[FoRK] This Man Wants To Control the Internet

J. Andrew Rogers <andrew at ceruleansystems.com> on Fri Nov 2 11:14:03 PDT 2007

On Nov 2, 2007, at 9:18 AM, Jeff Bone wrote:
> On Nov 1, 2007, at 6:49 PM, Aaron Burt wrote:
>>
>> He's applying Robust Control Theory to routing protocols, basically.
>> Great idea.  I can't imagine that many folks with significant  
>> exposure
>> to Control Theory also do work in CS or IT.
>
> Yeah, I still don't know that I buy that this guy's really had any  
> novel insight or has anything other than a sort of cheerleader- 
> overview.


As far as I can tell (and feel free to disabuse me of this notion),  
the only thing destabilizing the Internet as an optimized network  
with the current protocol designs is the assumption that marginal  
cost pricing of the network segments is a good proxy for the total  
transaction cost i.e. when you include the resources consumed behind  
the network.  When network was very expensive and applications  
primitive this was a nice simplifying assumption.  As network is  
rapidly becoming very cheap there are application spaces where the  
network cost is not remotely representative of the transaction cost  
and so you quickly end up with degraded global performance (Nash et  
al) because the protocols allow a significantly asymmetric imposition  
of cost.  It boggles my mind that in 2007 there are still people  
designing globally decentralized network protocols whose well- 
functioning is predicated on non-greedy behavior, as though that  
worked out well every other time it was tried.  On the other hand,  
BGP4 and the backbones have evolved strong strategies that stably  
optimize flow at the network level.  The protocols on top of that  
network are another matter.

This is an interesting problem.  How do you design a protocol to  
route a network when the transaction cost is not significantly  
embedded in the network itself?  In theory, the answer is relatively  
straightforward (all manner of micropayment schemes), but for a  
global network this is impractical on many different levels and not  
remotely as convenient as using network pricing as an optimizing  
proxy.  A narrow solution to this issue is ubiquitous advertising,  
which is a micropayment scheme of sorts but it is not universally  
applicable to this problem space.  At some point, this will need to  
be dealt with for practical purposes but any usefully general  
solution would undoubtedly cause a backlash.

Cheers,

J. Andrew Rogers





More information about the FoRK mailing list