History of Communications Edited by Mischa Schwartz
Total Page：16
File Type：pdf, Size：1020Kb
LYT-HISTORY-August-Kleinrock 7/20/10 10:28 AM Page 26 HISTORY OF COMMUNICATIONS EDITED BY MISCHA SCHWARTZ AN EARLY HISTORY OF THE INTERNET LEONARD KLEINROCK INTRODUCTION also offer a personal account of the stages of development in Internet histo- same events, as an autobiographical ele- ry. I present these threads and phases It is impossible to place the origins of ment in this story. In doing so, I aim to chronologically so we can revisit the the Internet in a single moment of time. further contextualize publications from history as it unfolded. One may find One could argue that its roots lie in the the period — my primary source materi- elaborations on this history in two earli- earliest communications technologies of als — with details from firsthand experi- er papers [4, 5] centuries and millennia past, or the ence. This perspective may add to our beginnings of mathematics and logic, or depth of historical understanding, in THE RESEARCH THREAD even with the emergence of language which the extent of personal detail does In January 19571 I began as a graduate itself. For each component of the mas- not imply a greater importance to the student in electrical engineering at Mas- sive infrastructure we call the Internet, events presented. In focusing on the sachusetts Institute of Technology there are technical (and social) precur- work of individual researchers and (MIT). It was there that I worked with sors that run through our present and developers, I rely on the various publi- Claude Shannon, who inspired me to our histories. We may seek to explain, cations that followed the work of these examine behavior as large numbers of or assume away, whatever range of individuals to link this story to the factu- elements (nodes, users, data) interact- component technologies we like. It is al historical record we will follow. There ed; this led me to introduce the concept equally possible to narrow Internet his- are, of course, many important personal of distributed systems control and to tory down to specific technologies with and institutional stories that have yet to include the study of “large” networks in which we are the most familiar. be told. The University of California at my subsequent thesis proposal. In that There are also many individuals that Los Angeles (UCLA) is heavily men- MIT environment I was surrounded by may be said to have “predicted” the tioned in this column, as it was the site many computers and realized that it Internet. In 1908, Nikola Tesla foresaw of so much foundational work. I view would soon be necessary for them to [1] a technology that would allow “a this period as a synergistic surge of tech- communicate with each other. Howev- business man in New York to dictate nology, engineered by a magnificent er, the existing circuit switching tech- instructions, and have them instantly group of researchers and developers nology of telephony was woefully appear in type at his office in London amidst a defining period of challenge, inadequate for supporting communica- or elsewhere” and would allow global creativity, invention, and impact. tion among these data sources. This access to “any picture, character, draw- was a fascinating and important chal- ing, or print.” Thirty years later, H. G. lenge, and one that was relatively unex- Wells articulated [2] his idea of a BEFORE THE BEGINNING: plored. So I decided to devote my Ph.D. “World Brain” as “a depot where WO HREADS THAT EET research to solving this problem, and to knowledge and ideas are received, sort- T T M develop the science and understanding ed, summarized, digested, clarified and The Internet did not suddenly appear of networks that could properly support compared.” These ideas were followed as the global infrastructure it is today, data communications. by a 1945 essay [3] by Vannevar Bush, and neither did it form automatically Circuit switching is problematic predicting a machine with collective out of earlier telecommunications. Dur- because data communications is bursty, memory that he called the memex, with ing the late 1950s and early 1960s, two that is, it is typically dominated by short which “Wholly new forms of encyclope- independent threads were being woven. bursts of activity with long periods of dias will appear, ready-made with a One was the research thread that even- inactivity. I realized that any static mesh of associative trails running tually led to the packet switching net- assignment of network resources, as is through them, ready to be dropped into works of today’s Internet. This thread the case with circuit switching, would be the memex and there amplified.” followed three possible paths to the extremely wasteful of those resources, These predictions, however, do not technologies that eventually emerged; whereas dynamic assignment (I refer to help us understand why the specific the researchers involved were, in this as “dynamic resource sharing” or events, innovations, people, and