The following text is copyright 2004 by Network World, permission is hearby given for reproduction, as long as attribution is given and this notice is included.


The Internet is not a railroad, or is it?


By Scott Bradner


Andrew Odlyzko is at it again.  He keeps putting out tenaciously researched papers that make it very clear that one "common wisdom" or another is fundamentally flawed.  His latest target is the telephone industry assumptions about what types of technology to invest in.  While he is at it, he describes a significant conflict between the architectural assumptions that brought us the Internet of today and the ability for the Internet service providers to make a living.


Andrew's most recent paper, "Pricing and architecture of the Internet: Historical perspectives from telecommunications and transportation" (, was published in late December.  The paper combines a history of transportation pricing with an exploration of which lessons from that history might apply to the Internet.


Andrew points out that the Internet may be an anomaly in the transport business because the cost of Internet transport does not currently depend on what is being transported.  Many other transportation systems, including railroads and freight haulers, charge different rates for transporting different types of goods.  Currently Internet service providers pricing does not take into account what is in the packets transported over the network.  Surfing the web, downloading music or watching a NASA telecast of the Mars landing all cost the same.


The Internet's basic architectural design assumed that Internet service providers just carried packets.  The ISPs did not know what was in those packets and did not care.  But as Andrew points out in this paper, there are now a number of reasons to think about changing this architectural assumption.  These issues are more fully explored in a 2002 paper by MIT researchers Dave Clark, John Wroclawski and Karen R. Sollins along with USC researcher Bob Braden called "Tussle in Cyberspace: Defining Tomorrow’s Internet" (  The reasons that it may become important that ISPs begin to know more about what is going on over their networks include economic issues, for example, ISPs need to figure out a way to benefit from high-value traffic to improve their ability to be profitable, and regulatory, for example, law enforcement agencies feel they need to be able to wiretap Internet-based communications. Both Andrew and Clark et al see significant conflicts going forward over these issues.  These conflicts may result in an Internet far less open for innovation than the historical one.


One of the parts of Andrew's paper that has gotten the most attention, at least on slash dot (, is his observation that "essentially all major networking initiatives of the last decade, such as ATM, QoS, RSVP, multicasting, congestion pricing, active networks and 3G, have turned out to be duds." I think this may be a bit overstated, maybe for effect.  He is quite right that none of these technologies have turned out to be successful carrier-based customer services but many of them (not including congestion pricing, active networks or 3G) have turned out to be quite successful within enterprise networks. 


Andrew's papers ( tend to be clear, thought provoking and annoying.  Annoying in that they tend to poke holes in too many of my not well thought through assumptions.  Sometimes annoying is the right thing to be.


disclaimer: The words "annoying" and "Harvard" do seem to show up in the same sentence rather often but the above positive view of annoyance is my own.