[rrg] No liveness requirement in the ID/Loc Split concept

[Noel Chiappa]

    > From: jnc at mercury.lcs.mit.edu (Noel Chiappa)

    > a LISP EID does _not_ name an interface, but rather a 'stack' - i.e. an
    > endpoint, with a collection of TCP connections.

A private reply pointed out that a LISP EID in fact _also_ names an
interface, just like an IPv4 address. That's a consequence of the fact that a
LISP design goal was to allow unmodified hosts, so one can only change the
semantics of existing namespaces a little.


Interesting historical note: the Nimrod deployment plan:

  http://ana-3.lcs.mit.edu/~jnc/nimrod/deploy.txt

looked a lot like LISP's deployment plan - the existing IP layer was 'jacked
up' and a new layer inserted underneath; unmodified hosts were unaware that
this all was happening, and continued to emit and receive classic IPv4
packet; Nimrod agent-routers were responsible for invisibly translating the
IPv4 'addresses' in those packets into Nimrod locators, and doing the
encapsulation. Although the Nimrod deployment plan spoke of turning IPv4
'addresses' into endpoint names (EIDs, to be exact - ironically), it's clear
that there, IPv4 addresses also retained minimal interface naming semantics.


It's almost inevitable, in any plan which supports unmodified hosts, that
the IPvN addresses used by those hosts will retain the basic semantics of
IPvN addresses.

	Noel