circum- chronological order, myself, Paul Baran, “demand access”) would be highly effi- stances that formed our Internet and Donald Davies. Below we explore cient. This was an essential observation, emerged when they did. Doing so is not these three paths, which were indepen- and in 1959 it launched my research possible from the scale of centuries or dently pursued in the quest to provide thread as I sought to design a new kind single individuals. This column’s focus is data networking theory, architecture, of network. Its architecture would use on the defining inventions and decisions and implementation. The second thread dynamic resource allocation to support that separate early technologies that was the creation and growth of the the bursty nature of data communica- were clearly not the Internet, from a Advanced Research Projects Agency tions, and eventually provide a structure wide range of recent inventions that may (ARPA), the institution that funded for today’s packet-switched networks. help characterize our Internet, but were and deployed these technologies — a also built within it. Thus, in this column process that, as we will see, was by no we trace both the early history of the sci- means automatic. These two threads 1 Later that year on October 4, I experienced a ence and infrastructure that emerged as merged in the mid-1960s, creating the widely shared feeling of surprise and embarrass- the ARPANET, and the trajectory of historical “break” that led to the ment when the Soviet Union launched Sputnik, development it set for the even broader ARPANET. Once these threads the first artificial Earth satellite. In response, construct that we now call the Internet. merged, the implementation and President Eisenhower created ARPA on Febru- As one of many individuals who par- deployment phase began, bringing in ary 7, 1958 to regain and maintain U.S. techno- ticipated in the Internet’s early history, I other key contributors and successive logical leadership. 26 0163-6804/10/$25.00 © 2010 IEEE IEEE Communications Magazine • August 2010 LYT-HISTORY-August-Kleinrock 7/20/10 10:28 AM Page 27 HISTORY OF COMMUNICATIONS This concept of resource sharing was As we see below, I was able to apply 1962, which turned out to be the first emerging at that time in a totally differ- Jackson’s result to represent the data paper [11] to introduce the concept of ent context: that of timesharing of com- networks of interest by making serious breaking messages into smaller fixed- puter power. Timesharing was based on modifications to his model. length pieces (subsequently named the same fundamental recognition that So the stage was set: There was a “packets,” as explained below). In it I users generate bursty demands, and thus need to understand and design general- provided a mathematically exact analysis expensive computer resources were wast- purpose data communication networks of the mean response time, and showed ed when a computer was dedicated to a that could handle bursty data traffic, the advantages to be gained by utilizing single user. To overcome this inefficien- there was an emerging approach based packet switching for this new network.3 cy, timesharing allocated the computer on resource sharing in timeshared sys- Note that the fixed length packets I to multiple users simultaneously, recog- tems, there was an existing special-pur- introduced did not match the randomly nizing that while one user was idle, oth- pose network that suggested it could be chosen lengths of the model, but fortu- ers would likely be busy. This was an done, and there was a body of queueing nately, the key performance measure I exquisite use of resource sharing. These theory that looked promising. solved for, the overall mean system ideas had roots in systems like SAGE [6] As a result, I prepared and submitted response time, did not require that and in the MIT Compatible Time-Shar- my MIT Ph.D. thesis proposal [9] in assumption, so the mathematical model ing System (CTSS [7]), developed in May 1961, entitled “Information Flow in properly reflected the behavior of fixed 1961 by Fernando Corbato (among the Large Communication Nets” in which I length packets as well. first timesharing systems to be imple- developed the first analysis of data net- I also developed optimal design pro- mented). The principles and advantages works. I chose a queueing theoretic cedures for determining the network of timesharing were key to my realiza- model based on Jackson’s model to char- capacity assignment, the topology, and tion that resource sharing of communi- acterize a data network as a network of the routing procedure. I introduced and cation links in networks could provide communication channels whose purpose evaluated distributed adaptive routing for efficient data communications, much was to move data messages from their control procedures, noting that net- like the resource sharing of processors in origin to their destination in a hop-by- work/routing control is best handled by timeshared systems was accomplishing.