Re: architected shim layers [was Re: threats ID]

[Dave Crocker <dhc@dcrocker.net>]

Brian,

BEC>  a better model would have been a recursive
BEC> one, with no predefined number of layers. Since that's not the
BEC> model the industry chose to adopt, we can approximate it by
BEC> inserting shim layers when we need to. 


I believe there are two things that can make your point more palatable.
The first might sound like quibbling, but I believe it really does help us
think and talk about these types of enhancements.

1.  Sub-layers.  These are entirely comfortable in the OSI model, and
frankly we have always had them in the Internet model.
Notably the network layer is actually 3 sub-layers.  Namely,
Convergence, Network, Inter-network.  IP-over-* is the convergence
layer. We have not really had a "network" layer in IP, I think.  But the
possibility of adding this new module to the architecture is pretty natural.

2.  Enhancement vs. hack. I've increasingly come to believe that the
addition we are talking about is a legitimate architecture enhancement,
rather than a special-case hack.  The distinction should be between IP
that is in every node along the path, versus IP that is only (or almost
only) in the endpoints.  What is nice is that the multiaddressing work
is not the first example of this distinction.  Any practical
implementation of IP has always made that distinction.  In case folks
disagree, then we can fall back to IPSec.

So, the Internet Model can get:

     Transport
     Network
        Host
        Transit
        Media Convergence (IP-over-*)

d/
--
 Dave Crocker <dcrocker-at-brandenburg-dot-com>
 Brandenburg InternetWorking <www.brandenburg.com>
 Sunnyvale, CA  USA <tel:+1.408.246.8253>