[e2e] New approach to diffserv...

Sean Doran smd at ab.use.net
Sun Jun 16 10:36:15 PDT 2002


| >         demonstrate the greater utility for the greater number of people
|
| The Internet (pre-middleboxes) already did that, relative to the 
| innovation-free telecom network that is still unable to do more than 3 kHz 
| analog phone calls (because of its "middlebox" oriented solution of 
| requiring mapping "in the network" rather than enabling smart instruments!)

No.  The great innovation of the Internet was in the statistical
multiplexing gain achieved via incremental improvements to the TCP
congestion avoidance algorithm.   This algorithm continues to be 
end-to-end even in the presence of middle-boxes of almost every variety,
since the only ways in which they can influence the distributed processing
between the sender and receiver is to drop/delay/introduce-error/over-ack,
and there are ways of mitigating against the latter two behaviours
within the communicating hosts themselves.

It is that TCP's congestion-avoidance algorithm *can* run over
stupid networks such that those stupid networks do not descend
into congestion collapse, that is a strength.  This does not mean 
that in all cases stupid networks are superior, since TCP's
congestion-avoidance algorithm runs quite nicely over "intelligent" 
networks too.

To put it in another way: end-to-end congestion avoidance distributes
the work of congestion management away from the network operator and
into the hosts.   This to-the-end-system distribution was not needed
in the classical POTS network because the end systems (i.e., phones)
would send a fixed amount of data whenever active.

| There's no moral superiority here, nor "imposition".  The imposition 
| happens every time some network manager presumes that he (not she) is doing 
| good by imposing a middlebox on all of his users.        

							   ^^^^^^^^^^
??? uh, who is the chair of IETF's midcom wg again?

Or are you saying that she is so highly evolved that
she presumes she is doing evil by imposing a middlebox
on all of her users?   I'm confused.

| The market would decide, if "bottleneck barons" didn't feel empowered to 
| decide for the good of their selves (er, users).

How would you propose to disempower them?

	Sean.




More information about the end2end-interest mailing list