Inventing The Internet, Abbate 1999

Download as pdf
Download as pdf
You are on page 1of 39
Inventing the Internet Janet Abate “The MIT Press Cambridge, Massachuseus London, England —. WHARTON REPROGRAPHICS an se 6 idan Wide Web are prominent examples of informally created applications that became popula, not at the result of some central agency's mar- keting plan, but dhrough the spontaneous decisions of thousands of, independent users. In reconstructing the history ofthe Internet, have been struck time land again by the unexpected twits and turns its development has taken, Often a welllad plan was abandoned after a short time and replaced by a new approach from an unexpected quarter. Rapid advances, such af the introduction of personal computers and the invention of localatea networks, continually threatened to make existing network technologies obsolete. In addition, responsiblity for operating the Internet changed hands several times over the course ofits first thirty yeas oF so. How; inthe face of allthis change and ‘uncertainty, did the system survive and even flourish? I believe that the key tothe Internet's success was a commitment to flexilty and versity, both in technical design and in organizational culture. No fone could predict the specific changes that would revolutionize the ‘computing and communications industries a the end ofthe twentieth ‘entity. Anetwork architectre designed to accommodate a varity of computing technologies, combined with an informal and inclusive management syle, gave the Internet system the ability to adapt to an ‘unpredicable environment. “The ler deity 8 a communion medium wa nother cent in the technology, was constructed through a series of social Choices The ingenuity of the systems builders and the practices of ts users have proved just ar crucial as computers and telephone circus in defining the structure and purpose of the Internet, That is what the te ofthis book, Zmsetng the eat is meant to evoke: not an isolated act of invention, but rather the idea thatthe meaning ofthe Tternet hal to be invented—and constantly reinvented—at the same time asthe technology ill. I hope that this perspective will prove tuseful to thore of ur, experts and users alike, who are even now ‘engaged in reinventing the Internet. aL White Heat and Cold War: The Origins and Meanings of Packet Switching (OF all the ARPANET's technical innovations, perhaps the most cele- brated was packet switching. Packet stitching wae an experimental, ‘ever controversial method for transmiting data across a network. Its ‘proponents claimed that it would increase the efcency reliably, and speed of data communications, butt was also quite complex to im ‘mers, and some communications experts argued thatthe technique would never work, Indeed, one reason the ARPANET became the focus of s0 much atention within the computer science community vas hat it represented the fis large-scale demonstration ofthe fea bility of packet switching The succesfl ute of packet switching in the ARPANET and in other early networks paved the way for the techaiqu's widespread adoption, and atthe end of the twentieth ‘century packet switching continued to be the dominant networking, pracice. It had moved from the margins to the enter, from experi ‘menal to “normal” technology? Many computer professional have seen packet switching a8 having ‘obvious technical advantages over alternative methods fr transiting data and they have tended to teat its widespread adoption 36 a natural result of these advantages. In fat, however, the succes of packet switching was not a sure thing, and for many years there was ‘no consensus on what its defining characteristics were, what advan- tages it offered, oF how it should be implemented-—in part because ‘computer sciemists evaluated it in ideological at well as technical terms Before packet switching could achieve legitimacy in the eyes of data communications practitioners, te proponents had to prove that itwould work by building demonstration networks. The wide disparity in the outcomes of these early experiments with packet switching demunsrats tha the concept could be realized in very diferent ways, and that far ffom being a straightorward mater of 2 superior 8 Chap ‘technology's winning out, the “success” of packet switching depended ‘ready on how i wat imterpreted, Packet switching was invented independently by ‘wo computer researchers working in very diffrent contexts: Paul Baran atthe Rand Corporation in the Unived States and Donald Davies atthe National Physical Laboratory in England. Baran was frst to explore the idea, around 1960; Davies came up with his own version of packet witching ‘few years later and subsequently learned of Baran’s prior work Davies was instrumental in passing onthe knowledge of packet switch- ing that he and Baran had developed to Lawrence Roberts, who wis in charge of creating the ARPANET. This chain of invention and sisseminaton has become & standard element of origin stories about the Imermet; indeed, itis easy wo get the impresion that packet ‘witching simply cook a detour through the United Kingdom before re-emerging, unchanged, inthe United States to full ts destiny at the underlying technology of the ARPANET? However, while Baran's and Dave's versions of packet switching had some basic technical similarities, heir conceptions of what defined packet switching and of what it was good for were very diferent. Much ofthis difference was due tothe strong political preures that were brought to bear on computing research inthe United Kingdom and inthe United States. Large computer projects in both countries were developed in 2 context of government finding and control, and national leaders saw computers a a strategic technology for achieving ‘important politcal goals But in the very diferent policy contexts of the United States andthe United Kingdom, packet switching took on different meanings for Baran, Davies, and Roberts. Packet switching ‘wat never adopted on the basis of purely technical rtria, but always ‘because it fi into a broader socio-technical understanding of how data reworks could and should be used, [Networking Dr Strangelove: The Cold War Roots of Packet Suitching fn the United Seater AAs the 1960s opened, relations between the United States and the ‘Union of Soviet Socialist Republics were disney chilly. The USSR hhad launched its Sputnik satelite in 1957, seting off alarm in the United States over 2 "science gap" and prompting a surge of govern- ‘ment investment in science and technology. A series of events Kept the We Hat ond Cod War 9 Cold War inthe public consciousness: an American U-2 spy plane was shot down over the USSR in 1960, the Bern Wall went up in 1961 and 1962 brought che Cuban Mise Crisis. The shadow of nuclear vrar loomed over popular culture. The novels On the Beach (Shute 1957) and Fai-Sofe (Burdick and