Re: [RAM] Re: Ramblings about "locator"

[[e-mail] (Noel Chiappa)]

    > From: RJ Atkinson <[e-mail]>

    > my take is that the IEEE MAC address (either 802 or 1394) in an Ethernet
    > frame is an address (i.e. neither identifier, nor locator).

I was going to assume that what you mean by "identifier" here is 'a name that
simply identifies something, without saying anything else about the thing'
(which is what I thought you were saying in your previous draft taxomony
message). However, that's clearly not the definition you're using, because
under that definition, an IEEE MAC address *is* an identifier; it calls out
one specific interface, without saying anything else about it (e.g. where it
is).

(Technically, I know, it's not true that it saying nothing else, because it
call tell you who the manufacturer is; yet another semantic axis. But I think
we can ignore that axis for the moment.)

So I am forced to conclude that you must mean something else by "identifier";
I was thinking that perhaps you implicitly mean that an "identifier" names a
transport-level entity - but I don't think you mean that either.

This is an example of why I don't like the term "identifier". (To me, all
names "identify" *something*; the only question is *what* does it identify,
and what are the interesting properties of that name.)


And as for using "address" - that term is so hopelessly damaged at this point
(except for specific modifier instances like 'IPvN address') that I have no
idea what it means (although I like your earlier suggestion that it be defined
to be 'a name with the mixed semantics of an ''internetwork locator'' and a
''transport identifier'' ', which is what an 'IPvN address' is) - but if we
use that definition, I absolutely cannot see how an IEEE MAC address can be an
"address".


    > {The claim that "an IEEE MAC is an Identifier" would be more true if one
    > considered Ethernet-II rather than IEEE 802.3.  Ethernet-II said there
    > was one MAC address per node, while 802.3 has one MAC per interface.

The issue of whether the namespace allows more than one name to be bound to a
single object is yet another independent axis which one can use to describe a
namespace.

It is not one that I had considered important in deciding whether a particular
namespace as a "locator", "identifier", or whatever. Your comment indicates that
you do think it is important?


    >> Maybe the essential point is that a locator can at least in principle
    >> be mapped to topology and an identifier can't.

    > A locator has location semantics, but not identity semantics.

And what exactly are "identity semantics"?

Reading that term de novo, I would assume that it simply means 'a name that
identifies an object'; or, from the mathmetical roots, perhaps 'a namespace
where names have a one-one correspondence to objects'.

    > An identifier has identity semantics, but not location semantics.  An
    > address has both kinds of semantics.  So an MAC address in an Ethernet
    > frame is an address.

What do you mean by "location semantics"?

Do you mean 'a name that includes some information about where the thing is
is, encoded *directly* into the name'? I don't think you can, because an
Ethernet address doesn't have the property, but by your proposed definition it
has "location semantics".

Do you "specifies a location"?


Also, just because a name identifies a thing which is at a particular location
(e.g.  an interface), that does not mean that the name has 'location
semantics'. To me, there has to be *information* in the name (so that e.g. it
can pass the 'are these two close' test).

    Noel

_______________________________________________
RAM mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/ram