one of these features "breaks" some of bindings that can make public IPv4 addressing uniquely flexible and adaptable, doesn't mean that things can't become "more broken" when more than one is applied concurrently.

Finally, I think UE deserves independent consideration because it is not (or perhaps no longer) reducible to your other essential features (uniqueness, consistency, persistence, etc.). For example, it's quite conceivable that the next-generation of (IPv6) addressing will permit each and every user, interface, edge and core device everywhere to be supported with one (or possibly many!) unique, consistent, and persistent public IP addresses. It's equally conceivable that this achievement will be accompanied by the deployment of next-generation (IMS) hardware that will effectively break all of the old "given" bindings, and subject each and every packet and flow to close inspection and potential interdiction by any operator along the network service path.  Given the fact that almost all Internet service paths must still traverse at least one critical bottleneck (the "last mile" facilities platform), competition alone cannot be expected to sustain the transparency and flexibility of the "old Internet" unless and until it reaches down to that level. The proximate cost and benefit proposition for such a move might be very appealing (especially to any commercial entity that commands one of those bottlenecks), but what would be lost—flexibility and adaptability, decentralized freedom of innovation—would literally be priceless.

Of course, the cynical realist in me recognizes that embracing UE as a design principle for IP addressing could lead to many hypothetical contradictions and paradoxes—routing is never guaranteed, every operator is autonomous, preferences and policies are defined locally, notions of "equivalence" can be taken to absurd extremes (if only for rhetorical purposes), etc., etc. Even so, the naive idealist in me is quite happy to engage in those debates; an Internet that does not even aspire to UE would likely be a sadly impoverished place…

On Layer Three, no one knows you're a dog (or a host, or a router).