I think Scott and I are mostly on the same page. One lingering
distinction is whether something like a MAC address should be called a
"locator", or whether (perhaps independently) a MAC can be considered
to "name a location" even if we don't call it a "locator".
Intuitively, a MAC seems to me like it names a location on an
ethernet, even though it is a "flat" name. I think this is a similar
property to that of an IP address once you get to the "final subnet"
on the way to a destination. There the IP address is essentially a
flat name as well (i.e., all the topological information in the name
value has been consumed). This basically maps to the host part of the
IP address, but gets a little complex to define in general because the
length of the mask is variable and locally configured.
Returning to definitions, I am happy to say that these types of names
indeed name locations, but because they do not encode any topological
information in the name value, we not define them as "locators".
Pragmatically, this definition of "locator" should also allow for co-
existance with the opinion that such flat names do not name location,
only interface.
In other words, not all names of locations are also locators. More
generally, it seems we are converging on a common understanding - that
is good.
R,
Dow
On Mar 30, 2009, at 10:09 AM, Dino Farinacci wrote:
But you don't want to have irrelevant definitions.
Dino
On Mar 30, 2009, at 9:49 AM, Scott Brim wrote:
Excerpts from Dino Farinacci on Mon, Mar 30, 2009 09:10:26AM -0700:
I think it's both because of the point made in the last few
messages that the issue is usability. Noel pointed out that MACs
as administered are not _usable_ as locators because of their
structure.
If MACs are used to find a location of an end-system, then they are
locators.
When a MAC moves from one location to another, the layer-2 network
readjusts to forward packets *in a new direction*. If that is not a
locator, then what is. Just because the value of the 48-bit address
doesn't change doesn't mean it's not a locator.
As long as calling something a locator doesn't exclude it from being
an identifier, that's okay ... but what good does it do you? What
good does that definition of a locator do if it's not exclusive? The
important thing you just said is not about the thing (locator,
identifier) itself, but about what a forwading function things of it.
What your sentence is really about is the first half: MACs are used
to
find a location of an end system (i.e. determine a next hop). MACs
are also used to identify end systems regardless of where they are in
the topology.
We do find cases where IP addresses move away from their subnet and
attach to another cable which is assigned another subnet. In that
case,
the IP address value of the host doesn't change and with a more
specific
route, there is a new location for that IP address.
See other messages about that. Locators don't always, necessarily,
change when the attachment point changes ... but unlike with
identifiers, a function can't assume that locators will NOT change.
They are better off assuming they will change. On the other hand one
of the inherent properties of identifiers is that any changing they
do
is completely decoupled from any attachment point change. They don't
have to change ever due to topology changes, and they can change
whenever they want even if there are no topology changes.
See? That's an essential difference that makes the definitions
worthwhile having. Otherwise why bother having them.
Scott
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