Wheeler 1962)—both made into ‘movies inthe early 1960s—presented chiling accounts of nuclear war andis aftermath. And in 1964, movie theaters across the United States, presented a briliant Black comedy of Cold War paranoia Dr Sirange dove (Kubrick 1968). ‘Dr Strangelove, though humorous, highlighted the vulnerability of the United States’ communications channels to disruption by a Soviet tack, which might make them unavailable just when they were rneeced most, In the movie, a psychotic Air Force commander named Jack D. Ripper sets a nuclear holocaust in motion by invoking a ‘Mratgy of mutual assured destrucion called “Plan R.” This plan— whi allows Ripper to circumvent the president's authority to declare vwar—is specifically designed to compensate for a wartime failure in ‘command, control, and communications. In the movie, an Air Force ‘general exphins: Plan isan emergency war plan in which alowerechelon commander may order nucearreaaon aera nek tacky the normal hin of command has teen daruped. The idea war to dacourage the Rushes fom any hope tha they could knockout Washington =. a8 part of general sneak iMac and ecape realist bees of ak of proper command and enol Plan R allows Ripper to launch a “retaliatory” attack even though no firs rike has actully ocurred, In realty (asthe film's disclaimer sate), the US Air Force never had any such strategy Even before Dr ‘Strangle opened, the Air Force was exploring avery diferent solu tion :o the threat ofa frst strike: building a communications system that would be able to survive an attack and so that “proper command and control” could be maintained. As Edwards (1996, p. 133) has ‘documented, Cold War defense analysts saw robust communications networks at a necessity in any nuclear confrontation: “Flexible- ‘espense strategy required that political leaders continue to commu cate during an escalating nuclear exchange... Therefore, presere- ing central command and control—poliicl leadership, But also recornaisance, data, and communications links—achieved the highest riltry priori" The need for "survivable communications” was ge cally recognized bythe early 1960s. Amang those intent on filing it toe 10° Caper 1 veas a researcher atthe Air Force's premier “think tank,” the Rand Corporation Founded by the Air Force in 1946 as an outgrowth of operations research efforts inated during World War Il, Rand (originally RAND, derived from “research and development”) was a nonprofit ‘corporation dedicated to research on military strategy and technology. Rand was primarily funded by contacts from the Air Fore, though it served other government agencies as well. It attraced talented minds though a combination of high salaries, relative autonomy for researchers, and the chance to contribute to policy decisions of the highest importance (Baran 1990, pp. 10, 1), Edwards (199, p. 116) notes that "Rand was the center of civilian intellectual involvement in ‘defense problems of the 1950s, expecially the overarching issue of ‘nuclear politics and strategy.” Rand's role was visible enough to be reflected in popular culture—for example, the ional Dr. Strange- love turns t9 “the Bland Corporation” when he needs advice on ruclea strategy Because is approach to systems analysis emphasized ‘quantkative models and simulation, Rand was also ative in computer science research (Edwards 1996, pp. 122-124). {In 1959 a young engineer named Paul Baran joined Rand's com- puter science department. Immersed in a corporate culture focused fon the Cold War, Baran soon developed an interest in survivable ‘communications, which he felt would decrease the temptation of mi tary leaders to launch a preemptive frst strike Both the US and USSR were building harsiger muclear ballistic mile syteme Ihe trate weapons command and contol systems could be tore survivable, then the county realtor capably coud ater allow to wkhasnd an stack nd il fnction; a more able posison. Bu this at nota wholly feaable concept, because long datance communications networks at time were exwemely vulnerable and ot able to survive atack. That tra the tue Here a mom dangerous situation was created bythe lack of 3 Survrable communication system. (Baran 1990, p. 11 Baran was able to explore this idea without an explicit contract from the Air Force (iid, pp. 12, 16, since Rand had a considerable amount of open-ended funding that researchers could use to pursue projets they deemed relevant to the United State’ defense concerns? ‘Baran began in 1959 with a plan for a minimal communications system that could transmit a simple "Ga/No go" message from the president to commanders by means of AM radio. When Baran pre- Sented this ida wo military officers, they immediately insisted that they Whe Hat ond Cad Wer 11 needed greater communications capacity. Baran spent the next three yeas formulating ideas for a new communications system that would Condine survivability with high capacity (ibid, pp. 14-18). He envie sioned a system would allow military personnel to carry on voice comersations oF to use teletype, facile, or low-speed computer terminals under wartime conditions, The Key to this new syste was a technique that Baran (1960, p. 8) called “distributed communica: tions” In a conventional communications system, such as the tele phone network, switching ie concentrated and hierarchical. Calls go first toa local office, then to 2 regional or national twitching office if «connection beyond the local area is needed. Each user is connected to only one local office, and each local office serves a large numberof users, Thus, destroying a single local ofce would cut off many users fiom the network. A distributed sytem would have many switching nodes, and many links atached wo each node. The redundancy would mabe it harder to cut of service to user. In Baran’ proposed system, each of several hundred switching nodes would be connected to other nodes by as many as eight lines (gure 1.1. Several hundred multiplexing stations would provide a ineface berween the users and the network. Eich mukiplexing sta tion would be connected to two or three awitching nodes and to a8 many a 1024 users with data terminals or digital telephones. The Switching was distributed among all the nodes in the network, 40, knocking outa few important centers would not disable the whole network. To make the system even more secure, Baran (19642, volume VIL, section V) planned to, locate the nodes far from population centers (which were considered military targets), and he designed the ‘muliplexing stations with a wide margin of excess capaciyy (on the assumption that tacks would cause some equipment to fai, Baran added such military features as cryptography and a priority system that would allow high-level were to preempt meses from lower- level users 1 adapted a technique or “soreand-forward switching” A common example of message twitching is the