[e2e] New approach to diffserv...

Scott Brim swb at employees.org
Mon Jun 17 09:31:54 PDT 2002


On Mon, Jun 17, 2002 11:33:08AM -0400, Jonathan M. Smith wrote:
> But aren't routers middleboxes? The contain both routing functionality and
> forwarding (the "IP" stuff). It seems to me, compared say to  the
> venerable source routing technology(!) that they are!

We thrashed with this when putting together rfc3234, and finally said
middleboxes are those which perform functions other than the ones you
would expect from a normal IP router.  Circular, eh?  My sense is that a
middlebox is called a middlebox because it does something unexpected
from the requests actually made by the endpoint, in ways which may
invoke the e2e argument.  

> > >3. Why is the network engineered in isolation from applications?

We try really hard to have architectural fundamentals which make this
possible.  The rise of middleboxes is indicative that the needs of
applications have changed and we haven't caught up yet in our
architectural principles.

> Sure. I was being extreme to provoke discussion. But I think of
> distributed applications with multicast semantics and how hard it has
> been to get multicast into place in the network, in spite of (what
> appear to be) good designs? How can you explain that? 

It's there, but we don't yet have deployed solutions for scaling the
accounting and other control functions.  Look at the group key
management problems, for example.

..Scott




More information about the end2end-interest mailing list