Re: [RAM] some draft proposed definitions

[Scott W Brim <swb at employees.org>]

On 06/11/2007 09:39 AM, Joel M. Halpern allegedly wrote:
> Lets look at two closely related examples:
> Is an IEEE 48 bit "address" really an identifier, a locator, or a
> hybrid.  As far as I can tell, it is strictly an identifier.  It is
> assigned to a particular thing in a way that has no regard to where that
> thing is in any topology.  However, being excessively clever engineers,
> we have then built solutions which allow us to forward packets based on
> that identiifers.  Sometimes even very large solutions.  (bridging in
> all its myriad forms.)  That usage does not change the nature of the field.

Exacto.

Ultimately you need a mapping to a nexthop.  In 802.1d you map
directly from identifier to nexthop.  In other environments you map
from a locator to a nexthop.  Sometimes you need both, sometimes you
it takes multiple lookups, sometimes you do multiple mappings at
multiple points along a path, and so on (a locator need not be the
ultimate end-system locator or even prefix -- it might be an
intermediate point like a tunnel ingress).   But how you get those
mappings can be done by any means, using any kind of information, and
the only thing you can be sure of is the nexthop.

> Hence, I think the difference between an Identifier and a Locator is in
> the semantics, not in whether it can be used to deliver packets.  It
> doesn't seem to me that an identifier suddenly becomes a locator if in
> some region we choose to keep track of where the identifier can be
> found.  

Right, it's about intended use.  Just because something is used to map
to a locator doesn't make it an identifier.

There are two phases to routing, path discovery and path selection.
Path discovery can be aided by dissemination of information and
distributed processing (including most of the routing protocols we
like).  It can be done by probing (PNNI, PCEP, NUTSS etc.).  It can be
done by asking someone who knows more than you do (like DNS).

In the case of SIP you start out with an identifier.  You resolve that
to a locator, and you send an INVITE including your identifier and
locator as payload.

In the case of LISP, you start out with a locator.  At some point that
locator is sent to "someone who knows more", as payload, to map to an
ITR (another locator) and to a nexthop.  This does not make the
original locator an identifier.


On 06/10/2007 12:12 PM, RJ Atkinson allegedly wrote:
> Identifier:    An object that is used only for identification,
>         never for forwarding packets or determining location.

I like this except to avoid ambiguity I suggest leaving out
"forwarding packets".

> Locator:    An object that is used only for forwarding packets
>         or determining location, never for identification.

Certainly the last part is true, but locators are used for all sorts
of things.  I propose "A locator is a name intended to have
topologically meaningful semantics, in order that it can be used to
determine a next hop for a packet."  That allows other uses.

> Address:    An object that with mixed semantics, where it is
>         sometimes used for identity and sometimes used
>         for packet forwarding.  Examples include but are
>         not limited to IP addresses, which are sometimes
>         used to forward packets and sometimes used for
>         identity (e.g. in TCP session state).

Only if we're talking about the current Internet.  Let's avoid using
this in new architecture discussions.

> Scoped Locator:    A locator that has non-global scope.  Note that
>         a scoped locator only has location semantics,
>         never identification semantics.
> 
> Scoped Identifier:  An identifier with non-global scope.  Note that
>         a scoped identifier only has identity semantics,
>         never location semantics.

Don't like them.  Everything has scope by nature.  Sometimes that
scope is global, sometimes not.  Just because we've assumed global
scope in the past doesn't mean that was the right thing to do.

> Identity:    A possible property of an object.  Addresses and
>         Identifiers are examples of objects that have
>         identity semantics.

Let's stay away from identity unless we really need it.  I just spent
two years listening to arguments about it.


On 06/10/2007 23:34 PM, Tony Li allegedly wrote:
> This would be a fine addition to the taxonomy draft, if it isn't there
> already...

Here is my current draft text.  Lixia hasn't had a chance to hack it
up.  All comments appreciated ... :

      <section title="Name">
        <t>A "name" is "an identifier (of no specific syntax or
	  properties) for an object" <xref target="ENDPOINTS"></xref>.
	  "Name" is a general term, and includes more specific subsets
	  such as "locator" and "identifier".</t>
      </section>

      <section title="Locator">
        <t>A locator is a name intended to have topologically
	  meaningful semantics, in order that it can be used to
	  determine a next hop for a packet.  </t>
	<t>Just because a locator is intended to be used in forwarding
	  packets does not mean that it can be used to do so in every
	  part of the Internet.  Locators have scope.  Even when a
	  locator is out of its scope and thus cannot be used for
	  forwarding (it cannot be resolved to a next hop), it is
	  still a locator if it is intended to be used in that
	  way.</t>
	<t>Locators which cannot be resolved to a next hop are said to
	  be "not routable".</t>
	<t>A locator may name an ultimate destination attachment
	  point, or it may name an intermediate point on the way to
	  that attachment point. For example, a locator may name a
	  tunnel ingress point which can be used to reach the ultimate
	  destination.</t>
        <t><list hangIndent="7" style="hanging">
            <t hangText="NOTE:">It has been suggested that "locator"
	      by itself implies global scope, and that if the scope of
	      a locator concept is less than global, it must have an
	      adjective associated with it, for example "routing
	      locator". However, it has also been suggested that
	      "locator" by itself is a generic term for all locators,
	      regardless of scope. To be resolved.</t>
          </list></t>
      </section>

      <section title="Identifier">
	<t>An identifier is a name that is used for identification,
	  never for determining location.  It does not have
	  topologically meaningful semantics.  It names an endpoint of
	  a communication. </t>
	<t>An "end system identifier" names the ???TBD??? layer for
	  that end system. </t>
	<t>"ID" is an abbreviation for "identifier".</t>
      </section>
      <section title="Address">
	<t>Before the idea of separation of identifier and locator
	  became common, "address" was used primarily to refer to
	  locator functions but also to identifier functions.  Because
	  the term is ambiguous, its use is deprecated except when
	  talking about the current Internet</t>
	<t><list style="hanging" hangIndent="7">
	    <t hangText="NOTE:">It has been suggested that "address"
	      should continue to be used for a name that functions as
	      both a locator and an identifier.  To be resolved.</t>
	  </list>
	</t>
      </section>




_______________________________________________
RAM mailing list
RAM at iab.org
https://www1.ietf.org/mailman/listinfo/ram