poral system. In a message switching system, each message (eg. a letter) is labeled with its origin and ie destination and is then patted from node to node through the nework. A message is temporarily stored at each node (4, a pos office) until it ean be forwarded tothe next node or the final destination. Each succesive node uses the addres information woe flesmz. oo Pett Pat! Baran’ design feauriog highly connected switching nodes, Source: ‘Baran 196, volume VII We Hat and Cad War 13 to determine the next step ofthe route. Inthe 19905, mestage switch. ing came into we in telegraphy: a message was stored on paper tape at each intermediate station before being transmitted to the next station. Ar Bist, telegraph messages were switched manually by the telegraph operators; however inthe 1960s telegraph offices beyan to ‘us computers to store and route the messages (Camphell-Kelly 1988, p22, For the postal and telegraph systems, message switching was more ‘Bicent than tansmiuing messages or lewers directly from a source toa destination. Letters ate stored temporarily ata post office so that large number can be gathered for each delivery rout. In telegraph, Imessage ritching alo addrested the uneven flow of traffic on the ‘expensive long-distance lines. In periods ofight aff, exces capacity ‘wat wasted; when the lines were overloaded, there waa rsk that some mataget would be lost Storing mesiages at intermediate stations ‘made it posible to even out the flow if line was busy, messages could be stored atthe switch until the line was free. In this way, message ‘itching increased the efficiency, and hence the economy, of long. siancetlegraphy”™ ‘Besides appreciating the efficiency ofered by message switching, Baran saw it a @ way to make his system more survivable, Since the nodes ina message svtching system act independently i processing the messages and there are no preset routes between nodes, the nodes ‘an adapt to changing conditions by picking the route that is bes at ‘ary moment. Baran (1964b, p. 8) described it this way: “There is no ‘ental control only a simple local routing policy is performed at each rode, yet the overall system adapts.” This increases the ability ofthe sytem to survive an atiack, since the nodes can reroute messages around non-functioning parts of the network. Baran realized that survivability depended on more than just having redundant links; the odes mut beable to make use of those extra links. “Survivability.” Baran wrote (1964a, volume V, section 1), “i a function of switching Sexbilty” Therefor, his neework design was characterized by disib> ed routing a well ae distributed links Departures from Other Contemporary Sytem Paul Baran wat not the fist to propose ether message switching ‘or survivable communications to the militar. Systems of both types aleady existed of were in development A look atthe state ofthe art inthese areas makes it easier to see what aspects of Baran's ideas were oe 1 Chapin 1 fom con- realy innovative and why he sw opportunist dps temporary prac era reas Mowage ‘switching systems were nothing new to the Department of putant bur the estung sens wee chmberome and inet Brn donered th on he ered +a meer of Deparment Se pucee commigee daged wi examining several existing oF Sropred rend forward dat ema inthe ety 1860, Thee Pepe ad sack low cpu ha backlogs of mesg tended 1 Und op ae chs Therefore, theses had to be bul ith sae wete capac to bold the mesages chat were wang (0 be stain the itching computers ended up being large and (Sipe Baran nas convinced tha ahetwork could and shuld be ‘Rog much higher tenanison sped, eliminating the Bate aks Nth noes Bese the obvious Bnei o geting megs Tinered fer highspeed, low-orage tem coul ave viching SEE that nere'mich Sper and chesper than thot ed in com rattan more-anorward data syems: As Baran (1964, p 8) Trea oat, ahough the highsperd sym would be woreand- PRO is deage in race menage would spend lite me Tide ored nthe nodes tothe wer thereoe connection wold ee te reaiime Bara agument (190.2) tha 8 os Jae ees mcmgestching nctwork wih fst end-oend ans cee meager and tal expensive sites wae» a “ange wo the sing undertading of sch stems Th Socetof are commuricasns” or “dted et wah so predated Baan indod hx pobeaos ce example of Tea fom mathemati, artic inlignce nd clan and i communeatoms (Baran 186, ome V seton 1 In a TNR” itary planer ad sendy proposed a vari fastens Ci nanctwort ot dcenrlaed nested by mule cone: SES Gd secon 1, Though they shved the dea of daibuted erticions however, tee oer sem dled i ent aortom Berar’ props In paras mst ofthe ee 1 have TEGNe tac of ample brstes techniques, with every menage Sh rey dena, wheres arn’ sem woud vote etna through the network? Tina ofthe dou systems Baran deseibed were nl propo we forking tens. However thre was one ge aired car ctatond seeork under actual development in the aly ‘ous Ths was AUTOVON, designed and operad forthe Depa Whee Heat and Cld War 15 ment of Defense by the American Telephone and Telegraph Corpora- tien. In 1961 AT&T had provided the Army with a communications network called the Switched Circuit Automatic Network, and in 1963 the corporation provided a similar network for the Air Force called North American Air Defense Command/Automatic Dial Switching “The Defense Communications Agency, which was charged with coor inating the provision of communications services throughout the armed services, decided to integrate these networks into anew system Called the Continental United States Automatic Voice Network (CONUS AUTOVON). AUTOVON was not a message switching sy tem; it was a special military voice network buil on top of the existing Grilan telephone network. It went into service with ten switching ‘odes in Apri of 1964 (Schindler 1982, pp. 266-269). Describing the AUTOVON system, AT&T's magazine Long Lins (2965, p. 8) noted: ‘The top requirementis thatthe system can survive tdsasten” Survivability was sought in part by placing the switching caters in "hardened" sites, often underground, away from major ‘metropolitan targets, The min survivalliy feature, however, was that the network was arranged in what ATT called a "polygid" with each svitch connected to eeveral links and withthe links istrbuted evenly throughout the system (rather than having all connections routed trough a few central switches). ATI's publicity stressed that this redundant, decentralized system represented a sharp departure from the hierarchical structure used in the ordinary tall network. AUTO: VON had one node for every few hundred lines, whereas in the regular tll sjstem a node ypcally served afew thousand lines. “The polygrid network, according othe system's architects, "plas a major Fel in the survivabiliy of AUTOVON. Along with its other virtues of fexibilty and economy, polygrid represents the best method that technology can now offer for the rapid and