[3B2-9] man2011030004.3d 27/7/011 16:26 Page 4

Why the Arpanet Was Built

Stephen J. Lukasik Georgia Institute of Technology

The who, what, when, and how of the Arpanet is usually told in heroic terms—Licklider’s vision, the fervor of his disciples, the dedication of computer scientists and engineers, the work of graduate students, and so forth. Told by one of the key actors in this salient part of US and Internet history, this article addresses why the Arpanet was built.

The who, what, when, and how of the Arpa- agencies, as well as to international research net is usually told in heroic terms. J.C.R. and operations communities. Licklider’s vision, the fervor of his disciples, Numerous people made substantial contri- the dedication of computer scientists and butions to this endeavor. At the very begin- engineers, the work of graduate students, ning, networks of people (US friends and and the attraction of the Arpanet to early par- foreign adversaries) provided assistance and ticipants carries with it a sense of inevitabil- pressure to succeed. In the development ity. But why the Arpanet was built is less mid-phase, the number of people contribu- frequently addressed. Writing from the view- ting their expertise, covering a number of point of the person who signed most of the technical fields, grew rapidly. And as the checks for Arpanet’s development, in this ar- bounds of both the research and user net- ticle, I detail the rationale for investing US works expanded beyond the US, the people Department of Defense resources for research participating in networking continued to and development of the first operational grow exponentially. packet-switched network. The goal was to ex- The individuals identified in Figure 1 and ploit new computer technologies to meet the in the references provide chronological tie- needs of military command and control points in a network of events and people, against nuclear threats, achieve survivable some tightly coupled, some quite loosely control of US nuclear forces, and improve coupled, and some completely accidental. military tactical and management decision Events are tied to the textual discussion making. Although not central to the decision with letters corresponding to those bracketed to pursue networking, it was recognized that in the text. these capabilities were common to non- defense needs. The Cold War The relationships among the various events The first step on the journey to networking recounted here can be better understood in computers was taken on 29 August 1949 terms of their organization into six sequences when the Soviet Union detonated its first nu- of events. ARPA’s existence and sole purpose clear weapon, a major event in the postwar was to respond to new national security con- confrontation between the US and the Soviet cerns requiring high-level visibility. In this Union.1,[A] Because aircraft were the only way case, it was the command and control of mil- to deliver a nuclear weapon to a distant target itary forces, especially those deriving from the at that point, air defense became an immedi- existence of nuclear weapons and deterring ate US concern. their use. The translation of these needs to The US took two actions. In 1948, the Air programs and capabilities were the joint re- Force had already studied the need for an sponsibility of the technical community air defense early-warning network of radar and the agency’s managers. This included stations and had an 85-station network of both network theory and engineering the limited capability underway. In December critical experiments central to making practi- 1949, the Air Force Science Advisory Board cal progress. The next steps were dependent convened an Air Defense System Engineering on the transfer of the R&D results to users, Committee (ADSEC), chaired by Massachu- both in the DoD and related US government setts Institute of Technology professor

4 IEEE Annals of the History of Computing Published by the IEEE Computer Society 1058-6180/11/$26.00 c 2011 IEEE [3B2-9] man2011030004.3d 27/7/011 16:26 Page 5

1950 1960 1970 1980

USSR/RAND/Baran

ARPA/C2/Licklider

MIT/UCLA/Kleinrock

ARPANET/Taylor/Roberts/Kahn

TRANSFER/Foster/Rechtin/Lukasik

UK/Davies/Kirstein

Figure 1. Chronological development of the Arpanet. The points highlight when individuals, organizations, and random chance intervened. Events are tied to the textual discussion with letters corresponding to those bracketed in the text.

George Valley, with the intention of recom- In early 1950, ADSEC recommended auto- mending a system that would overcome the mating this process. Radar data would be many deficiencies of the ad hoc radar net- digitized, transmitted over telephone lines, work hurriedly being assembled (tellingly and analyzed by the new digital computers called Lashup).2 The second was on 31 Janu- being developed. The MIT Whirlwind com- ary 1950 when President Truman decided to puter, funded by the Office of Naval Research, accelerate the development of fusion weap- was seen as pointing to the technology ons to deter Soviet use of their less powerful needed for such a system. Although the cal- fission weapons.3 culations involved in correlating and dis- MIT’s Radiation Laboratory had been a playing radar data were simple, slide-rule major center of radar development in WWII level, the mass of data and the need for and some of the participants in that work, real-time display for the command and con- in the now-renamed Research Laboratory trol of response forces dictated something for Electronics, naturally turned to the new more powerful. problem. Current air defense technology The urgency of US air defense increased was the same as in WWII. Data from radars when North Korea invaded South Korea on scanning air sectors was posted by people 25 June 1950 and when China intervened with grease pencils writing in reverse on on 26 November of that year, thus bringing transparent boards. From this display, tacti- the specter of nuclear war with the commu- cal air defense commanders determined nist world even closer. bomber routes, projected positions, avail- Discussions between the Air Force and able response time, and the targets to be MIT to establish an Air Defense Development attacked. Decisions included alerting anti- Center in the mode of the Radiation Labora- aircraft units, dispatching fighter aircraft tory proceeded. MIT convened a study group, and guiding them to intercept the incoming Project Charles, consisting of its faculty and a bomber streams, providing warning to civil- number of outside experts to vet the pro- ians to take shelter, and alerting civil defense posal. The result was to approve the idea for workers. an off-campus classified Project Lincoln,

July–September 2011 5 [3B2-9] man2011030004.3d 27/7/011 16:26 Page 6

Why the Arpanet Was Built

namedforitsproposedlocationnearLin- flexibility in management and contractual coln, Massachusetts. A significant demon- arrangements, enforcement of the highest stration occurred on 20 April 1951 when a level of excellence from its contractors, and ‘‘bomber’’ was intercepted in real time by a openness to technical ideas from all direc- ‘‘fighter’’ using actual radar data transmitted tions.4 It sought results that redefined how over telephone lines and analyzed by Whirl- one looked at the problems it tackled— wind. The result of this effort was the Semi- game-changing solutions to what today are Automatic Ground Environment (SAGE) air called ‘‘DARPA-hard’’ problems. Modest evo- defense system. lutionary improvements were left to others. Thelaunchofanearthsatellitebythe ARPA’s early years were characterized by Soviet Union on 4 October 1957 was another ‘‘Presidential Issues,’’ assignments coming major development in the Cold War con- to it directly as the result of White House frontation.[B] If the USSR could launch sat- concerns about its most critical technical ellites of that size, they could also deliver concerns. ARPA initiated programs in large missiles armed with nuclear warheads to boosters to close the missile gap highlighted the US. Besides initiating a space race to ac- in the 1960 Presidential election; reconnais- company the advancing arms race, the sance, communication, navigation, meteoro- event resulted in rethinking the US policy logical, and missile launch warning satellites; for science and technology and in substan- ballistic missile defense; a worldwide satellite tial increases in research support for univer- tracking system; establishing a scientific basis sities and the education of students in in seismology for negotiating and monitor- science and engineering. Business-as-usual ing nuclear test ban treaties; counterin- was put aside. The best minds in the coun- surgency and unconventional warfare; and try were enlisted in the equivalent of the creating an interdisciplinary science of mate- WWII Manhattan Project, and the national rials from the separate disciplines of physics, security establishment went into crisis chemistry, metallurgy, ceramics, and engi- mode. neering. Shorter-term efforts were in solid One US response, DoD Directive 5105.15 propellant chemistry, energy conversion, dated 7 February 1958, established the Presidential protection following the Ken- Advanced Research Projects Agency (ARPA, nedy assassination, and other matters that now DARPA.) The directive is marvelously drew on its contracting lexibility, speed, ex- general: ‘‘to be responsible for the direction pertise, and innovative approaches. or performance of such advanced projects in the field of research and development as The Command and Control the Secretary of Defense shall, from time to of Military Forces time designate by individual projects or ARPA’s history and character lies at the heart by that category.’’ Over the years, this has of networking. To appreciate why things are come to be interpreted as seeing that techno- as they are, a deeper look at their origin is logical surprise would never be repeated, a ra- needed. The history of ARPA for the period tionale and guiding principle the agency has 1958–1974 covers its development and has unwaveringly followed.[I] What preventing been used extensively to aid in reconstruct- technological surprise means, in practice, is ing the events I recount here.5 that the agency does the surprising, not the Following the agency’s establishment in other way around. 1958 as the presumed savior of dysfunctional Over the next 10 years, ARPA developed a US national space efforts in the DoD, it was modus operandi that still distinguishes it decided that the US national space effort from other parts of the federal government. should not be lodged in the DoD.[K] Instead, It is quite small: typically approximately in 1959 it was moved to a new nondefense 150 employees, with only modest internal agency, NASA, created around a greatly structure. Program managers, who typically expanded National Advisory Committee are recruited for only a few years, are masters on Aeronautics (NACA). By this time, the of their subjects, equal to the most expert White House was having second thoughts specialists with whom they work. The agency after ARPA’s first director, Roy Johnson, pub- does not have a large budget. When adjusted licly disagreed with it over the civil space for inflation, the current one is only a factor decision. Similarly, the scientific establish- of two larger than its first budgets. What ment that had initially promoted the distinguished the agency from other parts agency’s formation now decided it would of government was its freewheeling style, rather have its funding from other sources

