[Catching up...]
This might be seen as good in the short term, but this kind of mixing of
fundamentally different functions in one mechanism is exactly the kind of
thing that, in the long run, can be a problem.
It is not mixing different functions. There is a big difference between
what forwarding uses and point of attachment names. Forwarding can use
anything, and does (not all of it in the packets). Granted it is
placing the service name in a field where one traditionally finds
addresses, but no actual functions are mixed. Forwarding does what it
always does. The service names cleverly look like addresses.
I don't much care how cleverly it's dressed up, it's still an
overloading of the token space. That's not an unthinkable direction,
but it should be done consciously, with care and forethought.
And regardless of much lipstick is applied, it still doesn't make the
'anycast group name' into a locator. It is simply an identifier for a
group, with 'one of this set' semantics. We also use it as a forwarding
selector.
Similarly, the token that we use for a 'multicast group name' is an
identifier for a group, with 'all of this set' semantics.
Perhaps one clear way of distinguishing these is to look at the embedded
topological semantics embedded in these independently of the routing
protocols. For a locator, there is a specific, assigned location in the
topology. For these group identifiers, _all_ of the location semantics
is necessarily embedded in the routing protocol.
If an architectural component is
fundamental, what someone does with it doesn't change what it is.
I like that, sort of, but it's missing something. After all, all tokens
are simply bit strings. It's only in doing something with them that we
assign semantics. Perhaps another way of saying this is that if you
overload semantics on an object, you don't obviate the original
semantics of the object.
So I ask you: if I had locators to start with, why wouldn't I still have
locators after someone else claims to have them as well? If the answer
is that I still do, how is that different from anycast?
A locator points to one assigned spot in the topology. If through
misconfiguration or maliciousness it is abused, then yes, it will be
mapped to the wrong place, but the fact that we can say 'right' and
'wrong' makes it clear that it has a semantic assignment to a given
point. Not so with anycast...
Tony
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