relible connection of munity use of computers” "Cooperative programming,” he contin. ‘ved, “would be stimlated, and in particular fields or disciplines it will, bbe postbe to achieve a ‘critical mas of talent by allowing geographi- cally separated people to work effectively in interaction with a system.” (Roberts 19676, p. 2) Roberts aso saw the network as a chance 10 advance the state of the art in computer science. Ashe knew from his ‘own experience in the Seld, networking techniques were sill in a rudimentary sage, and many theoredcal and practical questions remained unexplored. For ARPA's managers, then, the network project represented a ‘chance to purme advanced research in a new branch of computer ‘cence, potential fnancal savings for the agency andthe fulfilment ‘of vision ofintracsive computing. These goals set the general outine ‘ofthe proposed network, Aswe saw in chapter 1, Roberts decided that, the network should be a disributed packet switching sytem, so a8 to reduce transmission cot, increase reliability, and potently further the military objective of developing sophisticated and survivable com- munications systems. The network would extend across the United States, matching the distribution of ARPA site. t would lnk time sharing computers to support both remote terminal acess to distant comptters and high-volume data transfers between computers. Since the network represented an experiment in data communications tech niques, the scope ofthe project would include not only building the system but also testing and analyzing its performance. Final in order to maximize the resourees avaiable and to save ARPA money on computer facies, Roberts required all IPTO ste to parsicpate by ‘connecting their computers tothe network—whether they wished 10 Roberts bore most ofthe responsibility for seeing the ARPANET project through to a successful conlusion, and his management skills proved invaluable, Building a long-distance packet switching network te connect diverse computers would be a formidable task, even for an agency with ARPA’s resources and its mandate for advanced research Beyond its sheer siz, the ARPANET was one of the most complex computing projects ofits time, pushing forward the state ofthe art in Duin he ARPANET 47 uta communications, To keep the project on rack, Roberts deployed ‘unique eet of technical and managerial strategies. Both the ARPA- INET elf and ARPAs approach to building it would have a lasing influence on the emerging field of computer networking. Ital Challenges ‘As Taylor, Roberts, and other members ofthe ARPA computer science ‘ommunity began working out the design ofthe ARPANET, it became ‘ear that building a network according to ther specifiaions would prevent enormous technical challenges Packet switching was @ risky Choice for the ARPANET; using this novel technique would increase the uncertainty and complexity of the system design and hence the project's chances of falure. In 1967 the world’s first packet switching ‘Omputernetvork was sil n the planning stages (at Britain's National Physical Laboratory, and many experts were openly skeptical that sich a system could work. The fact that the eventual succes of [ARPANET was widely interpreted asa proof ofthe feasibility of packet switching indicates thatthe technique had not previously achieved vide acceptance. Roberts (1988, p. 150) found that telephone engi- heers questioned his credibility for even suggesting such a radical epartre from established practice: "Communications profesional ‘acted with considerable anger and hostility, usually saying T didnot lnow what I was talking abost." Communications experts were fai lar with the dificulty of routing messages individually through = retwork, and i was clear to them that breaking messages into packets ‘would add to the complexity of the system. From their perspective, the activites required of packer switching nodes seemed too dificult tobe performed quickly, reliably, and automaticaly. The communica tons experts atthe Defense Communications Agency were no more ‘pmpatheic, acording to Roberts (1989). Even within the field of Computer scence, criécs pointed out dffcules. Packets sent through the ARPANET would have to be reordered and reassembled into ‘complete mesages at their destinations. Somme experts predicted that this would require excessive amounts of computer memory. Others ‘gue that a routing eystem that changed rapid in response to traffic ‘conditions might rend packets loping endlesly dhrough the network (Rinde 1976, p. 271; Roberts 1978, pp. 1307-1308). Some ofthese problems did, in fact, occur in the ARPANET, and they took consid ‘rable effort to fix owe “Another unusual and potentially troublesome characteristic of the ARPANET was the great varery of computers it would connect Besides machines commercially available from IBM, DEC, GE, SDS, and UNIVAG, the proposed ARPANET sites had various one-of-a-kind ‘machines, such as ARPA's experimental ILLIAC supercomputer (Dick son 1968, p. 182). These various types of computers were incompatible with one another, which meant that users who wanted acess pro- {grams or data at other sites often had to reprogram the software or reformat the data. In 1969, ARPA direcor Eberhardt Rechtin told Congress: "When one user wants to take advantage of another's developments, he presently ha litle recourse except to buy an appro: priate machine or to convert all ofthe orginal software to his own machines." (US Congress 1969, p. 809) Incompatibility wasted time and programming resources, and it remained an obstacle to colabo- rative work. For Roberts, one aim of the ARPANET project was to ‘overcome thete obstacles, Roberts viewed the diversity of computers ‘not a5 an unfortunate necessity but as a strength of the system, since ‘a nework that connected heterogeneous systems could offer users a ‘wider range of resources. But geting this asortment of machines to communicate would require an enormous effort in hardware and software redesign. “Almost every conceivable item of computer hard- ‘ware and software willbe in the network," Roberts pointed out adding “This isthe greatest challenge of the sytem, a well at its greatest ‘imate value” (quoted in Dickion 1968, p. 131) Roberts view was bated on his experience as one ofthe it people to auempt to establish 2 connection between different types of com: puters. After receving his Ph.D. in Hlecrcal Engineering ffom MIT in 1950, Roberts began working atthe Lincoln Laboratory, where he became interested in the possibilty of networking computers for time sharing during discussions with J.C. R. Liklider, Donald Davies, and others in 1964 and 1965 (Roberts 1988, pp. 148-144) Roberts found a Kindred spirit in Thomas Mari, who had studied under Licklider and had founded a time sharing company in Cambridge called the Computer Corporation of America. In 1965, with