6 IEEE Annals of the History of Computing [3B2-9] man2011030004.3d 27/7/011 16:26 Page 7

than the DoD.[H,J] ARPA space programs were emerged. The armed services’ command transferred either to NASA or returned to the and control functions were central, for they military services. provided the basis for what military forces The services were relieved. They had been should and should not do. OSD welcomed concerned from ARPA’s beginning that a the opportunity to gain deeper insight into strong R&D arm under the direct control of the issues with the intention of making mil- the Secretary of Defense through his immedi- itary capabilities more useful to the Presi- ate staff in the Office of the Secretary (OSD) dent and the Secretary of Defense. Unlike would threaten their weapon development the other marriages of convenience in the prerogative. The joint chiefs of staff (JCS), early days of ARPA, the command and con- reflecting this position, opposed the agency’s trol assignment took on a life of its own. formation.[H] Now the services returned to The original ARPA program description the attack with renewed vigor, arguing that called for the agency to ‘‘support research ARPA was never intended to be a permanent on the conceptual aspect of command and agency, but simply a now-completed transi- control and to provide a better understand- tional phase in the reorganization of US ing of organizational, informational, and space efforts in response to the Soviet satellite man-machine relationships.’’6 launch. The appointment of the second The event that turned this small assign- ARPA director, Army Brigadier General Austin ment into the major national capability Betts, was seen as a holding action until the that evolved from it was when Jack Ruina agency’s fate could be decided.[J] In that hired J.C.R. Licklider (Lick) in October 1962 uncertain period, ARPA took on ‘‘filler’’ to run the program.[N] A psychologist work- tasksthatreliedasmuchonitsstatutory ing on human factors in WWII, a participant funding flexibility as on its technical in the Project Charles study that laid out the capabilities. conceptual foundations for the SAGE net- The new Secretary of Defense, Robert worked air defense system, a professor of McNamara, and his new Director of Defense psychoacoustics at MIT, and a computer Research and Engineering (DDR&E) Harold researcher at Bolt Beranek and Newman Brown decided to retain ARPA as part of (BBN) before going to ARPA, Lick saw the OSD. Betts was replaced by Jack Ruina, earlier command and control problem as having an MIT professor, as the agency’s third direc- two parts: the machine processing of infor- tor in 1961.[L] Its project in Command and mation and the presenting of that informa- Control Research, the foundation for all its tion to humans in a form suitable for use in subsequent information technology research, making decisions. was one of the assignments of convenience Another incident played a role in formu- during the Betts period. An expensive backup lating the ARPA program. The Defense computer at the System Development Corpo- Science Board, noting the success with ration (SDC) for the SAGE air defense system, which the physical sciences were being ap- the AN/FSQ 32XD1A, was declared surplus in plied to national security problems, pushed 1963 as concern shifted to missile defense.[M] for some DoD attention to the behavioral The DDR&E was interested in war gaming as sciences beyond the narrow focus in the the DoD sought to understand the dynamics department on human factors. An appropri- of nuclear war to establish requirements for ation for work in the behavioral sciences, weapon systems development and procure- small in size but large in establishing legiti- ment. SDC found itself not only with a macy for such work, was received by ARPA spare computer, but also with a decreasing near the end of the Ruina directorship. role in air defense software development Licklider’s office provided a convenient fortheAirForce.Sotheproblemofthe place from which to administer it. Thus, large valuable surplus computer, its soft- Lick had both a technical focus on informa- ware,andthepeopletorunitwasturned tion technology and a charter that could over to ARPA to address OSD’s system cover the decision side of his integrated requirements needs. vision. With the merging of the Departments of The ARPA environment provided ideal War (the Army) and Navy, the formation of ground for making fundamental advances theAirForceasaseparateservice,andthe in command and control. Although previ- management of all three under a civilian Sec- ously the agency’s work was carefully retary of Defense by the National Security Act watched by the Secretary of Defense, the of 1947, tensions that persist to the present White House, and the President’s Science

July–September 2011 7 [3B2-9] man2011030004.3d 27/7/011 16:26 Page 8

Why the Arpanet Was Built

Advisory Committee, in this endeavor Lick Lick made several direct efforts at net- was left alone. In his words, working, although no one quite knew how to turn networking into an academic re- [TheDirector]seemedtoobusy,hewasjust search area, and certainly the military phrase relieved to get somebody to run the office... ‘‘command and control’’ was not part of the I talked with him periodically [and] he would explanation. One exploratory effort at UCLA make suggestions about directions of things, looked into networking separate computers but pretty much let me do what I wanted to do. to function as a virtual parallel processor. A Therewasabeliefintheheadsofanumber much larger effort was to capitalize on the of people—a small number—that people could presence of three large IBM mainframes on become very much more effective in their theUCLAcampus.TheWesternDataPro- thinking and decision making if they had the cessing Center (WDPC) in the Business support of a computer system, good displays, School served a number of business schools and so forth, good data bases, computation at your command. It was the kind of image in the western part of the US to help intro- that we were working toward the realization duce computation into their curricula. A sec- of ... It really wasn’t a command and control ond was in the Health Sciences School and research program [anymore]. It was an interac- was supported by the National Institutes of tive computing program. And my belief was, Health. The third served the rest of the and still is, you can’t really do command and UCLA community, including the ARPA- control outside the framework of such a supported work. The plan was for a large proj- thing ...of course, that wasn’t believed by peo- ect, with IBM 360/40s to serve as the network ple in the command and control field. front-ends for the host mainframes. The proj- ect, which was to include substantial IBM What Lick and his successor, Ivan Suther- contributions in personnel and hardware as land, did was initiate a number of activities well as ARPA support, was submitted for that ultimately served both purposes well. funding in February 1965. It collapsed when Lick expanded the command and control re- UCLA decided not to renew the basic search program at SDC into a high-quality WDPC contract with IBM. By this time, computer science program based around Sutherland had replaced Lick as director of Centers of Excellence. MIT was one such cen- IPT. A noteworthy feature of Sutherland’s ter, and one step in the direction of computer thinking was that the network was to be ro- support for user interaction with data was bust, on the ground that a DoD-like network time-sharing. A centerpiece was Project MAC— should be able to ‘‘survive a bombing raid.’’9 variously understood to be an acronym for Starting from this state of development, Machine-Aided Cognition, Man and Com- and fully implemented by his successors, puter, and Multiple-Access Computer. Lick initiated a major change in thinking A contemporary perspective on time- about networked computers, a change much sharing was that the speed of computers broader than simply connecting computers, was so great compared to the speed with a capability implicit in some defense systems which people think that they could be shared, and applications at the time. Existing com- apparently simultaneously, among several puter networks were functionally specific users. But users could only interact with data networks, special-purpose networks designed in their host computer and only with other for a particular application. They were in that people at their same institution who were sense much like the special-purpose com- connected to that computer. One could have puters of that period, or application-specific several terminals connected to different com- chips today, optimized to do certain things. puters, but the transfer of data from one com- What Lick conceived was a general-purpose puter to another was awkward. The need to network that could be used for as many aid cognition by facilitating the interaction decision-support purposes as its users had of people and data as broadly as possible— the imagination to conceive. To Lick, ‘‘net- what Lick called the Great Intergalactic work’’ had three meanings: the network Network—was being explored by a number ofhardware,wires,andsoftware;theman- of people in different places but was not machine network, where each performed yet fully articulated.7,8 This combination of functions for which each was best equipped motives, interactive cognition, and manage- to reach a larger goal; and the network of ment efficiency persisted through the Infor- people, with their skills, experiences, and in- mation Processing Techniques (IPT) program formation resources that could be focused on in different areas. larger cognitive goals requiring interaction to

