[ih] IP addresses are not phone numbers, was Some Questions over IPv4 Ownership

[Scott Brim]

On 10/17/2010 06:12 PDT, Dave CROCKER wrote:
> 
> 
> On 10/17/2010 7:34 AM, John Curran wrote:
>> There have been discussions for geographic addressing in IPv6, but
>>    it doesn't improve routing (actually, the converse) unless the path
>>    of connectivity actually follows the geography.
> 
> Deering conducted discussions as part of his original proposal effort
> (the first SIP) that fed into the current IPv6.  As I understand it,
> there was followon research in the area (via Lixia?).  Note that none of
> this surfaced all that publicly, which says a lot about how it progressed.

Tony Hain probably carried it the furthest and most recently.

> My own view is that IP Addresses need to be treated strictly as locators
> and that the infrastructure needs to be tailored solely for that.  This
> is a variant of saying that the current Internet IP model is fine and we
> should not mess with it, except to make it more efficient at doing what
> it has been doing so well.

Well, yes, the current IP model is fine but that doesn't mean that all
routing/forwarding must be restricted to it.  There should at least be a
topological space with aggregatable locators that are globally unique
and globally routed like the current Internet.  That doesn't mean that
there can't be other topological spaces with locators that are not
globally routed (e.g. LISP classic).

> When Deering started his discussions, I felt that geographic addressing
> was essential, in order to buy customers freedom from ISP lock-in.

Joel Halpern and I once had a nice long walk-and-talk on the beach in
Waikiki where I became convinced that while metro-based addressing is
better when everything is simple, it becomes "grubbier" very fast as you
add policy, operational exceptions, etc. to complicate things.

> Per the above view, I now believe that the most powerful path is to get
> "identifier" functions entirely out of the IP Address

Absolutely.  The fundamental problem in locator/identifier separation
has nothing to do with location.  Rather, it's to wean identification
functions away from depending on topology-dependent inputs.  Regardless
of where or at what layer those identification functions are placed.

Scott