funding from PTO, Roberts and Marill undertook to build a rudimentary network linking two experimental computer: the TX-2 at the Lincoln Lab and the (Q-82 atthe System Development Corporation in Santa Monica. A ine leased from Western Union provided the communications link, and “Maril and Roberts wrote ther own software to manage the connee- Duding be ARPA tion. They published their results in the fll of 1966, just before Roberts left Lincoln Lab for ARPA In describing their experiment, Marill and Roberts articulated some important concepts. In ther view, the “elementary approach” 1 con- recing two computers was foreach computer to treat the other as @ terminal. Such 2 connection required litle modification of the com- puters, but it had severe imitations. The connection was slow, since terminals operate at much lower data rates than computers, and there nas no general-purpose way to acess a remote system, since each separate application program had to manage its own connections rnther than having the operating system handle the connections for al applications, Marland Roberts thought that forgoing the elemen- tary approach and taking on the harder task of modifying the com- puters operating systems would make it posible to create a Figher-speed comiputer-to-computer interfice instead of relying on the ordinary terminalto-computer interface. They proposed that each ost computer implementa general-purpose set of rules for handling network connection, which they called the “message protocol” (Sarill ‘nd Robert 1965, p. 428). Roberts applied what he had learned fom Us experiment to the design of the ARFANET. He decided that all the hott computers should follow a standard protocol for network interactions, Having a ttandard protocol would help overcome the iacompatbiltes between different types of computers. However, this approach alo created a huge task for the people maintaining the toss, who would have to add this new networking capability «0 the cperatng systems oftheir various computers, (Creating a heterogeneous, packet switching, continent-spanning. ‘omputerto-computer network would be 2 significant technical {achievement for ARPA: the challenge would lien Keeping thee same features from leading the project imo chaos. The technical and mana- frial ditfculier of the ARPANET project became apparent when ‘Taylor and Roberts presented the network concept at IPTO's annual aeeting of Principal Investigators (cents heading research pro- jes) 3 the University of Michigan in April of 1967. Roberts had ‘tread discussed the idea informally wth several ofthe Pls, but atthe necting he announced thatthe project would definitely go forward “The Pls, who would have to design, implement, and use the proposed retwotk, did not greet the network idea with the enthusiasm it would rective i ater years Mott Pls at the meeting reacted with indifference we 50 Chapin? ‘or even hostlity tothe ide of connecting their computer centers 10 the network. Some of them suspected—correctly—that ARPA saw the network as an alternative to buying them more computers. Roberts (1960) reate: Although they knew inthe back of thi mind thai as » good den and ‘were supportive on a phlosophial fone, fom a practi Pont of view, they—Minaly, and McCarey? and everybody wil ter own machine = ‘ame [9 continue having ther own machine, I was oly couple yeas aler they had goten on [de ARPANET] tat they tare raving abou how they could now share research, and oily publish paper and do he things that they could never do before Many Pls did not want to lose control oftheir local computers to ‘people a other sites, and they saw the network at an intrusion’ Since “ther” machines were actually paid for by ARPA, the Pls had litle choice in the mater; however, they were not eager to join in the network. Even those who agreed on the general advantages of devel ‘oping computer networks had practical objections to implementing the ambitious sytem envisioned by Roberts and Taylor. Some of these Pis were unwilling to undertake the massive effort that seemed to be required; others were convinced thatthe project would fal alkogether, esides reminding us that even those atthe forefont of computer science in 1967 could not foresee the astounding popularity of the ARPANET and its successor, the negative reactions of the Principal Tnvestigatorsillustrate the ewo major challenges that ARPA faced. Fist, it was dar thatthe complerity of the network's design would require imaginative technical solutions. Second, ARPA would need to find ways to gain te cooperation of prospective network members The Pls were inally more concerned with continuing their own local projects than with collaborating on a nework. In order for the projet to succeed, Lawrence Roberts would need to create tome sense of common purpose. Sytem: Building Strategies (Of the many problem-solving strategies that Roberts and his team of, contractors would employ in building the ARPANET, two were erpe- cially significant. One was an approach that came to be known af layering, which involved dividing complex neworking tats into modu lar building blocks. The second was an informal and decentralized aiding he ARPANET 51 management syle, Layering and a decentralized, collegial approach 1 management came to be seen by members and observers of the project as estental characteristics ofthe ARPANET, and were later held up as models for success project development; this gave these ‘techniques an influence beyond their role ax management tol forthe ARPANET project? Leyering layered s)stem i organized aa set of discrete functions that iteract cording to specified rule. The functions ae called “ayers” because they are arranged in a conceptual hierarchy that proceeds from the ‘most conerete and physical functions (auch as handling electrical sig "asf the most abstract functions (interpreting human language ‘commands from users), Each higher-level function bulls onthe capa- tiles provided by the liers below. The idea of layering seems to tv occurred independently to many people working on networks 3¢ they drew on concepts of modularity and functional division ofsytems that were current in computer science® {In the ideal layered system, the opportunites or interaction among layers are limited and follow set rues. This reduces the complexity of the system, making it easier wo design, test, and debug. The designer of particular layer needs to know how that layer is expected (0 interact with other layers but docs not need to know anything about the internal workings of those layers. Since the layers are independent, they can be created and modified separately as long as all those working on the system agre to use the same interfaces between layers, ‘Thus, layering has bth technical and socal implications: t makes the technical complexity ofthe system more manageable, andi allows the ‘system tobe designed and builtin a decentralized way. ‘The ARPANET's builders did not wart out witha specific plan for tow functions would be divided up among layers or how the interfaces and protocols would work. Rather, a layered model evolved as the ARPANET developed. The firs step toward a layered approach was ‘uken at the 1967 meeting of Principal Investigators. One of the contractors’ main concerns on frst heating about the project was that Cceating the necessary packet switching software for their computers would require to much efort on their part. 1PTO research sites ured wide varity of ime sharing operating systems; ifthe host computers tad to perform packet switching, someone would have to program we 52 Chater? 