8 IEEE Annals of the History of Computing [3B2-9] man2011030004.3d 27/7/011 16:26 Page 9

solve problems too large or complex for an individual. Not only did Lick see information tech- The technical solution to nology and behavioral and cognitive science issues as connected, but in his view research avoiding decapitation and military needs also had much in com- mon. He notes in his April 1963 memo, involved new ways of

I am hoping that there will be, in our individ- routing and switching ual [research] efforts, enough evident advantage in cooperative programming and operation to in decentralized lead us to solve the problems and, thus, to bring into being the technology that the mili- tary needs. ...many of the problems will be es- and distributed sentially as important, in the research context as in the military context. communication systems

It was more a matter of luck than manage- that could survive ment that the JCS and the armed services did not realize the potential implications of damage from an attack. ARPA’s work in command and control. If they had seen the ARPA effort as eventually changing the way military forces would op- an independent Behavioral Science Office. erate and communicate based on, to use The latter office did not initially focus on deci- the current jargon, net-centric warfare,they sion making. The head of the Behavioral would likely have objected to so fundamen- Science Office had been in ARPA’s counterin- tal an intrusion into the essence of their surgency field office in Thailand and moved missions. the program into anthropology and political Even before the first Soviet nuclear weapon science. Later, when the office did focus on de- test, the newly established Air Force saw the cision making, it was in decision theory. An need to address its strategic role in the projec- early application supported political negotia- tion of US power. In 1948, at the beginning tions for transfer of the Panama Canal from of the Cold War, Soviet aims in Europe US to Panamanian sovereignty. The new be- were becoming clear. The Air Force recog- havioral science effort was intended to extend nized the need to enlist the US scientific the ARPA research approach to other parts of and technical community. One part of this the social sciences, a theme that continued process was the establishment of the RAND in ARPA for a number of years, despite the Corporation.10 This new ‘‘think tank’’ looked controversy the DoD’s presence engendered. into issues such as bombing accuracy, air-to- air combat, and the operation of air defense The Arpanet from Concept centers. This work, involving both mathemat- to Realization ical analysis and analog simulation facilities, By 1959, U2 flights over the USSR established was done in the System Research Labora- that the Soviets were deploying interconti- tory.11 SRL eventually undertook software nental ballistic missiles (ICBMs). An alleged production for the SAGE system but had to missile gap was a prominent issue is the spin it off to a new organization, the SDC, 1960 Presidential campaign. The concern when it became too large an effort for RAND was that US air defenses were powerless to manage. RAND did, however, continue against the high speeds and consequently technology efforts such as a pen tablet and short decision times implied by ICBMs. handwriting recognition to facilitate man- ARPA was directed to examine technical machine interaction. options for a ballistic missile defense system When Lick left ARPA at the end of 1964, to avoid a situation prone to instability the integrated character of the original com- from operational errors and misreading in- mand and control program was lost and the tent by either side. The 1962 Cuban missile Behavioral Science Command and Control crisis was a case in point. Research office was divided in two. The infor- In addition, the DoD was concerned with mation technology side of command and how the national command authority could control stayed with the now-renamed IPT of- survive a first strike and still have sufficient con- fice, while the behavioral charter shifted to nectivity to order a retaliatory second strike.

July–September 2011 9 [3B2-9] man2011030004.3d 27/7/011 16:26 Page 10

Why the Arpanet Was Built

Were the Soviets to believe a decapitating at- have scaled, how easily it could have been tack was impossible, the logic went that they adapted to new technology, and how well it would be deterred from initiating it.12 The would have performed in a nuclear attack technical solution to avoiding decapitation will never be known. Baran’s work is impres- involved new ways of routing and switching sive for the scope of the design specifications. in decentralized and distributed communica- It was the first packet-switched network tion systems that could survive damage from ‘‘invented.’’ ARPA’s later effort was a second an attack. pass by DoD at packet switching. Its network RAND was at the forefront of these policy efforts met with greater acceptance, largely and technology studies. Paul Baran joined becausetheywerenottightlycoupledto the organization in 1959 from Hughes Air- the defense command and control com- craft and synthesized many ideas for reliable munity. This decoupling was reciprocal. priority communication into a proposal for a The bulk of the ARPA contractors involved distributed system based on what he called were unaware of, nor did they particularly ‘‘message blocks.’’[C] The theoretical work care, why the DoD was supporting their was largely completed by 1962 and published research.15 in 1964.13,[D,E] The fifth report of this series One of the networking efforts indepen- summarizes 11 routing concepts examined dent of the earlier DoD work was that of between 1952 and 1961. The ninth report Len Kleinrock, a graduate student in the on security, secrecy, and tamper-free con- MIT Electrical Engineering and Computer siderations addresses architectural issues of Science Department. In his May 1961 PhD network security—deficiencies in the Inter- thesis proposal, ‘‘Information Flow in Large net we are only now coming to regret. As a Communication Nets,’’16,[P] he examined high-water mark in thwarting malicious questions of scalability, performance evalua- users, Baran notes that his system ‘‘assumes tion, control, routing, and related issues in a potential infiltration by enemy agents hav- distributed networks. His thesis was com- ing access to portions of the system and the pleted in 1962 and was published in cryptographic keybases.’’ 1964,[Q] coincident with the publication of RAND made a formal recommendation to Baran’s 1960–1962 work at RAND.17 Like the Air Force in 1965 to deploy Baran’s sys- Baran, Kleinrock recognized that information tem. Based on an outside review chaired by would have to be transmitted as blocks of Mitre, the Air Force concurred.[F] At that bits. Baran and Kleinrock later assisted point, the organizational structure of DoD ARPA in synthesizing the current technical intruded. Under the Defense Reorganization thinking into the Arpanet design.18 Act of 1949, the Defense Communication The next step in realizing Lick’s vision Agency (DCA), not the Air Force, had the re- occurred when Bob Taylor joined the agency sponsibility of developing, deploying, and from NASA in 1965.[O] Lick’s university Cen- managing DoD communication systems. ters of Excellence were central to the develop- DCA at that time was unfamiliar with digital ment of his approach to command and technology. It was assisted by AT&T, a major control: time-sharing; flexible and conve- provider of communication services and nient interaction with computers through deeply committed to analog voice circuits text, graphics, pointing, and handwriting; and their switching. The Air Force decided and artificial intelligence. Taylor added an giving the Baran/RAND proposal to DCA for additional push into networking, and later implementation was a plan for disaster and IPT directors added language, speech, and the proposal was withdrawn.14,[G] image understanding to the mix that enabled The system Baran and his team proposed the power of computers to expand into tasks looked remarkably like the Arpanet eventu- where humans could be augmented or ally did, but it had many features still sorely replaced. lacking in the public Internet 40 years later, Another illuminating story of the ARPA such as the redundancy to withstand heavy management environment is Taylor’s ac- attacks yet fail gracefully as links were sev- count of securing ‘‘seed money’’ to study ered; high reliability; security of the informa- networking—some $1 million in 1965—on tion within it; and the ability to defeat the basis of a 20-minute discussion with the malicious users inside and outside the sys- then director of ARPA, Charles Herzfeld.[R] tem. Baran’s network was intended to con- I recall a later conversation with Taylor, nect the national command authority to US in December 1966, following another semi- nuclear forces. How well the system would nal event, securing the cooperation of his

