Re: [RRG] Terminology redux

["William Herrin" <bill@herrin.us>]

On Tue, Jul 8, 2008 at 10:15 AM, RJ Atkinson <rja@extremenetworks.com> wrote:
> In at least some technologies that are being discussed on
> this list, the "Network-layer" Identifier does not have
> L2 Locator semantics.
>
> Whether one considers such an Identifier to have L2
> Identifier semantics depends mostly on how one defines
> the layers; just as it can be hard to distinguish
> between the "top of the network layer" and the "bottom
> of the transport layer".  Is ARP at the top of layer
> 2 or the bottom of layer 3 ?  What layer is MPLS in ?

Ran,

Rather than get sidetracked by terminology and arguments over the
definition of an identifier, lets return to the original point: in IP,
layer 3 and above are not tightly bound to the layer-2 address.
Because of this, multiple layer-2 addresses can be used during the
delivery of an IP packet and the layer-2 addresses in use can change
willy-nilly without impacting the function of the IP protocol.

Some examples of this in action are:

* Proxy ARP
* ARP-based failover
* Linux will by default send ARP responses for any of it's IP
addresses on any interface where it receives an arp request, even if
the address is not configured on that interface.

On the other hand, layer 4 is tightly bound to layer 3: a session is
identified the ports plus the layer 3 addresses. The session does not
survive either of the layer-3 addresses changing.

-IF- the layer 4 protocols were redesigned in such a manner where this
was not the case then the layer-3 address could change willy-nilly as
well. This has major implications for address-based route aggregation.
I claim they're major enough to solve the route scaling problem.


What's the counterexample? Is it impossible to design a protocol that
permits layer-4 on different hosts to find each other dynamically as
with ARP at the layer2/3 boundary? Is it impossible to design a
wide-scale aggregable addressing protocol? Given an optimal address
layout, would the topology of the Internet still require too many
routes? Would this sort of design induce the same sort of route-policy
violations that geographical aggregation does? Such a change would
break the hell out of many of the existing network security models.
Would the resulting protocols be unsecurable?

What's the counterexample? Why -won't- decoupling layer 4 from the
layer 3 address permit a -successful- solution to route scaling
through topological route aggregation?

Regards,
Bill Herrin



-- 
William D. Herrin ................ herrin@dirtside.com bill@herrin.us
3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004

--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg