[e2e] a means to an end

David P. Reed dpreed at reed.com
Mon Dec 1 10:23:07 PST 2008

some points about the non-generality of your definition, which is tied 
to a very specific systems architecture.  The need for such a "tie" has 
always been my main point.

I guess wireless communications systems are not part of your model.  
Many such systems don't have a "communications graph".  (if you stand on 
your head you can try to construct one if it is needed for some 
discussion that cannot take place without one in hand, but Maxwell's 
equations don't construct such a graph).

I don't know what an object name is:  content addressable memories hold 
information, but don't necessarily have "object names", nor is an 
"object name" the primary retrieval means.

Why "reliably translated"?   Xerox PUP used addresses that could be 
mistranslated (48 bit UIDs), but the PUP protocol system worked 
reliably, having been constructed in a manner that followed von 
Neumann's constructive proof that you could build reliable systems out 
of unreliable parts.

Just as there is no need for "reliability" at the base of the network, 
there is no need for "location" at the base.

Craig Partridge wrote:
> In message <49342367.7070503 at reed.com>, "David P. Reed" writes:
>> So define "well defined place in the network" as YOU define it.
> A stable object name that can reliably be translated to an address in a
> communications graph.
> Explaining a bit (since we seem to have an impedance mismatch):
>     * stable object name -- it is a name that has longevity -- it isn't
>       a temporary name created for 5 minutes on-line.
>     * reliably translated to an address -- the address of the object
>       can change but the object always exists -- you can address it (I'm being
>       careful here to permit intermittent connectivity ala DTN -- 
>       what I'm saying is you can address the object and send it a request,
>       that doesn't mean you can communicate with it in real-time).
>     * communications graph -- I'd like a better term (as graphs
>       model nodes and edges, which for wireless isn't the abstraction
>       I want) but the point is that the combination of RF/wires/fiber/nodes
>       that make up our communications networks are reasonably independent
>       of geography. (I can have a subnet spanning continents.)
> Note there's some clumsiness here -- I originally used "node" and then
> chose "object".  Neither word is quite right.  I hope the broad point is
> clear.
> Thanks!
> Craig

More information about the end2end-interest mailing list