10 IEEE Annals of the History of Computing [3B2-9] man2011030004.3d 27/7/011 16:26 Page 11

principal investigators to pursue Lick’s net- earlier work on digital networks but high- working ideas. lighting the importance of the survivability The IPT contractors, for the most part aca- of defense communication networks under demics, but including not-for-profit organi- severe stress. zations such as RAND, SRI, and SDC, would In a 1999 presentation, Roberts recalled meet periodically to hear the research under- that taken by the graduate students who were the office’s intellectual ‘‘shock troops’’ and to en- Packet switching was new and radical in the courage the sharing of ideas among them. 1960s. In order to plan to spend millions of Networking was new, few were working in dollars and stake my reputation, I needed to understand that it would work. Without the area, and the matter was seen by the Kleinrock’s work on Networks and Queuing IPT principal investigators either as engineer- Theory, I could never have taken such a radical ing, not computer science, or a diversion of step. All the communications community funds that might be better spent to expand argued that it couldn’t work. This [Kleinrock’s] their own, self-evidently more important, book was critical to my standing up to them areas of research. Taylor’s relief in getting and betting that it would work.21 them to embrace networking and adapt their computer science research to a net- In his review of the history of network- worked environment might be difficult to ing, Roberts notes, ‘‘the discussions and in- understand when the normal buyer-seller re- terest Licklider spawned had an important lationship can be characterized as ‘‘He who motivating impact on the initiators of the pays the piper calls the tune.’’ But ARPA oper- first two actual network projects, Donald ated on the basis of partnership with, and Davies and myself.’’22 Donald Davies headed consensus among, its world-class contractors. the UK Data Communication Network pro- It was not about to pull rank and tell national gram at the National Physical Laboratory experts they were wrong, a position that (NPL).23 Commenting on the relationship would be inconsistent with why they were betweentheUSandtheUKwork,Roberts part of the ARPA program. noted, Taylor had the ball rolling and at least a grudging team, but he needed someone It was in good part due to this [Licklider] in- to translate the dream into reality. For this, fluence that I decided, in November 1964, he recruited Larry Roberts. Roberts was at that computer networks were an important Lincoln Laboratory where, in June 1966, he problem for which a new communications system was required. Evidently Donald Davies was working on an implementation of net- of the National Physical Laboratory (NPL) in working, a packet connection between the the United Kingdom had been seized by the TX–2 computer at Lincoln Laboratory, and same conviction, partially as a result of a sem- a Q32/PDP–1 computer at SDC in Santa Mon- inar he sponsored in autumn 1965 which ica. An account of the work (a message proto- I attended with many M.I.T. Project MAC col for the link) was presented at the Fall people. American Federation of Information Process- ing Societies (AFIPS) Conference in October The next need for Arpanet funding came 1966.19,[S] Roberts was reluctant to leave to in May 1968 for a contract to design and run the ARPA program, but did so when Taylor build 20 interface message processors enlisted Herzfeld to do some arm-twisting at (IMPs). The IMPs sat between time-shared Lincoln Lab. The story attests to the extra- host computers to which a user was con- ordinary effort ARPA spent on recruiting the nected and the public telecommunications best people for its programs. network. In fairness to the voice-circuit- The period 1965–1968 was seminal for the oriented communication industry at the synthesis in Roberts’ mind of then-current time, packet switching did seem to require ideas about distributed networks. His contri- substantial overhead in software, computing bution, amounting to an independent ‘‘in- time, and the use of a number of different vention,’’20 was in combining his own work computers (managed by separate organiza- with that of others and converting it into tions), as well as a dismaying number of specific decisions to turn the ideas into a interfaces to navigate to transmit a single working prototype network. He consulted ASCII character. Roberts clearly understood with others working on the subject. Baran how rapidly microelectronics technology was one contributor, bringing not only his wasmoving,hisearliernetworkrationales team’s technical contributions from his notwithstanding, and that eventually the

July–September 2011 11 [3B2-9] man2011030004.3d 27/7/011 16:26 Page 12

Why the Arpanet Was Built

cost of computation, subject to economies of a network would have to contend with, scale, would become trivial compared to would slow down the program with little communication and its dependence on capi- likelihood that what would be learned tal investments. would be relevant to understanding real It was a high-risk gamble that fit ARPA’s traffic in real networks. The program was approach to difficult but important areas of approved but structured with an initial four- national security. It involved having faith node test network to examine routing algo- in creative people allowed to operate with a rithms and other potential show-stoppers minimum of bureaucratic micromanagement before committing to the larger roll-out and in the dedicated work of many others in requested.[T] This was what led to the first and out of government. Even this computer- Arpanet configuration, the UCLA–SRI–UCSB– enthusiast physicist must own up to raising Utah net. The four sites were selected not an eyebrow when hearing the communica- only to economize on leased lines but tion protocols involved in packet switching because each had an area of expertise to con- explained. tribute to this first phase of the implementa- By this time, Eberhardt Rechtin replaced tion. The first node was at UCLA, where Herzfeld in November 1967 as the agency’s Kleinrock was now on the faculty, to provide director, and I became ARPA’s deputy direc- a capability to measure network performance tor. In a presentation for the ARPA FY 69 bud- and to test his ideas on packet delay and net- get, Roberts brought Herzfeld’s foundling work congestion. The other three nodes simi- back to the front office. Despite the contin- larly made important contributions to the ued commitment of the agency to high-risk four-node test network. programs to prevent technological surprise, BBN was selected as the contractor to mod- by 1968 the environment for ARPA was ify Honeywell 516 minicomputers to serve as quite different. Although the need for surviv- IMPs.24 ThecontractwasletinDecem- able networks had not changed, political ber 1968 with the first delivery 1 September forces were tightening the screws on ARPA. 1969. The first IMP was delivered to UCLA The agency’s budget was shrinking, partly three days early, ruining yet another week- to help pay for the Vietnam War and partly end for Kleinrock’s graduate students.[U] By because opponents to the war in the Con- December 1969, all four of the test sites gress were using every opportunity to cut were operational.[V] Finding no reason to DoD programs. ‘‘Fraud, waste, and abuse’’ change our plans, the Arpanet implementa- was a frequently invoked litany of sins. Lack tion of the remaining 16 nodes proceeded of relevance to national security was another. seamlessly. Seven additional nodes became Senator William Proxmire periodically identi- operational between April and July of 1970. fied projects for a Golden Fleece Award, proj- I do not recall Roberts asking for approval ects with scientifically accurate but easy-to- to proceed, but he did not have to. The way ridicule titles. Senator Mike Mansfield was ARPA operated, he would have come back soon to demand ‘‘relevance statements’’ for for guidance only had he felt it was needed. every R&D line item in the DoD. Maps of the early development of the As ARPA’s deputy director, the mind- Arpanet tell a story of the demand side of numbing chore of overseeing the preparation networking. Beyond the growth in number of these statements fell to me. The enormous of nodes, they reveal something of the nature amount of time with word-by-word review of the demand for networking by user com- enhanced my normal interest in applying munities (see Table 1). ideas to help fix problems and the experience Eb Rechtin left ARPA in December 1969 to exerted a strong influenced on my subse- become principal deputy to the DDR&E, and quent actions as director. I became ARPA’s acting director and then di- So the question of the degree of risk in the rector, eventually leaving at the end of 1974. proposed Arpanet arose, both technical and In addition to the several histories cited, political. Too little risk and the task could Arpanet’s history as part of the broader IPT be left to others, but too much and perhaps program is recounted by Arthur Nordberg it was a poor investment of limited funds. and Judy O’Neill.25 The history of time- Roberts was asked if the concept could be sharing, graphics, and artificial intelligence tested less expensively at a single site with are also used to illustrate the way in which simulated IMPs and synthetic network traffic. theinnovativeatmosphereofARPAputin We agreed that such an approach, lacking the place many other parts of information tech- real-world accidents and user errors that nology as understood today.