2 sitcom sore are eat ieee cy see cid ae Me econ ipegtedaiets teaer tardiio ee re Si arr iaomeesatiot peereenrar Sg vy Crk ting vee oath agit le neat ara i ener eat nk a aoa kee oe rena Maines nie sclan pn conta a ees ote reeen eas e or ae a ete ae Sa rstcnctaty mabe sion ed acne ie aie ger ted at ean ge la ce aa pe (a Hye 3 oe Se ee ec cider ec oma Sate pa eh load we ue [owe tne Figure 23 ‘Nevock mode with communications sobre Duling the ARPS 53 Table 2. ‘The twolayer model af the ARPANET. Layer mame ntons Tox ‘Handles wer inverts ites and manne ‘connections between par of ante Communications Moves data through subnet wing packet switching; ‘nmuesreable rarumion on hove IM and IMP. connections ffom one par ofthe network to another, and the hosts, which were responsible for the content of those packets. Packet switching pro- ‘ams could now be writen fora single ype of IMP computer rather than many diferent types of hosts. Host administrators could treat the entire subnet asa “black box” that provided a service without requ ing them to know how it worked, and could focus their energies on roving host resources. The ARPANET team began tose the system being divided conceptually into two layers: communications layer, cnisting of packet switching IMPs connected by leased telephone lines, and a host lyer, which would coordinate interactions between host processes and provide user services (Heart etal. 1970, p. $51). ‘This model is summarized in table 2. In depicting the network as a "sack" of layers or protocols, the ‘welayer model (able 21) suggests two kinds of relations between system functions. First, the functions become increasingly abstract at ‘ose moves from the bottom to the top of the stack—from moving cecrons over wires to interpreting commands typed by terminal ers. Second, the order ofthe layers represents a temporal sequence from top o bottom: frst the user types a command that invokes the host program, then the host protocol sends packets to the communi- cations subnet. ‘The “protocol stack” model would quickly come to dominate the ‘ay people thought about organising networks precisely because it ‘fered a blueprint for reducing the complexity of network compo- ‘nents while increasing the predictability ofthe system as a whole? Before the ARPANET was finished the model would be expanded to three layers, and in later yeas sill more layers would be added to keep pace with new capabilites and new ideas about how to organize esworks, we 54 Chur 2 Informal Management “Whereas the layering approach stresed separating the sytem’s ele: rents, ARPA's management syle was aimed at fostering the coopera: tion required to integrate chow elements into a coherent whole AARPAs unmatched nancial sources drew many computer scientist ito is projecs, but ARPA managers did not conduct relations with their researchers on a purely financial, contractual basis, The organ lnational culture surrounding the ARFANET was notably decental ined, collegial, and informal. In coordinating te contractors, ARPA relied largely on colaborative grrangements rather than contract obligations, and technical decisions were usually made by consensus ‘The network iself provided 2 new way to coordinate dispersed activ ties and came to function as 2 meeting pce for the computer rience ‘community. Though conflicts ometimes rote among the contactors, the ARPANET culture enhanced ARPA‘ silty to enlist the support ‘ofthe research community and to respond to the echnical challenges thatthe project pore. ‘The collegial management syle of Taylor and Roberts was typical of IPTO in the 1960s and the 1970s. IFTO recruited mott of ite directors and project managers ftom the ranks of active researchers at university and industrial research centers IPTO managers hep in ‘ouch with ther ealeagues by touring contract sites to evaluate the progress of programs learn about new ideas, and reruit promising "esearehers. Not career managers, hey generally stayed at ARPA only 2 few yeas before returning to academia or private busines in part because ARPA salaries were modes). Though ARPA a an oranintion had financial power over is contactors, most ofthe individuals who ‘actually managed IPTO projects were dravn from those contactors. Howard Frank of the Nework Analysis Corporation, an ARPANET contractor, observed: “Its easy to say the government. or ARPA, oF something ike that, but they are individual that you deal with” (rank 1990, p. $00) ‘ARPA tended to aad contracts through an informal proces, d= ing individuals or organizations who were already known to IPTO. managers for thet expert in particular are, The ARPA approach ‘exhibited the weaknesses and th advantages ofan “old boy" network Many alented computer sient found themlies lft out of the Sls biggest funding opportuni, bu those who were included en- joyed an extremely supportive environment. Wesley Clark (1990) commented: “In the ARPA system, once you were in, you were 2 Duling te ARPANET. 55 member of cub... wth ...a prety good sense of community with tether people who were receiving support from that ofc.” Robert “Taylor made a spedal point of providing ongoing funding for gradu ate students at contract sts, and he arranged special meetings and ‘working groups for ther (Taylor 1989, p. 1). Graduates ofthe 1PTO- funded programs at MET, Sanford, Carnegie Mellon, and elewhere ‘became a major source of computer science faculty at American uni verte, thereby extending ARPAs social network into the next gen- cation of researchers (Norberg and O'Neill 196, pp. 200-201) " Tn view ofthe expense of computing machinery in the 1960s and ARPA‘s large role in funding computer science, IPTO managers had real power over ther contractor, and they were willing to use this power when they elt ic necesary. As has already been noted, Robert "Taylor exerted pressure on Lawrence Roberts 0 leave his position t the Lincoln Lab and jin ARPA. Once in charge ofthe project, Roberts id not hesitate to make reluctant contractors share inthe ARPANET fore “The universes wer beng faded bys, and we sid, "We are going build s nework and you ae ging wo pareipate int And you are pang ws connect ‘eto your machines By vito