[e2e] on the need for more research in core networking

Joe Touch touch at ISI.EDU
Tue Jan 21 15:20:06 PST 2003


George Michaelson wrote:
> On Tue, 21 Jan 2003 08:01:00 -0800 Joe Touch <touch at ISI.EDU> wrote:
>  
> 
>>Although overlays began as a transition technology, it is becoming 
>>clearer (at least within my project team) that they are part of a more 
>>general extensions of the Internet architecture.
>>
>>Joe
> 
> I entirely agree. I think the slide from the network 'stack' to
> 'stackable networks' introduces a very nice abstraction, that you can soft
> reconfigure any (apparent) n-way end-to-end configuration on top of any
> other, for some cost in packet overhead, delay, etc.
> 
> Does it introduce much complexity in a router to hold two or more models of
> network routing? is this the kind of thing virtual routers inside one crate
> might address?

That depends primarily on whether the layers are ships-in-the-night or 
not. In the former case, it's just demuxing, though not many currently 
available software routing systems (gated/mrtd/zebra) handle it well.

> Actually, I think that might also point to the geographic
> addressing model as having some usefulness as well: if we explicitly expect to
> have to do layered networking, then we might as well make the underlying
> address topology follow the beaurocracy. Another unpopular proposal for IPv6
> dropped by the wayside...

FWIW, we've applied overlays for this purpose already too:

"Network Construction and Routing in Geographic Overlays,"
Greg Finn, Joe Touch, ISI Technical Report ISI-TR-2002-564, July 2002.
http://www/touch/pubs/isi-tr-2002-564/

I.e., overlays provide opportunity to use alternate naming, addressing, 
forwarding, and routing, though I'm always wary of applying them to 
'priviledged' routing that assumes there is no tragedy-of-the-commons as 
a result.

Joe





More information about the end2end-interest mailing list