12 IEEE Annals of the History of Computing [3B2-9] man2011030004.3d 27/7/011 16:26 Page 13

Three trends in the growth of networking Table 1. Growth of early Arpanet nodes by type 26 can be seen. The first is a build-out at a lin- of sponsoring organization. ear rate to meet the needs of ARPA/IPT net- working R&D, supported for the first two IPT ARPA R&D Military Non-DoD years by the initial IMP procurement and Date R&D (Non-IPT) R&D R&D Total the provision of network service by ARPA as January 1970 4 0 0 0 4 a free good. This was largely completed by June 1970 9 0 0 0 9 September 1973. The number of ARPA R&D December 1970 13 0 0 0 13 nodes and contractors then began to satu- September 1971 15 0 1 1 17 rate. The next trend in the Arpanet’s growth March 1972 17 0 6 2 25 was driven by military R&D sites and a few August 1972 17 2 7 3 28 closely related non-DoD sites; the NASA/ September 1973 20 3 11 9 41 Ames node, for example, followed from the June 1974 21 3 14 9 45 decision to locate ARPA’s Illiac IV parallel July 1975 24 3 18 10 55 processor there. As early as March 1972, the July 1976 24 3 21 10 58 number of new nodes added for internal July 1977 24 3 24 10 61 ARPA purposes was exceeded by the number of military sites exploring the utility of net- working technology for their constituencies and missions. The third trend is the satura- efforts, IPT’s next step was to break the tether tion in overall growth of nodes, although to fixed locations defined by the telephone users and traffic continued to increase. system. Of obvious importance to military A 1974 Mitre study of Arpanet usage forces is mobility. This reasoning led to the showed that about three-quarters of the traffic ARPA packet radio net (PRNET). Like so was email.[e] This was somewhat different from much in the IPT networking program, there Roberts’ expectations on how the net would were the same set of mixed motivations. Any- increase the efficiency with which computer one spending DoD money would see the scientists would work. At a time when every need for extending network connectivity calculation had to be programmed for a spe- to mobile users. The researchers’ goals— cific computer, the early belief was that users deriving from the Licklider vision of network- would send their input data to the site best ing as encompassing everyone, at all places able to handle it. In the 1966 Thomas Marill and all times, with their information, and and Lawrence Roberts paper,19 the authors with the widest possible set of input and out- argued that users with large matrices to invert put options—were completely congruent would send the problem to sites that already with military needs. had matrix inverters. Those having a problem Not only did the PRNET have to function requiring natural language understanding independently of the Arpanet, but to meet would send their calculation to a machine al- the intent of networking, it had to connect ready programmed to do that. But in any to the Arpanet so users in both domains event, the ensemble of researchers utilizing could communicate and so the contents of the Arpanet, supported by the growing utility one network were available to users of the of the network, became a self-organizing col- other. Hence, the PRNET introduced the lection of users that evolved in different need for internetwork protocols. directions from what was anticipated. These The technical aspects of the PRNET were other uses, for example, included the packet detailed by Robert Kahn and his colleagues voice work that eventually became what is in 1978.27 The experimental PRNET employed now voice over IP (VoIP). spread spectrum transmissions, common for OneexpectsR&Dprogramstohaveun- military applications requiring covertness, expected results; that is the nature of any ex- but not authorized for civilian use at the ploratory enterprise, especially programs in time. The spectrum allocation for the experi- an agency that explained itself in terms of ment was approved by the Interdepartmental high-risk and high-payoff goals. The result Radio Advisory Council (IRAC), the nation’s was the Licklider vision of cooperative prob- frequency coordination body. But when it lem solving, not management efficiency in came to doing a packet-radio field experi- the utilization of hardware and software, be- ment in the San Francisco area in 1974, the came what the Arpanet was about. Federal Aviation Administration raised objec- Having developed and demonstrated the tions for reasons of flight safety at area air- utility of networking for facilitating group ports. The minor flap, remedied by a new

July–September 2011 13 [3B2-9] man2011030004.3d 27/7/011 16:26 Page 14

Why the Arpanet Was Built

spectrum allocation, was one of misunder- The Arpanet was providing email service standing, or perhaps inattention, but is illus- to its contractors and supporting a network- trative of the problems encountered when ing research community. The Mitre study innovation meets the real world of hardware, documenting that did not sit well with OSD regulations, and organizational missions. auditors. It meant ARPA had created an oper- ational communication system under their Breakout from the ARPA ‘‘Beachhead’’ very noses, without following the rules dic- In the late 1960s, John Foster, the DDR&E to tating the DCA as having that mandate. whom the agency reported, felt ARPA had HadARPAsaidinitiallythenetwouldbea slipped into a role of being ‘‘DoD’s NSF,’’ mail system utilizing packet switching, it and after almost 10 years in existence, there should, to conform to DoD policy, have seemed to be little coming out. Upon Eb gone to DCA to get it built—notwithstanding Rechtin becoming director in 1967, address- the fact that no one at the time knew how to ing this concern became an ARPA manage- do that. The request would have engendered ment priority. At that time, the US was much study and debate culminating, if planning the deployment of the Sentinel approved at all, in a system that would likely missile defense system. The DDR&E decided have been late and over-budget, the same to augment the Army’s technical capability concerns that dissuaded Baran and the Air by transferring ARPA’s Defender program, Force in 1965 from utilizing DCA to con- including current funding and staff, to the struct their proposed message-block system. Army Ballistic Missile Defense Agency. In In the long run, the auditor’s recognition his first year in ARPA, Rechtin negotiated of ARPA’s violation of policy had a positive this transfer, amounting to approximately effect. It took ARPA, an R&D agency, out of 25 percent of the total agency’s budget and the business of managing an operational net- more than half its strategic technology pro- work far from its research objectives and grams and personnel.28 required creation of a management infra- Beyond addressing ballistic missile defense structure dealing with details that were and OSD’s perception of the agency’s value, distractions to the agency. It also forced the Defender program transfer underlined a DCAtopayattentiontopacketswitching. need for its staff to shift some of its attention DCA handled the Arpanet management task from the input stage, formulating new pro- well and was soon converting to the new grams to address new problems and threats packet-switched environment for its own and managing high-quality projects, to the programs.[Z] output stage, moving their results to those In 1973, I was prodding the non-IPT offi- who could use them. Good research was not ces in ARPA to incorporate the network idea enough. Successful research had to have more into their thinking. One approach was simply than potential utility. Its actual utility for to use the net for internal ARPA operations. national security had to be demonstrated to The flexibility of email for management traf- Executive Branch budget managers and Con- fic and creating records of transactions was gressional appropriators through ‘‘rubber on appealing, not simply to boost networking the road.’’ It was the beginning of a coupling but because it worked better than what we of the 6.1-funded ‘‘Research’’ programs in had: seven-ply carbon paper, copiers, phones, ARPA to the 6.2 ‘‘Exploratory Development’’ and snail mail.30 The normal competitive category.29 It also led to closer and more posi- instincts for the director’s attention, and tive relations with the military services that hence funding for their projects, made easy had previously seen ARPA as a thorn and a access to the front office through the net waste rather than an effective partner in na- attractive. tional security. This was more natural for the Another opportunity to use the Arpanet 6.2-funded offices in the agency, but was a for operational purposes arose from the Nu- less familiar perspective for the research offices. clear Test Detection office, which I directed Although networking was shifting from a when I joined ARPA in 1966. In 1963 the managed and fully subsidized part of IPT, US entered into a Limited Test Ban Treaty and was being used by other parts of govern- banning nuclear tests in the atmosphere, in ment, those organizations were still engaged space, and underwater.[W] But the inability in networking R&D but focused on their to distinguish between underground explo- own mission areas. There were no purely op- sions and earthquakes prevented an agree- erational users who cared only about the ment banning those tests pending further quality and reliability of the service. research on how to verify compliance,

