[e2e] Open the floodgate

[David P. Reed]

At 07:56 AM 4/22/2004, =?gb2312?q?Jing=20Shen?= wrote:
>So whatever theoretical model used, mechanism for
>transmission protocol is  just methods people take in
>a big game. So, if someone could modify the
>transmission mechanism to make me a better place in
>competing game why shouldn't he take it?  So, fairness
>between new protocol and TCP is not a problem. Maybe
>"game theory + feedback control" is a better way with
>transmission protocol analysis.

I commend all on this list to study the end-to-end protocol called 
BitTorrent - with an open mind.   I think that it is a very interesting 
example of game theory + feedback control, and sad to say, it has captured 
no interest from the theory community and the measurement/simulation community.

It's a tremendously creative idea to solve a practical problem - and almost 
certainly has flaws that can be discovered by a "disciplined" approach to 
thinking about it.   By analogy, it might teach us about how congestion 
control in routers can be managed as a game - and perhaps how we can deal 
with bad behavior, just as some of the work by people trying to build 
algorithms that work under the "byzantine" assumption or the "exponentially 
spreading attack" assumptions are improving the network.

If we had Cannara's world, where the entire network was owned by a 
benevolent government monopoly (or one owned by a small number of 
corporations who are free to collude, but never conduct network wars to 
screw each other up), perhaps we could apply traditional control theory 
because errors would be small linear excursions, noise would be white and 
uncorrelated and  the network would be managed by a high-priesthood of (ah 
why pretend?) AT&T employees who are never incompetent, never wrong, and 
powered by lead acid cells in windowless buildings.