[Tsvwg] Re: [e2e] e2e principle..where??....

[David P. Reed]

At 03:50 PM 6/2/01 -0700, Randy Bush wrote:
> > As long as the intermediaries preserve protocol semantics while changing
> > headers (adding cookies for instance) or perform actions (load balancing)
> > that are transparent to the end points, there should be no violation of 
> e-2-e
> > principles, correct?
>
>as long as we are careful when we paint your car pink, change the doors
>and seats, and convert the fuel injection to an su carb, there should be
>no violation of the warranty and you should not care or notice.
>
>end to end is pretty literal.



To be more specific, tinkering with the content does NOT preserve the 
semantics in most cases.  Let me be very specific.

HTTP is a protocol that is used for a variety of applications.  The vast 
majority of those applications involve a "browser" and a "server" of the 
sort you expect.  However, HTTP is used to support other types of apps 
(e.g. RPC calls between clients, such as SOAP & XML-RPC, and remote control 
of certain devices).   In these other apps, one *cannot assume* that the 
endpoint device understands all the potentially rich set of operations that 
HTTP understands.  So, for example, assuming that the remote endpoint that 
is normally a "browser" can always handle cookies is a bad assumption - 
using the cookie facility because it "is transparent to the semantics" in 
normal browsers may in fact cause a non-browser application to 
malfunction.  It's the same mistake as assuming that an intermediary on a 
telnet connection knows the character set that can be displayed on the 
terminal.

Yet this is exactly the kind of intervention that fraudulently-advertised 
transparent middleboxes are promoted for.


- David
--------------------------------------------
WWW Page: http://www.reed.com/dpr.html