Andrew Odlyzko Interview
Paul DiPerna:
Andrew.. I understand you worked three summers at the Jet Propulsion Laboratory in Pasadena, CA, very early in your professional life. What kind of work did you do there?
Andrew Odlyzko:
I attended Caltech for undergraduate work, where there was the opportunity to work at the Jet Propulsion Laboratory during the summer. The math group there was assembled from very talented people, some of whom did not function well in a traditional academic setting, but who often were brilliant at addressing specific problems. The students in the math group at the Jet Propulsion Laboratory were able to get first hand experience in techniques on the edge of mathematics and computer science. This experience was very useful to my Ph.D. thesis work in number theory. The Jet Propulsion Laboratory influenced my choices in graduate school and beyond.
Paul DiPerna:
You received your PhD in Mathematics from MIT in 1975. Can you describe what that graduate school experience was like for you?
Andrew Odlyzko:
At first, I was attracted to combinatorics and the group assembled there by the late Professor Gian-Carlo Rota. In a thoroughly charming incident, Rota offered the students a bounty of one dollar for each mistake they found in a manuscript that he was finishing rapidly to meet a deadline. Over a few days time I reported to Rota that I had found sixty mistakes. Rota paid me (a tidy sum for those days) and, more important, invited me to join him as co-author of the paper.
In time, I became more attracted to number theory and did a thesis on bounds for discriminants of number fields under the direction of Professor Harold Stark. I was able to adapt some of the techniques of linear programming, learned at the Jet Propulsion Laboratory, to achieve my results. I concentrated my research in number theory, and in 1986, I had the honor to be an invited speaker at the International Congress of Mathematicians at Berkeley.
Paul DiPerna:
Do you think demands are any different for graduate students today?
Andrew Odlyzko:
The main difference I see is that there has been huge progress over the last three decades, so that there is much more to mastered before launching on independent research.
Paul DiPerna:
Can you describe further about what you think needs to be mastered in graduate school before going off into independent research?
Andrew Odlyzko:
Obviously, a huge amount of technical material has to be mastered. But even more important, graduate school is where researchers get socialized, in the sense of learning how their fields function. Just what is the reward structure? How does one function in a challenging area? How does one cooperate with others?
get socialized"
Whether we like it or not, research is becoming increasing a team activity, with increased emphasis on interdisciplinary work. And, in a world where so much can be outsourced, it is the jobs with substantial human components (such as interactions with customers) that are least likely to be sent overseas.
Paul DiPerna:
You served for a long time at AT&T Bell Labs, from 1975 - 2001. What were some of your favorite projects over that time? What was the culture like?
Andrew Odlyzko:
Yes, I conducted research at Bell Labs in Murray Hill, New Jersey, and then at AT&T Labs, mostly in Florham Park, New Jersey, for quite a long term. Bell Labs was set up in the 1920s to do research that could benefit the communications industry. It was owned, in a complex relationship, by AT&T and Western Electric. When I went to Bell Labs in 1975, it had a mathematical sciences research center with 70-80 Ph.D.s. I joined the department in this center led by the famous mathematician Ron Graham. While I expected to stay there for two or three years, the appointment worked so well that I remained for twenty-six! I eventually headed one of the departments in the center. Under the pressure of emerging technology and government policy, Bell Labs and AT&T Labs went through a series of breakups, and I decided to go to the University of Minnesota in 2001
Paul DiPerna:
Can you describe your past work on the size and growth rate of the Internet? (see paper) What got you interested in that research question?
Andrew Odlyzko:
I have been collecting information about the size and growth rate of the Internet, primarily about traffic (as measured in bytes transmitted) and capacity, since late 1997.
The original motivation for this study came from my work on Internet pricing. In early 1997, I proposed Paris Metro Pricing. It is a very simple scheme that would at low cost, and without violating the principles of what is today called Net Neutrality, provide differentiated levels of services on the Internet. (This is QoS, the still-controversial and still-sought Holy Grail of the telecommunications industry as well as of many networking researchers.) But some comments I received from people running networks led me to question the basic assumptions that I had made, assumption that were universally held in the industry, and that I had absorbed from the literature and people around me, assumptions that held that data networks were chronically congested, and so on.
