RE: Comments on draft-nordmark-multi6-threats-01

[jnc@mercury.lcs.mit.edu (Noel Chiappa)]

    > From: "Fleischman, Eric" <eric.fleischman@boeing.com>

    >> identifier  - an IP layer identifier for an IP layer endpoint
    >>			(stack name in [NSRG]).

    > I am bothered by this definition. The way I read it, it implies that an
    > IP layer endpoint is the protocol stack. That is true of every protocol
    > system that I know (including OSI) except for the Internet protocol
    > suite. For the Internet protocol suite, the IP layer endpoint is a
    > network interface.

To start with, terminology nit - and lest you think terminology nits are not
worthy of notice, let me yet again roll out one of my faourite quotations:

  "I am far from thinking that nomenclature is a remedy for every defect in
  art or science: still I cannot but feel that confusion of terms generally
  springs from, and always leads to, confusion of ideas." 

    -- John Louis Petit, "Architectural Studies in France", 1854

I'm going to complain about your use of the term "IP layer endpoint" above,
because you've seriously warped the definition of "endpoint".

Where you said "IP layer endpoint" I think you mean to say something like
"the thing(s) named by the name(s) of the existing IP layer" - so, to restate
your line above: "the thing named by the names of the existing IP layer is a
network interface". (Which is not quite true either, but I'll ignore that for
now.) Or perhaps you meant "the fundamental entity of the IP layer is a
network interface"; I can't be sure which.

In either case, an "endpoint" (as originally defined in:

    http://users.exis.net/~jnc/tech/endpoints.txt

about a decade ago) is a name for the entire communicating entity, including
the transport protocol(s), etc. So to use "endpoint" in any other way, as you
did, is really confusing.


    > Here is where the difference is manifest: for OSI and other protocols
    > with a NSAP (network service access point -- i.e., a stack interface at
    > the network layer), every network interface on that device has the same
    > network layer address. However, for the Internet protocol suite, every
    > network interface has a different IP address. This is why you need to
    > eliminate the text "(stack name in [NSRG])" from your definition.

This is like saying "we need to eliminate family names and replace them with
personal ID numbers".

The thing is that family names *aren't* names for individuals, so there's no
way any kind of name for an individual (no matter what its syntax or
properties) could replace them. Family names name groups of individuals, not
a single individual.

Similarly, these "identifiers/stack-names" name a *different kind of thing*
than network interfaces - which is why they have different names.


    > has multi6 actually considered redefining IP addresses to actually
    > become stack interfaces at the IP layer? If you have, I think that you
    > may agree with me that most/many of the *node* multihoming problems go
    > away.

Alas, we need to explicitly recognize (and name) interfaces *as well*, and
give different interfaces different names, for multi-homing to scale to large
sizes.

This is because if all the interfaces have the same name, then since
interfaces are where the routing system sends packets (i.e. it looks at
interface names in packet headers, to get packets where they are going), then
the routing has to track interfaces (because if interface "X" appears at N
widely-separated places in the network, it has to keep track of where at
least one "X" is, and how to get there) - and that doesn't scale.

Put another way, the routing has to have 'routing-names' which are
topologically sensitive; i.e. have to information about *where* they are
encoded into them. Therefore, the routing-names for two interfaces which are
connected to different parts of the network cannot be the same. Therefore, if
you want to have one name apply to an entire host, including several
interfaces, it cannot be a routing-name.

	Noel