14 IEEE Annals of the History of Computing [3B2-9] man2011030004.3d 27/7/011 16:26 Page 15

which is central to arms-control agreements. SAAC turned out to be more of a bureaucratic Because the nature of earth movements for problem than anticipated, not because of explosions and earthquakes are inherently technical issues but because it required inter- different, this research sought to substantiate acting with a number of new entities that their differences as recorded by instruments neither knew nor cared about packet switch- at teleseismic distances and to determine ing or arms control. under what circumstances the two could be The connection of Norway to the SAAC reliably separated.31 This ARPA program, provided an opportunity to add UK network Vela, was to provide data that US policy mak- researchers to the Arpanet. Davies’ indepen- ers could use in negotiations for a treaty ban- dent network research at the NPL net, con- ning all nuclear weapon tests. sisting of about 10 sites in the London area, Central to the Vela program were arrays of was approved in 1968, and went into opera- seismometers. Through beam forming, one tion in 1970.[f,g,h] It was first reported upon can improve the signal-to-noise ratio to sig- in 1972.35 nificantly improve detection in the presence When international borders are crossed, of seismic noise arising from natural and national common carrier regulations are sub- man-made sources. The first of these arrays ject to international agreements that are built by ARPA was the Large Aperture Seismic far more complex than those involved in Array (LASA) in Montana in 1965.[X] It dealing with a government-friendly monop- comprised345shortperiodand16three- oly carrier. New bureaucratic problems component long period instruments with surfaced, which Peter Kirstein reports on an aperture of 200 km.32 in detail.36 Although the technical issues The convenience of developing seismic proved to be tractable, the regulatory and po- array technology in the US had a downside litical roadblocks were more persistent, over- for studying teleseismic discrimination. come only after considerable delays. Because LASA was too close to the only well-defined the UK in the early 1970s was moving toward source of nuclear explosions available, those entry into the European Common Market, at the Nevada Test Site. Experience gained their political emphasis was on links to the in LASA’s design and operation was used European Informatics Network in preference in the installation of a second large array, to those with the US. Further impediments the Norwegian Seismic Array (NORSAR), were bureaucratic. To connect the UK net- placed at a teleseismic distance from both work to the Arpanet required the use of US weapon tests and those at the Soviet test a TIP furnished by ARPA. On arrival at site in Kazakhstan.33 It was developed in con- London’s Heathrow Airport, however, it was junction with the Royal Norwegian Council impounded by Customs and Excise for duty for Scientific and Industrial Research and in- to be paid and an additional £5,000 was lev- stalled by the Norwegian Defense Research ied for Value Added Tax, neither of which Establishment. NORSAR consisted of 132 had been anticipated. The UK Science Re- short-period instruments and 22 long period search Council (SRC) held that this connec- instruments. The array was approved in Janu- tion did not present a particularly fruitful ary 1967 and became operational in February opportunity and declined to provide the 1971.[a,b] LASA and NORSAR recorded signals funding needed. from nuclear events in the USSR, which Another complication arose when the allowed corroboration of data and the intro- Scandinavian Tanum satellite ground station duction of multiarray analysis to reduce the came online, which allowed the retrieval of detection threshold below a seismic magni- NORSAR data at 50 kilobits per second tude 4.2. (Kbps) for the first time. This upgrade of Data from the arrays were analyzed at the bandwidth obviated further need for the Seismic Array Analysis Center (SAAC) in Al- 9.6-Kbps cable line that passed through London exandria, Virginia, where a Terminal Inter- en route to Norway. A new London–Norway face Processor (TIP), essentially a standalone link was needed to connect to the now IMP that allowed terminals direct connection expanded Norway–Washington link, but the to the Arpanet, was installed by August tariff proposed by the carriers acting under 1972.34,[Y] The installation of a TIP at NORSAR the umbrella International Telegraph and occurred at the same time.[c] This center Telephone Consultative Commission (CCITT) could now perform multiarray analysis to was beyond any possible budget provision. compare signals from different azimuths. The French wanted the tariff to be priced Getting near-real-time NORSAR data to the at the number of multiples of 2.4 Kbps

July–September 2011 15 [3B2-9] man2011030004.3d 27/7/011 16:26 Page 16

Why the Arpanet Was Built

voice circuits to achieve the Arpanet speed in Aldermaston and Malvern, and the SRC of 50 Kbps. The British Post Office offered joined the enterprise. The diplomatic high- to provide the link free for a year. Later, an light of the Arpanet was when Queen Eliza- appeal to the British Post Office to provide beth opened the Malvern link to the US in a 50 Kbps link from their satellite station at 1976, thus marking, as Kirstein notes, the Goonhilly Downs was approved. They pro- first involvement of a head of state in a com- vided the use of the satellite link on a puter network. research basis and thus circumvented the Aside from the bureaucratic gymnastics French roadblock. arising from adding overseas sites to the Kirstein explains that because of the pol- Arpanet, the experience stimulated two addi- icy to favor European over US links, it was tional dimensions to ARPA’s networking necessary to shift the UK–Arpanet connec- R&D. The first was the addition of satellite tion from NPL to University College London links that enabled the creation of a global (UCL). This idea was contained in a proposal network. (Fiber optic cables were not in the Kirstein submitted to the SRC at the end of picture at this time.) The second was the 1971. It was turned down then as being too need, already seen in the concurrent packet- speculative and uncertain, despite successful radio network development noted earlier, to operation in the US for several years. The De- interconnect networks. The need to connect partment of Trade and Industry insisted on NORSAR provided the opportunity to gener- statements of interest from industry. After alize the network from a US to a global entity nine months of study, their conclusion was and was central to the Arpanet’s eventual that ‘‘one would gain more from a two-week commercial exploitation. Irwin Mark Jacobs visit to the US than from a physical link.’’ explained: Nevertheless, the lack of official UK enthu- siasm had several positive results. The UK The Atlantic Packet Satellite Experiment serves link had to be for research purposes (an as the development vehicle and experimental ARPA condition as well) and the net had to test-bed for the overall Packet Satellite Pro- be developed based on real traffic, thus open- gram. The experiment includes both the exper- ing up participation by a broad set of UK aca- imental satellite network, called SATNET, and demic researchers. Originally, UK access was the supporting development and measure- not completely open, requiring a password, ment activities and facilities. ... The basic philosophy of the experimental but this requirement was eventually dropped. program and the SATNET development efforts The first packet was transmitted from UCL to has been to achieve a flexible experimental en- the University of Southern California Infor- vironment which could serve as a measure- mation Sciences Institute (USC/ISI) in Marina ment vehicle for a spectrum of specific [i] del Rey on 25 July 1973. system realizations and applications. Further, The duty and VAT arguments were not development and measurement emphasis has settled until 1976. The UCL appeal was ini- been placed on those system aspects that are tially rejected, but more senior levels of the least understood and most in need of feasibil- 37 treasury ruled—after being threatened with ity demonstration. export of all equipment back to DoD and reimport under the US–UK Exchange of These developments marked the transi- Forces Agreement—that ‘‘the equipment tion of the academic computer scientists’ that you have imported, and any future Arpanet to a global facility encompassing op- equipment brought under the same agree- erational as well as R&D users, embraced by ment would be free of duty and VAT.’’ This the communications authorities of three had the beneficial result of preventing other countries, and with packet technology dem- UK government departments from taking onstrated as technically viable in a variety overtheprojectbecauseofthedutyand of transmission modes. The role of ARPA in VAT they would otherwise have to pay. The networking technology had entered its next UK Ministry of Defence, through this period, phase. was most helpful and supported the UCL re- Related to the scope and timeframe of this search on network protocols with the condi- memoir is the technology of wireless local tion that MoD unclassified research networks area networks (LANs). In 1979, I became the could be connected. The international log- chief scientist of the Federal Communica- jam was eventually broken and the British tions Commission. FCC Chairman Charlie Post Office, British Library, Department of Ferris brought me in, based on my ARPA ex- Trade and Industry, the MoD Laboratories perience, to speed up the FCC’s approval of