of that we are ging t reduce our computing demands onthe of So that you understand, we are not going buy You ‘ew computers und you have uted up al ofthe resources ofthe network ™ Sooner Ue we narted forcing them o be invelted. (Robert 1989). ‘But IFTO managers preferred to tae the informal approach when- ever posible. Having been researchers themselves, they subscribed 10 the vew that the bes way to get results in basic reeeach was t find talented people and give chem room to work as they saw ft They also tended to believe that differences of opinion could be debated ration: ally by the partes involved and decided on their echnical merits, and that they, a IPTO managers, would need to intervene with an execu tive decision only if the contaciors could not resolve diferences among themselves. Not surprisingly, IPTO contractors praised this ‘management syle as an enlightened and productive way to conduct research, The report of an outside consultant commissioned by ARPA to report on the project's watus in 1972 agreed thatthe project's informal syle had contributed to its success, noting that the process of building the ARPANET had "been handled in rather informal fashion with a great dal of autonomy and an indefinite division of| responsibilities among the organizations that address the various ele: rents of this function." The report continued: "Personal contact, we 36 Char? telephone conversations, and understandings are relied upon for day to day operation. This environment is a natural outcome ofthe pro- reste RED atmosphere that wat necessary for the development and Implementation of the network concept” (RCA Service Company 1972, p. 34)" In view of the nature of the projec, t made sense for Roberts to encourage ARPAs network contractors to work together as peers Dilferent tasks required diferent combinations of skis, and no one contractor had the overall expertise or authority to direct the others {subordinates Roberts informal coordination methods provided a ‘content in which the network builders could, for the most part, share ‘ils and insights on an equal and cordial basis Getting Started Roberts began the ARFANET projet informally. Rather than soliciting bids for contacts right away, he brought together a small group of Principal Investigators who had expressed interest in the network concept and began meeting with tem to dicuss design problems and ‘to work out posible soluions. He asked Elmer Shapiro ofthe Stanford Research Instte to lea thee meetings, and he recruited Paul Baran ofthe Rand Corporation (the man who had done the exit work on tiseibuted communications and parket switching) to advise the group. "The various members of thie group incorporated their own values into the ARPANET design. For instance, while time sharing entha- sins insisted on very fat response, so that users woul not be frus- trated by long delay, more analytically oriented researchers such as UCLA® Leonard Kleinrock insted on incorporating measurement software into the switches 40 that they would be able to study the retwork’s performance (Kleinrock 1900). The istutional homes of the members of this seltsleted group would also become the first four nodes ofthe network: the University of California at Santa Ba bara the Stanford Research Inuit, dhe University of Utah, and the University of California at Loe Angeles. tp June of 1968, Roberts submitted the plan his group had worked out to ARPA director Hers Feld In July, Roberts received an inal development budget of $22 nln and approval to sec contractors to develop the network * "The basic infrastructure ofthe ARPANET would consist of time sharing hosts, packet siching interface message procesors, and leased 56-Llbiteper-tecond telephone lines to connect the IMPs, The hosts were alread in place and the lines would be provided by ATT, Duling the ARPANEs 57 40 the main development task wat to build the IMPs, Unlike most PTO project, which were inated by contractors who had aleady shown interest and expertise ina given area, building a packet switch ing computer was @ new venture iniated by ARPA, and there was no comvious candidate forthe job. Therefore, Roberts departed from [ARPA'susal praice and solited competive bide fr the IMP con- tact from a number of computer and engineering firms. In carly 1060, ater considering bids from a dozen companies ofall sizes, Roberts awarded the contact to the Bok, Beranek and Newanan Corporation of Cambridge, Mateachusets, a relatively small company specializing in acoustics and computing systems ‘Though nota giant inthe computer business, Bok, Beranek and Newman had several advantages behind its bid. The company had previous tis with IPTO: JC. R.Licklder had worked at BBN before becoming the fist director of PTO, BBN had contributed to PTO’: carler time sharing efforts, and BBN researcher Robert Kahn had ‘dscusted networking with Roberts during the eariy wages in the planning of the ARPANET: BBN wes also known fr it strength in research, In 1990, Kahn recalled BBN as having been “a kind of hybrid version of Harvard and MIT in the sense that most of the people there were either faculty or former faculty of ether Harvard fr MIT" and as "sort ofthe cognac of the research busines, very send packets™ To find the shortest routes, the IMP kept 4 table vith an entry foreach ost onthe nework, showing how longit would take a packet sent from the IMP to reach that host and which ofthe [MPs links led to hat ost by the fastest route. When 2 packet came in, the IMP would look up the desination host in the table and [brvatd the packet vt the specified link The rowing sytem was also sotble for beng adaptive, continually responding to changes in net ‘ork configuration oF traffe. Every 2 second, the IMP would make {new eximate of how long it would take to reach the various host Aestnatons, and ie would send these routing eimates to each of is tseighbors The IMP used the information sent in by is neighbors to we 62 Caper? rae fx own routing abl, matching ech host destination with dhe EK thar hed reported the shorts travel time to that ost. This innovate approwch to routing served ARPA: goa of building a ag ge exile syuem, Disrbuted roving made the system more robust ‘eimimring is dependence on any ove component. piv rou {ig lowed IMPs to improve the peed and relabity ofthe network ‘praoldng congested routes and pode o ine lures, The pie of ‘eying on 80 many independent, constantly changing routing dec erehoncver waa eompler stem pron to unexpected mera: oon Avast Bt, Beranck and Newma's IMP team had torevise i rowing algrthn sever ties as experience or sulin ‘veled weakness (Oran e172. 