I quickly found that in order to get an understanding of where the Internet was going, and what technologies and services were likely to thrive, I needed to have a grasp on some of the basic properties of the physical Internet, such as its size and growth rate. Since nobody seemed to have any good data, I proceeded to collect this information myself, largely in collaboration with a colleague at AT&T Labs - Research, Kerry Coffman. And we did make many discoveries that were controversial at that time, but have been shown to be correct since. Internet traffic was doubling only about once a year, not every 100 days (and so the roughly $100 billion in investment in new long haul fiber networks was destined to vaporize, and duly did), data networks have light utilization (so almost all QoS technologies are irrelevant, and have duly been gathering dust), and so on.
I have been continuing these studies ever since, because they are key to understanding just what is happening. For example, in spite of all the hoopla about rapidly rising Internet traffic, growth rates appear to have slowed down to the 50-70 per cent per year range. At that level, they only barely compensate for improvements in technology, and so the industry should be striving to induce its customers to use more capacity, for example, and should give up on many of the systems they are developing.
Paul DiPerna:
Why did your recent co-authored article for IEEE, titled "Metcalfe's Law is Wrong", generate so much controversy?
Andrew Odlyzko:
Why so much controversy?
Most likely because we punctured this very comforting myth that had grown up over the last couple of decades. For techies as well as business people, having a simple quantitative measure, such as that provided by Metcalfe's Law, which promised rapid payoffs once a start-up reached a critical size, was a great motivator. It helped to justify much of what they saw happening on the Internet, especially the meteoric rise of companies like eBay and Google, and suggested that they could reasonably hope to achieve similar success. Having such a myth destroyed, and a much more conservative measure introduced, dampens the spirit of enterprise.
Paul DiPerna:
Was your article "Content is Not King" a building block of sorts that led to your refutation of Metcalfe's Law? To this day, that 2001 article continues to be one of the most widely read on First Monday
Andrew Odlyzko:
The observation that led to the "Content is Not King" article came at the end of 1999, when I was working on the manuscript "The history of communications and its implications for the Internet" (which was released for public distribution by AT&T in June 2000). As I was collecting a variety a statistics about communications technologies over the centuries, it suddenly dawned on me that the same pattern had recurred over and over again, namely that decision makers were pre-occupied with content (meaning material prepared by professionals for wide distribution), while what people cared the most about, and were willing to spend far more for, was connectivity, namely simple business or social exchanges. This observation became a prominent part of "The history of communications ..." and was then published separately in First Monday in February 2001 as "Content is Not King."
That connectivity is king, and not content, continues to be a contrarian view. Yet it predicted correctly that the over $100 billion that European telecom operators spent on 3G spectrum was going to be wasted, and that much of the investment in telecom technologies and services was and continues to be misdirected.
But it is hard to dislodge a deeply entrenched dogma!
"Content is Not King" did not contribute directly to the refutation of Metcalfe's Law. But it did reinforce a generally skeptical outlook on all "accepted wisdom." If something as deeply embedded in general thinking and decision making as the "content is king" thesis was false, what other assumptions might be wrong?
"Content is Not King" is very much in line with the Chris Anderson's "Long Tail" theory, as it says that the wide mass of small connectivity instances has more value than the few prominent pieces of content, where so much attention is devoted.
Paul DiPerna:
Back in March, your Digital Technology Center (DTC) hosted an open house to connect industry and academia to discuss the challenges and possibilities of "data mining".. Has there been progress in building partnerships?
Andrew Odlyzko:
The data mining open house was very successful, with close to 200 attendees. We are now building on the contacts we made there and on other occasions, and expect to announce some substantial data mining activities soon. There is tremendous interest in this topic, both at our university, and in industry, as everyone is facing a flood of data, and searching for methods to cope with it.
Paul DiPerna:
If we can provide here a trailer for your upcoming projects and research, what can we look forward to in the next couple years?
Andrew Odlyzko:
I have not published much recently, largely because I have been doing research that will lead to a series of books. The first one, which hopefully will get done in the next year, will be a comparison of the Internet bubble to the British Railway Mania of the 1840s, the greatest technology mania of all history. There are a number of amazingly close analogies, but also some interesting contrasts, which suggest how future technomanias might arise and develop.
For a sense of just how close some of the analogies are, please read the satirical short story, "The Glenmutchkin Railway," published in 1845 at the height of the Railway Mania. A copy is on my home page, and you can also find it elsewhere on the Web. The similarities to the dot-com promotions are striking.
November 23, 2006