16 IEEE Annals of the History of Computing [3B2-9] man2011030004.3d 27/7/011 16:26 Page 17

new communication technology. The organi- ARPA contractor community worked because zational challenge was similar to that facing the partners were moderately homogeneous DoD following the launch of Sputnik in in their level of knowledge and their willing- 1957. Some of what I accomplished at the ness to cooperate with one another. Al- FCC in the next few years included commis- though the networking community had its sion approval of the first spectrum allocation differences (largely technical disagreements), for cellular telephony, authorization of direct the broader DoD community that might broadcast satellites, and technical rules for have opposed ARPA’s work in command the use of unlicensed spread spectrum tech- and control was buffered from the perform- nology. The story of unlicensed wireless policy ers. The mediator, ARPA, managed the inter- is a similar mix of the technology, bureau- face, and was trusted by both sides. cracy, and politics seen in networking.38 There is nothing comparable today for addressing Internet security. There is no Lessons from the Arpanet trusted mediator with technical expertise Development and the resources needed to fix global prob- In 1983, the Arpanet was divided into two lems. The Internet has grown so large that parts, a research network that continued to the number of interested parties is probably be called Arpanet, and Milnet serving mili- too numerous to find enough common tary users. In 1989, the Arpanet was closed ground. Even if by some miracle govern- down. ARPA’s intent had been to demon- ments and industries worldwide were to strate the utility of packet switching for mili- agree on a course of action to remove threats tary command and control, and in that, it to the Internet and the services dependent on succeeded brilliantly. As a consequence it, substate groups—such as criminals bene- unanticipated by its sponsor, the Arpanet fiting from the current insecurity—are un- and its successor, the Internet, further dem- likely to cooperate. Neither will heads of onstrated the general utility of networking state forgo the intelligence-collection oppor- for ‘‘command and control’’ far beyond the tunities it provides, nor will military services needs of the DoD, just as Lick anticipated. abandon the precision of cyberweapons that Absent the subsequent metamorphosis of can be directed against national infrastruc- the Arpanet into today’s Internet, the story tures and other military targets. of the ARPA experiment would be a footnote The adoption of packet switching by com- in a technical history of military command mon carriers and the infusion of digital tech- and control. But because of its evolution nology into so many aspects of life have been into a public-access service based on the commercially successful, but the current same technology and protocols, the Arpanet quality of network service reflects the process development process deserves comment. through which this occurred, just as the orig- The Arpanet, as is typical of the output of inal net reflected ARPA processes. In the US, any process, bears the imprint of its origin. the FCC, which might have provided such Central was its open nature based on ‘‘rough guidance to protect users, has been largely consensus and running code,’’ a process absent after authorizing the resale of trans- recently described by Steve Crocker.39 Open mission capacity by nonregulated carriers. source code, open platforms, wikis, social There were two factors influencing the networks, text messaging, and the like are FCC’s lack of Internet regulation. Primarily, typical of the academic culture that produced the commission made a considered decision the Arpanet. But that openness has not scaled to not regulate what it called ‘‘valued-added well to the global community. services,’’ preferring the deregulatory view A frequent mantra of those concerned that regulation is more apt to constrain inno- with correcting the problems introduced vation than assist it. Although the FCC recog- into the Internet from its Arpanet origin is nized the impending information revolution the notion of public-private partnerships. and did not want to impede it, like most peo- This professed need follows from the private ple, it did not have an adequate understanding ownership of most Internet facilities and ser- of the technology and its policy implications vices. It reflects the joint responsibilities of to enable it to act responsibly in terms of pro- both private owners and governments for tecting users.40 Thus, it is a question whether correcting its security flaws. Public-private we could have avoided the insecure network partnerships work in many cases, the Arpanet we now have.41 development being a successful case in point. This is not unlike the current problem with The partnership between the DoD and the the proliferation of nuclear weapon states.

July–September 2011 17 [3B2-9] man2011030004.3d 27/7/011 16:26 Page 18

Why the Arpanet Was Built

In 1939, nuclear weapons seemed like a 2. R. Buderi, The Invention that Changed the World, good idea. By 1949, they had become a Touchstone/Simon & Schuster, 1997. nightmare, but the nightmare had the Inter- 3. R.E. Powaski, March to Armageddon, Oxford net as a consequence. In the case of net- Univ. Press, 1989. working, the technical community and the 4. A program manager recalls, ‘‘When I first met DoD focused on what could be done at the two-star Air Force general who was to the time. We did not attempt to follow all become the co-sponsor of my National Soft- possible moves ahead and plan how we ware Works project, his first words were, ‘I’ve would prevent unexpected and undesirable always admired ARPA,’ and I expected him to consequences.Toattempttodosobefore say something about our technical prowess. proceeding would have paralyzed us to inac- He finished the sentence with ‘... you are able tion, assuring that nothing positive would to spend money so effectively.’’’ be accomplished. On balance, ARPA and its 5. The history is based on official documents and developers protected the nation from one interviews with ARPA technical and manage- kind of technological surprise, only to create ment personnel. See the Richard J. Barber another. Similar problems of technology Associates report, The Advanced Research confront us today, in dealing with energy Projects Agency, 1958–1974, supported by and climate change, economic development the Advanced Research Projects Agency of and poverty, and genetics and its potential the Department of Defense under contract for misuse. number MDA 903-74-0-0096. It is available One can speculate about what would have through the National Technical Information happened without an ARPA or the accidents Service (www.ntis.gov). that caused it to improve capabilities for the 6. These and other quotes and details are taken command and control of military forces. from the Barber and Associates history. When Quite possibly, we’d have something similar serving as the agency’s deputy director, I found to what we have today. Computers would traces of this early history in the agency’s files still need connecting, carriers would still and was surprised to find we were originally want to sell transmission capacity, and indus- directed to work on command and control, not try would still want to sell the services thereby nurturing computer science in universities to enabled and the hardware and software to prevent technological surprise, as I had been provide them. The resultant network tech- led to believe. Military command and control nology might be structured as proprietary was not in the forefront of the minds of the islands of computers and users, perhaps like ARPA researchers by this time nor was com- today’s separate cellular networks. Today we mand and control research a recognized driver might be looking to public-private partner- in ARPA’s computer and behavioral science ships to bring about a more interconnected programs. It was seen simply as one area of state of affairs. The early ARPA decision to potential application. keep protocol developments open and in 7. J.C.R. Licklider, ‘‘Memorandum for Members of the hands of developers acting for the user the Intergalactic Computer Network,’’ ARPA Be- community as a whole would have been havioral Science Command & Control Research, more problematic in the face of private-sector 23 Apr. 1963; www.kurzweilai/articles/art0366. competition. html?printable=1 posted December 11, 2001. 8. M.M. Waldrop, The Dream Machine: J.C.R. Acknowledgments Licklider and the Revolution That Made Comput- ing Personal, Viking, 2001. In his bibliography, The following colleagues have reviewed Waldrop notes his use of the oral histories of this manuscript at various stages and pro- the University of Minnesota Charles Babbage vided encouragement and further insights: Institute. Paul Baran, Larry Roberts, Peter Kirstein, 9. This account was provided by C. Al Irvine, the Charlie Kline, Len Kleinrock, Barry Wessler, leader of the UCLA effort and the author of Bob Balzer, Steve Crocker, Charlie Herzfeld, the three-center network proposal, to Steve Eric Willis, Mike Marcus, Ed Feigenbaum, Crocker. and Bob Kahn. Their help is gratefully 10. The first staff members were a distinguished acknowledged. group: Henry Kissinger, John Forbes Nash, Stephen Dole, Herman Kahn, Harold Brown, References and Notes and Donald Rumsfeld. 1. Bracketed letters refer to events indicated on 11. W.H. Ware, ‘‘RAND and the Information Revolu- the summary timeline in Figure 1. tion,’’ RAND, 2008.

