[e2e] e2e principle..where??....

John Day day at std.com
Wed Jun 6 10:51:49 PDT 2001

At 13:44 -0400 6/5/01, afalk at PanAmSat.com wrote:
>  > I think this is where I'm confused, apparently people have no
>  > problems with caches(or proxies when they are used a caches).
>Of course I can't speak for the community, but one answer that occurs to me
>is that the issue, again, is Internet transparency. Web caches operate at
>application level. While shortsightedness in designing invisible web caches
>might inhibit future development in http, the notion of even an invisible
>web cache shouldn't affect future developments in non-http network services.
>  >
>  > basically e2e is but ONE argument in system design, it is quite
>  > likely that other factors can outweigh e2e arguments...
>This makes me uncomfortable. When new technology is deployed into the
>Internet is affects the shared resource that we all use. Breaking with the
>design principles that Internet community has adopted should be done very

This should make you no more uncomfortable than some applications 
deciding that UDP provides sufficient e2e reliability or that SMTP 
feels they have enough reliability that an e2e protocol on top of 
mail relays is not necessary.  E2E applies when you need more error 
control than UDP and tells you where it should go.

>reluctantly. Let me relate a short anecdote about my life in the IETF. When
>I first approached some folks in the IETF about addressing TCP over
>satellite performance issues, I was (somewhat ignorantly) seeking to get a
>special satellite-friendly TCPng working group created. Some patient folks
>in end2end advised me against this by saying that 1) it wasn't necessary and
>2)perturbing very widely used end-to-end protocols, like TCP, is like

And why should an e2e protocol do something special for one segment 
of its entire path or answer the question that if it did how it would 
do something different to account for the different technologies in 
each segment along its path and if that wasn't bad enough how does it 
do it when it doesn't know what they are?

When the question you should have been asking is how do I construct 
the protocols for the media I am interested in to meet the service 
expected by TCP?

>fiddling with a biological system. The likelyhood of upsetting the stability
>of the system in an unforseen way -- risking network collapse -- is high
>enough to justify making changes only in extreme cases. A message I took
>away from this is that the stability of the network is more important than
>optimizations for a single user or community. This doesn't negate your
>observation that the e2e principle is one of many considerations but one
>needs to keep in mind that other people will have live the impact of your
>design decisions. Choose wisely. :)

Take care,

More information about the end2end-interest mailing list