.

Interview With Leonard Klienrock

This is our exclusive interview with one of the people that helped to create the Internet. Leonard Kleinrock had developed the packet switching theory, which is used to transfer packets of digital informaiton through regular telephone lines. We have asked Leonard some questions about the internet, his answers are very interesting, you can read about them here. Please visit his website at: http://www.lk.cs.ucla.edu

Question:  What was your role in the creation of the Internet?  How did you come up with the idea for your theory?

Answer:  I had anticipated the need for packet networks longbefore anyone really cared about or saw the need for them (a decadebefore the ARPANET which, as you know, later came to be known as the Internet). This early work critically influenced the approach to, andthinking about, the development of the ARPANET.  I developed thebasic ideas for networking from 1959 to 1962 in the course of my Ph.D.research at MIT; this work culminated in the publication of mydissertation in 1962 (which was also published as a McGraw-Hill book in1964, and later reprinted by Dover in 1972), with the first article onthese early ideas published by me in July 1961. Here is what I did in my dissertation.  It had a number of keyideas.  My work was directly motivated by a desire to study how datanetworks could be designed, and to uncover the underlying principles ofthose networks.  I did then, and have always, approached a problemwith the goal of exposing its underlying structure; what makes a systemwork well or poorly; what are the basic tradeoffs; are there principlesto be found that explain the behavior and results.  I found thatqueueing theory provided many of the essential tools for analyzing anddesigning these networks; the theory allows one to evaluate throughput,response time, buffering, loss, efficiency, etc, namely, many the systemlevel metrics that determine the performance of data networks. People sometimes get the story backwards and erroneously think thatqueueing theory led me to data networks; it was quite the other wayaround. In my work, I uncovered the fundamental principles of demand multiplexing(demand access) and resource sharing.  I developed an analyticalmodel, did the analysis, developed the optimum design methodology, andproved the correctness of my approach in that research.  Further, Ideveloped the first distributed control routing procedure.  I alsoanalyzed the effectiveness of "time-slicing" and chopping alarge amount of work into smaller pieces in anticipation of what latercame to be known as packet switching.   Note that packetizationby itself does not lead to the underlying technology that supported theARPANET (or the Internet).  It helps, and is part of today'snetworking technology, but by itself it is not the whole story of theefficiency of networks.  Let me comment on what made the ARPANET networking technology so powerful.  In my mind, there are threebasic components:         
A. The keyconcept of demand access (i.e., resource sharing).                 Packetswitching is one example         
B. The keyconcept of large shared systems. Highspeed channels is one example         
C. The keyconcept of distributed control. One example is that of distributed routing algorithms A. The key concept of demand access: The inefficiencies of circuit switching, the major networking technologyfor voice communications, could not be tolerated in data networks. So the idea of demand access (examples are: polling, message switching,packet switching, asynchronous TDMA, CSMA, etc) where you only use theresource when you need it (i.e., when you actually have data to send) wasfundamental for data networks. No one had previously elucidated theprinciples underlying the need for such structures.  Moreover, noone had produced a model, much less an analysis, of how they performedunder stochastic loads.  Lastly, there existed no optimal designprocedures for laying out the topology, choosing the channel speeds andselecting the routing procedure and routes.  I did all of the abovefor the case of demand access to network resources.  This allows thenet to make efficient use of resources in the face of highly burstytraffic, which is the nature of data traffic.  I showed that thenetwork response time can be improved with packetization, a concept thatwas implied in my work which showed how the response time was affected bythe length of the message unit; note that this focuses only on responsetime, and has little to do with efficiency (whereas demand accessresource sharing focuses on efficiency). I addressed this issue byshowing that round-robin time slicing (essentially breaking a file intosmaller messages, later to be called packets) "...results in shorterwaiting times for short messages and longer waiting times for longmessages...".  Note that the word packet was not coined untillater in the 1960's.  I must emphasize that the totality ofunderstanding the full picture, and not just the issue of packetization,had to be developed before a convincing body of knowledge could beamassed to prove the case for data networks. B. The key concept of large shared systems.          I studied the trading relations among delay, capacity and load.  In that study, I showed that the larger the channel capacity, at constant load,the smaller the mean delay.  In fact the relationship is linear: ktimes the capacity leads to a reduction in mean delay by a factor ofk.  Large systems are more efficient. C. The key concept of distributed control: I introduced the idea of distributed control in my dissertation anddiscussed why it was so useful.  I analyzed the behavior ofdifferent distributed routing algorithms (alternate routing, randomrouting, etc.), and simulated them as well.  I showed the benefitsand pitfalls of providing dynamic alternate paths; this I did by analysisand by simulation.  This concept of distributed network control thatrapidly adapts to network conditions was one of the basic designprinciples for the ARPANET.   I cannot emphasize how important it was to the development of packetswitching networks as we know them today to have elucidated theseunderlying principles of the technology.  In order to do this, Ifirst had to create a tractable analytic model so that behaviors could beevaluated.  For this purpose, I developed the exact equation formean transit delay.  Then it was possible to elucidate the basicprinciples.   As I said above, the three principles I developedwere (1) demand access, or resource sharing,  (2) large sharedsystems and (3) distributed control (I did not call these principles bythose names in my dissertation, but the basics were therenevertheless). In addition to my early work as a PhD graduate student at MIT, I alsocontributed to the "fathering" the birth of the Internet byvirtue of being the first node ever to be connected to the net. Indeed, on Sept 2, 1969, in my laboratory at UCLA, the first piece ofInternet equipment (the Interface Message Processor - IMP) was firstconnected to the first operational computer in the outside world; it wason that day that the Internet took its first breath of life bytransmitting bytes between the IMP and our Host computer.  Further,on Oct 29, 1969, I supervised the first message transmission between twoHost computers (UCLA to Stanford Research Institute).