18 IEEE Annals of the History of Computing [3B2-9] man2011030004.3d 27/7/011 16:26 Page 19

12. This argument depends on a presumption of ra- the creation is not the individual but the group. tionality of the attacker. Concerns over irrational In this regard, the precedent of the 2007 Nobel attackers introduce further complication into Peace Prize Committee to recognize as awardees the decision processes. more than 450 lead authors, 800 contributing 13. The RAND work was published in a series of authors, and 2,500 expert reviewers of the reports as the design grew in detail. In 1964, Intergovernmental Panel on Climate Change is the team published them as a single series titled relevant. Nevertheless, the penchant for seeking On Distributed Communications. These can be simple causes for complex effects is strong. found at www.rand.org/about/history/baran.list. 21. L.G. Roberts, ‘‘The First Theory of Packet Net- html. The documents in the series are P. Baran, works,’’ presentation at SIGCOMM 99; www.lk. I. Introduction to Distributed Communication Net- cs.ucla.edu/first_packet_network_theory.html. works, RM-3420-PR; S.P. Boehm and P. Baran, 22. L.G. Roberts, ‘‘The Evolution of Packet Switch- II. Digital Simulation of Hot-Potato Routing in a ing,’’ Proc IEEE, vol. 66, no. 11, Nov. 1978, Broadband Distributed Communication Network, pp. 1307–1313. The independence of the UK RM-3103-PR; J.W. Smith, III. Determination conception of packet switching from that in the of Path-Lengths in a Distributed Network, US is unclear from this, but that is immaterial in RM-3578-PR; P. Baran, IV. Priority, Precedence, view of the continuing technical exchanges and Overload, RM-3638-PR; P. Baran, V. History, between the UK and the ARPA group that pro- Alternative Approaches, and Comparisons, RM- vided important stimulation to both efforts. The 3097-PR; P. Baran, VI. Mini-Cost Microwave, RM- drive to connect computers together between 3762-PR; Paul Baran, VII. Tentative Engineering 1962 and 1964 is unsurprising, with Baran, Specification and Preliminary Design for a Hgh- Kleinrock, Licklider, Davies, and Roberts con- Data-Rate Distributed Network Switching Node, verging on the idea almost simultaneously. This RM-3763-PR; P. Baran, VIII. The Multiplexing Sta- was the period when digital computers shifted tion, RM-3764-PR; P. Baran, IX. Security, Secrecy, from being one-of-a-kind devices to a commod- and Tamper-Free Considerations, RM-3765-PR; ity product of increasing size and power. Con- P. Baran, X. Cost Estimate, RM-3766-PR; and necting them was an obvious next step. The P. Baran, XI. Summary Overview, RM-3767-PR. devil, as always, was in the details. See also Baran’s account in the Willis Ware 23. In view of the natural interest in priority, Len RAND publication. Kleinrock, in his historical review, ‘‘Creating a 14. P. Baran, private communication with S. Lukasik, Mathematical Theory of Computing Networks,’’ 2008. See also J. Abbate, Inventing the Internet, Operations Research, vol. 50, no. 1, 2002, MIT Press, 2000, chap. 1. recounts the following chronology regarding 15. Early workers in the Arpanet project differ in the NPL work. In D.W. Davies et al., ‘‘Proposal their view of what they were engaged in, with for a Digital Communication Network,’’ unpub- few recognizing the fundamental military pur- lished memo, June 1966, www.archive.org/ poses of the program. All have various recollec- details/NationalPhysicalLaboratoruProposal- tions of the scope of their expectations. An IPT ForADigitalCommunicationNetwork, he Principal Investigator, working outside the net- described his proposed design for what he working area, noted ruefully, ‘‘We were all non- termed a ‘‘packet switched network.’’ The first visionary.’’ open publication was in D.W. Davies et al., ‘‘A 16. ‘‘Proposal for a Ph.D. Thesis, Information Flow Digital Communication Network for Computers in Large Communication Nets,’’ 31 May 1961; Giving Rapid Response at Remote Terminals,’’ www.cs.ucla.edu/~lk/LK/Bib/REPORT/PhD. Proc. ACM Symp. Operating System Principles, 17. L. Kleinrock, Communication Nets: Stochastic ACM Press, 1967. This network, which became Message Flow and Delay, McGraw-Hill, 1964; operational in 1970, was described by R.A. reprinted by Dover Publications, 1972. Scantlebury, ‘‘A Model for the Local Area of a 18. P. Baran, ‘‘The Beginnings of Packet Switching: Data Communication Network Objectives and Some Underlying Concepts,’’ IEEE Comm. Maga- Hardware Organization,’’ Proc. 1st ACM Sym. zine, July 2002, pp. 42–48. Problems in Organization of Data Comm. Sys- 19. T. Marill and L.G. Roberts, ‘‘Toward a Coopera- tems, ACM Press, 1969. tive Network of Time-Shared Computers,’’ Nov. 24. Industry giants such as IBM and AT&T declined 1966; www.packet.cc/fies/toward-coop-net. to bid, testimony to the high-risk nature of the html. ARPA network technology. 20. ‘‘Invention’’ is used here with some trepidation, 25. A.L. Nordberg, J.E. O’Neill, and K.J. Freedman, given the plethora of those who would be Transforming Computer Technology: Information pleased to accept such a mantle. In the case of Processing in the Pentagon, 1962–1986, Johns any large and diverse enterprise, the essence of Hopkins Univ. Press, 2000.

July–September 2011 19 [3B2-9] man2011030004.3d 27/7/011 16:26 Page 20

Why the Arpanet Was Built

26. The maps used are not referenced since they underground nuclear tests. The Norwegian can be found on the Internet in profusion. array was thus a continuation of the ARPA Vela 27. R.E. Kahn et al., ‘‘Advances in Packet Radio research program. It thus undercut the deploy- Technology.’’ Proc. IEEE, vol. 66, no. 11, 1978, ment of what would have been an unproven pp. 1468–1496. treaty verification system. 28. To head off what was seen as a large new 34. Later renamed the Seismic Data Analysis Center weapons program and a destabilizing technol- (SDAC); both acronyms are used on Arpanet ogy in the offensive strategic nuclear balance, maps as appropriate. the US and the Soviet Union signed an Anti- 35. D.L.A. Barber et al., ‘‘Operating Experience with Ballistic Missile Defense Treaty in 1972 provid- the NPL Network,’’ Proc. ACM Symp. Computer ing for a limitation of defensive systems to a Networks, IRIA, 1972. missile field protecting the national command 36. P.T. Kirstein, ‘‘Early Experiences with the Arpa- authority and one protecting a single missile net and Internet in the United Kingdom,’’ IEEE field to serve as a survivable deterrent force. Annals of the History of Computing, vol. 21, This formalized the strategic doctrine of mutu- no. 1, 1999, www.cs.ucl.ac.uk/jon/arpa/ ally assured destruction—defense had become internet-history.html. See also V.G. Cerf and offense, and offense was defense. This formula- P.T. Kirstein, Proc. IEEE, vol. 66, no. 11, 1978, tion was coined by J. Newhouse and quoted by pp. 1386–1404, for technical details. A. Wohlstetter in ‘‘Bishops, Statesmen, and 37. I.M. Jacobs, R. Binder, and E.V. Hoverstein, Other Strategists on the Bombing of Inno- ‘‘General Purpose Packet Satellite Networks,’’ cents,’’ Commentary, vol. 75, no. 6, 1983, Proc. IEEE, vol. 66, no. 11, 1978, pp. 1448–1467. pp. 15–35. 38. See Info, vol. 12, no. 5, 2009, edited by 29. This move to transfer research is even more C.L. Jackson at www.emeraldinsight.com/ strongly reflected in the current DARPA pro- 1463–6697.htm, particularly the papers by M.J. gram. The agency now receives 6.3 Advanced Marcus ‘‘Wi-Fi and Bluetooth—The Path from Development funds, thus helping push its Carter and Reagan–Era Faith in Deregulation to results even further downstream. Widespread Products Impacting our World,’’ 30. My experience in using email to manage the and S.J. Lukasik, ‘‘Unleashing Innovation: agency is detailed in the Waldrop’s The Dream Making the FCC User-Friendly.’’ Machine and in K. Hafner and M. Lyon, Where 39. S.D. Crocker, ‘‘How the Internet Got its Rules,’’ Wizards Stay Up Late: The Origins of the Internet, New York Times, 7 Apr. 2009. Simon & Schuster, 1996. It was not smooth. 40. S.J. Lukasik, ‘‘Unleashing Innovation: Making Quirks computer scientists worked past or fixed the FCC User-Friendly,’’ Info, vol. 12, no. 5, themselves were a much bigger problem to a 2009, pp. 76–85. physicist for whom writing Fortran programs 41. The FCC has recently inserted itself into Internet was a high-water mark in computing expertise. regulation but not to address its security flaws; 31. The key word here is teleseismic, meaning at a instead, it has taken up the political issue of net large distance from the nuclear event. Although neutrality. nuclear tests in a small country can be moni- tored from immediately outside its borders, in Stephen J. Lukasik is a dis- large countries such the Soviet Union and the tinguished senior research US, where suspicion of the integrity of data fellow at the Center for Inter- from internal sites for monitoring adherence to national Strategy, Technology, a test ban could not be satisfied, treaty verifica- and Policy at the Georgia tion had to be based on data collected outside Institute of Technology. His the borders of both parties. recent work deals with vulner- 32. The LASA design, and much of the signal abilities of industrialized soci- processing research, was the responsibility of eties to terrorist attack. Lukasik received a PhD in the MIT Lincoln Laboratory under Paul E. Green, physics from the Massachusetts Institute of Tech- Jr. It is described in the Nuclear Test Detection nology. He is a member of the International Insti- issue of Proc. IEEE, vol. 65, no. 12, 1977. tute for Strategic Studies, American Physical 33. At the time, the arms control community was Society, and American Association for the Advance- urging the installation of a network of 12 LASA- ment of Science. Contact him at [email protected]. type arrays around the world. The ARPA posi- tion was that the detection threshold and dis- crimination capability of seismic arrays had not yet been established to the point where the US could safely enter into a treaty banning

20 IEEE Annals of the History of Computing