SIGCOMM'99 Tutorial Session Notes
			      Boston, MA
			     31 August 99

			      Paul Baran
			    paul@baran.com

1. My work in this field was done in the 1960-4 Paleolithic or
   pre-ARPANET era, as a byproduct of learning how to build survivable
   networks. I realized that the reliability of such a network could
   be far greater than the reliability of the elements that comprise
   that network. The reason is that units connected in series must all
   work before the system can work, while redundant units connected in
   parallel must all fail before the system would fail.  This meant
   that it would be theoretically possible to build super reliable
   systems out of unreliable parts.

2. And, if so then an extremely powerful and inexpensive network could
   then become feasible.  The following page is from my 1962 RAND
   Paper, On Distributed Communications, P-2626 and summarizes the
   objectives that I had in mind. (Fig.1.)  

3. This is then followed by some illustrations taken from P-2626
   describing the evolution of the ideas towards a very large,
   all-packet based network.  The charts are:

4. Definition of centralized, decentralized and distributed
   networks. (Fig. 2.)

5. Definition of redundancy level. (Fig. 3.) 

6. Perfect switching in a distributed network.  (This showed that
   extremely tough networks could be built at only moderate levels of
   redundancy.) (Fig.4.) 

7. The concept of a "message block." (i.e. "packet".)  (Fig. 5.)

8. The concept of adaptive routing. (Fig. 6.)

9. The concept of decoupling the user's logical address from the
   physical address. (Fig. 7.)

10. This illustration dramatizes the concept of building a network
    composed of a mixture of totally different types of media and data
    rates. (Fig. 8)

11.  There were understandably many questions raised about the
     feasibility of the concepts.  In response I was forced to clutter
     the literature with a series of about a dozen more formal RAND
     Memoranda (also titled On Distributed Communications) with
     implementation details focusing on the most challenged parts of
     the concept.  

12. These Memoranda are on line at the RAND web site.
    http://www.rand.org/publications/RM/baran.list.html  

13. Since the purpose of this meeting is to review some old history, I
    have also included a few pages from Vol. V, History, Alternative
    Approaches, and Comparisons, RAND Memorandum RM-3097 to show the
    relation of this work to earlier work in this general field.

14. The following tables list alternative approaches considered in the
    50's and early 60's relevant to building more survivable
    networks. (Fig. 9 and Fig. 10.)

15. For example (Fig. 11) is a 1952 a machine designed by Claude
    Shannon to simulate a mouse learning a maze.  It provided me with
    an early existence proof that adaptive routing would be feasible.

16. One of the points that Vol. V makes is that there is a spectrum of
    alternative tradeoffs allowing systems to be built with some but
    not all of the characteristics of truly distributed networks.. 

17. To better appreciate what is new, and what isn't, let us look over
    some of the key concepts of the network's proposal.

18. The concept of digital regeneration used is based on the very
    early T-1 work at Bell Labs. (Fig. 12.)

19. The source of the concept of converting all modalities of analog
    signals into digital signals is unclear.  It might be found
    somewhere in the very early literature of pulse code modulation.
    (Fig. 13.)

20. A short, fixed length "message block" now better known as a
    "packet". (Fig. 14.)

21. The "virtual circuit" concept. (Fig. 15.)

22. Taking advantage of statistical multiplexing.(Fig. 16.)

23. The time transformer concept.  A single data rate throughout the
    system in not required and the illusion of a real time connection
    created if the data rate is high enough.  (Fig. 17.)

24. Error-free transmission concept.  (Fig. 18)

25. Universal cryptography.  Link by link and end to end.  (Fig. 19.)

26. Achieving high reliability by the use of inexpensive high data
    rate links. (Fig. 20.)

27. High data rate trunks allowing simple lower cost
    switches.(Fig. 21.)

28. Combination of switching nodes and multiplexing stations to create
    an overall user-to-user all digital network. (Fig. 22.).  Anyway,
    you get the idea.  

29. For the student trying to put this all into a present context I
    have listed three reference works by competent historians:

   A.  Arthur L. Norberg and Judy E. O'Neill, Transforming Computer
       Technology, Information Processing for the Pentagon, 1962-1986,
       The Johns Hopkins University Press 1996.

   B. George B. Dyson, Darwin among the Machines, The Evolution of
      Global Intelligence., Addison Wesley 1997

   C. Janet Abbate, Inventing the Internet, The MIT Press, 1999.