Question: When you developed your theory, did you imaginethat in less then 40 years the idea you helped to develop would change our lives in so manyways?

Answer: Yes and no.  If you go to http://www.lk.cs.ucla.edu/LK/Bib/REPORT/press.html you will find a press release that came out of UCLA two months before the Internet was born.  There I am quoted (esp in the final paragraph) as saying that when the network expands, it will be possible to get access to a computer utility from our homes and offices as easily as we gain access to telephones and electric utilities, namely, just by plugging into the wall.  However, I never anticipated that my 92-year old mother would be on the Internet today.

Quesiton: How do you think our lives will look in 40 years from now? How will the Internet change them?

Answer: Forty years is far too long to predict.  However, I can see over the next 5 years.  My vision is that of nomadic computing where one will be able to travel from one place to another and still gain access to IP services as easily as they did from their usual corporate or home environment.  I also foresee the emergence of smart spaces where IP services will be available everywhere, all the time, and most of all, invisible.  Internet services will be available not only wherever you go, but accessible from almost any device you carry; in fact the environment (walls, desks, pens, eyeglasses, refrigerators, your body) will all contain embedded technology that will make the net ubiquitous.

Do you think that the enormous growth of the Internet in the last few years is a good thing or a bad thing? Considering the option that in a few years we wont even have to leave our houses to do anything, because anything could be done via the Internet.

Answer: Clearly a good thing.  Nothing makes me happier to see the penetration of the Internet worldwide, to all citizens, regardless of race, economic status, geography, political or religious persuasion, etc.  On the other hand, creating "couch potatoes" as you suggest is not a good outcome.  One must find a proper balance between cyberspace and physical space.  6. How often do you use the Internet? Of course every day, 6-8 hours per day at least.  I am connected most of the time and depend deeply on the Internet to carry out my professional activities.

How often do use the internet?

We would like to thank leonard for his time, please visit his website at:
http://www.lk.cs.ucla.edu

Copyright © 2000
Team C005753