[e2e] latest spate of cruft postings to e2e

Wu-chang Feng wuchang at cse.ogi.edu
Thu Nov 13 11:57:05 PST 2003


I would argue that DoS prevention and congestion control are not 
special-case functions, but rather mandatory functions that must be 
placed in the core of the network (i.e. IP).

The problem with these functions (vs. something like reliability), is 
that they require universal implementation to be meaningful.  If you 
relied on just end-point implementations, then an IP flood is all one 
would need to perform to disable the function for everyone and every 
application (regardless if they've implemented puzzles or not).  This is 
what we like to call the "weakest-link" argument.

$0.02
Wu


David P. Reed wrote:

> At 04:54 PM 11/12/2003, Wu-chang Feng wrote:
>
>> IMO, TCP is the wrong layer to put it at.  It makes sense to have the 
>> puzzle mechanism placed in a single, common layer (i.e. IP) rather 
>> than in every individual transport and application protocol.
>
>
> Well, this is a true inversion of the end-to-end argument...
>
> I offer the opposite point.  TCP *is* the wrong layer to put it at, 
> but for all the reasons that justify placing function at the edge of 
> the network, rather than *in* the network, this mechanism should not 
> be included in the core of the Internet.
>
> PS: the point of invoking the end-to-end argument is not 
> "religion".    Every time I mention it on this list I get flamed for 
> that (incorrectly).   The point is that it is a "shorthand" for an 
> argument that we could have in detail (and often do have in detail) on 
> such proposals as this one.   Just like a subroutine call or a macro, 
> the end-to-end argument reference means - read the d***ed paper, and 
> then explain to me why this case is any different than all the other 
> calls to push special-case functions into the network.






More information about the end2end-interest mailing list