A flaw in the Internet architecture?

[David Farber <dave () farber net>]

________________________________________
From: Christian Huitema [huitema () windows microsoft com]
Sent: Thursday, July 10, 2008 12:49 PM
To: David Farber; Tony Lauck
Subject: RE: [IP] Re:   Richard Bennett Op-Ed in the San Francisco Chronicle

> From: Tony Lauck [tlauck () madriver com]
> Subject: Re: [IP] Richard Bennett Op-Ed in the San Francisco Chronicle
>
> "These complexes exploit a flaw in Internet architecture that enables
> them to seize more than their fair share of network bandwidth,
> effectively giving their owner a fast lane. A richly funded Web site,
> which delivers data faster than its competitors to the front porches of
> the Internet service providers, wants it delivered the rest of the way
> on an equal basis. This system, which Google calls broadband
> neutrality,
> actually preserves a more fundamental inequality."
>
> What flaw might that be in the Internet architecture?

The reference is typically to programs like BitTorrent, that use multiple TCP connections. The TCP congestion control 
algorithms ensure that each of these connections will get about the same share at the bottleneck. Many network 
configurations rely solely on end-to-end congestion control to organize the sharing of network resource. If that is the 
case, a program that uses many connections will receive a larger share of than one that uses few.

This is not exactly a new problem, or a new argument. Back in the days, when the Netscape browser came out, it was 
accused of "playing unfair" because it allowed more simultaneous HTTP/TCP connections than the Mosaic browser. There 
are also recurring arguments that voice and video transmissions do not react to congestion in the same way as TCP, and 
thus are attempting to get more than their fair share.

Some people will call that a "flaw in the Internet architecture", and that is questionable. However, there certainly is 
a flaw in the common practice. If you think about it, that practice delegates the management of network resource to 
programs running under the control of end users. There is a clear conflict of interest there, as end users, or program 
designers, are constantly tempted to "optimize" their own service... and grab more than their "fair" share.

The Internet Architecture, however, does not prevent network providers from implementing some forms of traffic 
management. There is a rich literature that discusses variations of "fair queuing". The idea is essentially to classify 
the packets in categories, and ensure that each category receives a fair share, for various definitions of categories 
and fairness. Providers could certainly adopt some of these technologies to organize the sharing of resource between 
individual subscribers. That would certainly not break the architecture of the Internet.

-- Christian Huitema





-------------------------------------------
Archives: https://www.listbox.com/member/archive/247/=now
RSS Feed: https://www.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com