[e2e] Open the floodgate

Cannara cannara at attglobal.net
Thu Apr 22 12:31:45 PDT 2004


I think you missed the point David.

Alex

"David P. Reed" wrote:
> 
> At 03:06 AM 4/22/2004, Cannara wrote:
> >What control theory does satisfaction with this
> >state derive from?
> 
> It's odd that "control theory" might be implied to provide satisfaction
> with the state of network control.
> 
> Traditional control theory applied to networks requires that they have
> perfect communications reliability - that is, bits get through reliably and
> with bounded latency.
> 
> I don't know of any physical situation where that assumption can be made
> true, anymore than that other assumption of communications theory in the
> 1970's refereed network theory papers: that network traffic was a random
> process, usually Poisson sometimes synchronous, that *did not respond* to
> network behavior (not "could not respond" but "did not respond").
> 
> Your precious theory is helpful, but what your language tells me is that
> you believe that the network design process should be one of developing
> analytic theories, and then *forcing the usage to match the assumptions
> which lead to a tractable theory*.
> 
> That's not engineering, even though it pretends to be.    It may give you
> satisfaction, because it lets you "prove systems correct".
> 
> A recent set of experiments on automobile control for intelligent
> transportation systems demonstrated that an architecture that delivered
> perfect information to the control loops holding cars in line, by using the
> best known reliable stream protocols, performed far worse on a reliability
> scale than a network that delivered corrupted measurements (random bit
> errors in position measurements) without correcting them.
> 
> This was done with the best possible "control theory" at the control level.
> 
> I don't mean to disparage theory - theory is enormously helpful in reaching
> an understanding, but it is far from a silver bullet.
> 
> In the case of the Internet's control problem, the actual problem - how to
> control a system that is not driven by random packet arrivals and random
> errors - is intractable, and the traditional control theory approach of
> forcing the system to obey simplified assumptions, is practically impossible.




More information about the end2end-interest mailing list