I see everyone has had a productive weekend.

First, about the definitions ...

Excerpts from Tony Li on Sat, Mar 28, 2009 08:44:11PM -0700:
>
> Hi all,
>
> Here's where we are after one round of comments.
>
> Tony
>
>
> locator  A locator is a name that has topological sensitivity
>          and must
>          change

"must change" is too strong, because small changes may not require a
locator to change.  The essential issue is that use of a locator is
topology-dependent, so perhaps the text could say "may have to change"
or "cannot be expected not to change".  Even better would be a whole
sentence explaining the point.

>          if the point of attachment changes.  By convention,
>          a locator refers to layer 3 by default.  It is also
>          possible to have locators at other layers.  Locators may
>          have other properties, such as their scope (local or global
>          (default)) and their lifetime (ephemeral or permanent
>          (default)).
>
> identifier An identifier is the name of an object; identifiers have no
>          topological sensitivity, and do not change, even if the
>          object changes its point(s) of attachment within the
>          network topology.  Identifiers may have other properties,
>          such as the scope of their uniqueness (local or global
>          (default)), the probability of their uniqueness
>          (statistical or absolute (default)), and their lifetime
>          (ephemeral or permanent (default)).

"do not change"?  Identifiers may and will change at any time, but the
point is that they NEED not change in response to a change of attachment
point.

So bringing these mods together,

  An identifier is the name of an object; identifiers have no opological
  sensitivity and need not change even if the object changes its
  point(s) of attachment within the network topology, and it may change
  at any time, independent of attachment point changes.  Identifiers may
  have other properties, such as the scope of their uniqueness (local or
  global (default)), the probability of their uniqueness (statistical or
  absolute (default)), and their lifetime (ephemeral or permanent
  (default)).


> address  An address is a name that is both an interface locator and
>          an endpoint identifier.

That depends on what "is" is.  A bit field is just a bit field, etc.  I
suggest:

  An address is a name that is used both as a locator and as an
  identifier.


Then on to some responses to comments leading up to these ...

Excerpts from Dino Farinacci on Sat, Mar 28, 2009 10:11:52AM -0700:
> A MAC address and a LISP EID are both identifiers which also have  
> topological significance within a scope. 

Forwarding can use anything it wants, and just because forwarding uses
something does not mean that thing is a locator.

Based on the definitions above I don't believe LISP EIDs are ever used
as identifiers (and that's the real test of whether one "is" an
identifier").  That is, a LISP EID is never used as a name of a packet
consumer/generator which never needs to change if its attachment point
changes.  Just because a locator is not usable for forwarding in some
scope does not mean it is used as an identifier.  In the core a LISP EID
is not used at all, it is payload.  During mapping, it is included in
the question "how do I get to this thing", and the function that is
asking that question is a forwarding function.

A MAC address does not NEED to change with topology changes.  And
again, just because it is used by forwarding doesn't make it a
locator.  Forwarding can use anything it wants, including SIP source
info.


Excerpts from Tony Li on Sat, Mar 28, 2009 11:47:56AM -0700:
> As I think we discussed, possibly after you left, a MAC address was
> originally conceived of as a pure identifier.  It's L2 locator
> semantics  were overloaded onto it when people conceived of
> bridging.  None of  these make the MAC address into an L3 locator.

Forwarding can use anything it wants.

> As far as I can tell, a LISP EID is simply a higher layer address in a  
> two level address hierarchy.  The RLOC the lower layer address.

I wouldn't even say hierarchy.  It's an address with just as much
global uniqueness as any other address.  The only difference is
where it is advertised by routing.  In fact it's just as routable as
any other address, it just isn't routED -- as demonstrated by the fact
that I could, theoretically, take any prefix currently routed on the
Internet and LISPify it.


Excerpts from Dino Farinacci on Sat, Mar 28, 2009 01:55:49PM -0700:
>> Dino Farinacci wrote:
> I realize that but after people mull over the RRG definitions, they
> will undoubly ask "so how does a MAC address fit into those
> definitions".
>
> So the RRG has the choice to say, it does in form X or a combination
> of form X and Y which are defined by the research group. Or, the RRG
> says we don't have a definition of it.
>
> So, to be practical, the RRG needs to decide if the definitions will
> be relevant to reality.

- An identifier does not need to change with attachment point changes

- An identifier MAY change at any time, independent of attachment
  point changes.  (and they do)


Excerpts from Tony Li on Sat, Mar 28, 2009 05:58:53PM -0700:
>>     > Identifiers may have other properties, such as the scope of their
>>     > uniqueness (global or local) and the probability of their uniqueness
>>     > (absolute or statistical).
>>
>> One 'word': "triple-barrel"! Actually, there's another important property,
>> which is the _lifetime_ of name. Presumably one property of an
>> 'identificator' is that it should be permanent?
>
> Ok, but we already know that some people, for the sake of privacy, want  
> to have temporary identifiers.

Also, some identifiers such as session identifiers are temporary by
nature.


Excerpts from Brian E Carpenter on Sun, Mar 29, 2009 03:14:09AM +1300:
> 1. An identifier can name various entities within the endpoint. It is
> important to qualify the identifier accordingly. In the RRG context,
> an identifier names the Sqglz entity within the endpoint.

Right.

> 2. The 'locatorness' and 'identifierness' of an address may vary
> between scopes and over time.

In my ideal world locatorness and identifierness are not properties of
the thing, but rather properties of the functions using them.  The
things themselves do not change (well, their values may change), just
how they are used.  An address is just a thing, while its locatorness
comes from being used as a locator, and its identifierness comes from
being used as an identifier.

> As for the exact definition of a Sqglz, I'm not sure we resolved
> that yesterday.

No but we're getting there.  Joel may write a draft, and I'll be glad
to help.


Excerpts from William Herrin on Sat, Mar 28, 2009 09:00:20PM -0400:
> On Sat, Mar 28, 2009 at 1:23 AM, Tony Li <[email protected]> wrote:
> > locator     A locator is a name that has topological sensitivity and
> >            must change if the point of attachment changes.
> 
> Hi Tony,
> 
> That soft-sells it a little. A locator specifies an entity's current
> attachment within the network topology.
> 
> Types of locators include:
> 
> * Local Locator - a set of names and/or numbers which specifies the
> point of attachment of a host, interface or service within the network
> topology of the local administrative zone.
> 
> * Remote Locator - a set of names and/or numbers which specifies the
> point of attachment of an administrative zone within the network
> topology of the Internet.

I would hold off on these.  As was mentioned in the face-to-face
meeting, every locator applies to a prefix, but in some cases the prefix
is a /32 or /128.  The locator definition as it stands now is general
enough to cover all of these cases.  If you want to define them now, why
call them "local" and "remote" when the real attributes are whether they
apply to a single endpoint or a group?

> The purpose of routing is, of course, to move packets from the
> initiator to a computer providing the service and then back to the
> initiator.

That's the purpose of forwarding.  Routing is to determine paths that
forwarding can use.

> > address     An address is a name that is both a locator and an
> >            identifier.
> 
> An address is a number used by the networking protocols to implement
> one or more types of locators and identifiers.

Not bad but what does it mean to "implement" a thing?  See what I
suggested above:

  An address is a name that is used both as a locator and as an
  identifier.

Scott
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg

Reply via email to