24; Slings 1991), The [ARPANET approach to routing reflected is designer commitment to exploring new techniques and bulding = high-performance ete Srorkeven atthe pie of resting aye that way at time, ile TDundertand nd contol “The imerfce and packet switching fancon jst dered had teen spced by the ARPA contac But the design ofthe IMP wat Tio aaped by the BBN group rong bebe about how ie thould Ferformin relation to heres of the network In partic the BBN Tm tied to enforce the disincions betwen network layers. AB ‘plnaion ofthe IMP design deciions by team members John NMeOuilan and David Walden arcusted thee bel thatthe set ‘ould be ocd fom any potenti iererence fom the hot com uterein ether words, cat the communica’ and ox hyers Should be separate: ‘teeing of funciona irrcy of cont een in» complex ‘ewer Suite For inc noe [IMP mast cont saknetwork ‘Suter and Hos mas corel Hot remuec or reat, de bac "Scene mu bw cn ore oe nde ogra, Tor anny, he andar message procengPo- oe thou be ale sire, rh canbe Gunged under coal el uch more fn tn ao props (tan snd en 197, rim Tn the BBN vision, te IMP subnet wa to be autonomous. Hosts woul ‘ie toate from tures inthe suet ule moss own fea IMP tere diabled, IMP would not depend on hon for any computing ours or lformaton, and the fencning ofan IMP would not Temps if localhost went down. Nor mould he people athe Dulin the ARPANET 63 hos ses be ale tonterfere wth the operation ofan IMP in ay way. Heat was paral eoncered tha nga graduate tens would vane to experiment wth thi lal TM. pedal since com ites fay sort were alla arc commodity at mon ser He was Zou o maketh IMPs selfcotained, otha would be hard or Solent to tmp with chem Gate and Lyon 199; pp. 12,150). Aspraciced by Here ands group. the ehuiquoflayrng become 2 ray to manage scl reabons a wells to Fete cca com Plety, Designing the ane o operate independent ofthe hosts tale the network more rou, ext the techn tak of te BEN team and alowed the team to fninain contol ever the sgn snd ‘pion o I “the BBN group tok sever! teps to make te operation of the Ps depend a dem pomible onthe hos. on oer IMP oto human eperatos Rather than counting on the netvork to fincion ‘cat, IMs alway checked fr lt oF Spat pct and ech IMP teed peridial for dea lines, aes nneghbering IMP, ‘o-fanesoning host, or dsinaons ade unreahae by terme. diate IMP orn fares The need for human neremon the subnet was. minimized by “oggeising™ the TMP hardware (Rigging a common procedure for spies of miltary com puter ented proteing the machine gan emperaure change ‘stn, radio interference, nd power sarges) The team bil the IMP capable for remote monitoring tnd contol ht allowed BBNsatfmenberto rn dlagnone procures ot ond safer on n IMP wihout making Bed vist orrehing on ea operators “The IMP ao was diged to recover rm aon aren IMP ‘hacen’ down aa ra of power flue would restart automat icy when power returned. Exch IMP checked perio taxi ‘sta operating program had been damaged io. woul request 2 nighboring IMP to send a copy ofthe program to replace the ‘ormped vein. Ifthe malfnctning IMP rn unable ela the new copy it woud automaticaly hut elf down to prove the nee vers frm any detrucne behavior the damaged ore might {aut Byaniptng and solving wh dy to-do minenance pre leh BBN tea ves praca fom thee of dred ‘cork on he commaictions ayer proceeded, Roberts made new get Bit any tine conning Teena et bot cure—ipety a up epee the nk ee Rye an eel ten Wiha BBN, here was teson btneen Alex Meese’ groupatthe MERE Conard Comer boxe priory wast keep theese up and Nene Sty snd the Ii developers who wanted o understand ne Sind nctwork maoncons 0 ao prevent Furnes Titan IMP ald, he development eam would often keep ot {7%Stuuon forever! hour we they debugged rater than weeny renoing to mri. Hear (1900) commented atthe ‘iir'acslopen cae under increasing pressure a the neswork rae eet ecme more heniy wit: "People began o depend Sana thr wasn problem, because hat meat when You Pett ort hed poems, they all got mad, So hat we 0 enki rd” The BBN group eventual resolved hs confit ree Be ew slvr wl tht wold mabe possible wo dig estat herping te ot of verve (Ores ta 107% ne 1, esimony tthe ecient of ARPS management ses ae sete de re onic finer between contac, the otis perdi remained oe of elaboration, In 172 confer saree repacnves ofthe tree main contacory—Howard FRAPU aC never abn of BBN, and Leoard Kloook of Uglied now the ARDANET had provided are oppor NGLBs cntaborain srs dips (rank, Kah, and Kldaock Tas ey perceed tr ot fir a something une cone sae Bese ourapprouhe and phibsophies” Frank tal (id, Payne, have fen ier rail anda rsa 28 hs 9 enor undsrting pce On the other band we have TESS aur cateoraon to be exremely revarding” Though they ie! nth prec fr ta maton, compte sar gineering expenmet ser two Jaro experience Taree Thing teconede taal of tee meds ae weal while Re apowerfe The mon valuable approach they contin nore bec te smutancus as fever of thee took” Gi, pa) ee ae of understanding among dicipnes was considered the ony ARPA aempe toring them tgeher i all he more edie uluwaung exiting socal netrork,cenng new mar Pung te ARPANET 73 agement mechanisms to promote system-wide ies, and insisting on ‘olbboration among groups all sided ARPA's socal and technical integration ofthe eytem, Informaity: The Netork Working Group ‘One of the mas important mechanisms for pooling efforts and build ing consensus among the sattered sites was the Neswork Working (Group. In assigning the NWG to create the host protocol, Lawrence Rater had esrsted an important aspect ofthe ter tf vey incxperncedvercarchere, Vinton Get then» Grane Student at UCLA, described ica flows" We were ju rank amaters, and we were expecting that tome author would nly come along and sy. Heres how we are ging todo K: And nobody ever came dong” (Cert 180, p. 11) Sepen Crocker recalled “The fst fw Inetngs were ite tenvou. We had o fica charter Mont of wer graduate students and we expected thts profesional crew ‘old show up evenly to ae over the problems we mee dealing {Tih (quoted in Reyols and Poel 198) The ck of exabched thre andthe newnes of the ld meant hat he NC's part pana had to formulate echnical problems and propor shone on {her own." We were al fing ott vay beet there wat any body ‘fearrent experi or knowlgeor anything” Alex Mek (100, P. 8 recabed When an oie sierer—he RCA Serie Company {igi A40)—asted why ARPA manages did nottake a more acne ‘olin defining the How prowl poled ou thats en ic fd the appropiate ent fr Shirai co lend of management and weal prblon in tht the dct at wold be mad are ery inporat Snd A any Svan re» ty ed oe oni ‘xpetecs (wel he ay fcr rh ge me pepe ‘to realize a successful implementation, — [AL one point, Roberts, disappointed with the slow progress of the [NWG, considered curing over the host protocols wo a profesional research team. Inthe end, however, he decided to sick with the NWG. in pare because he sensed that the group increased the contractors sense of involement in and commitment to the network, At Carr Crosker, and Cerf reported toa 1970 computing conference, the NWG: provided a unique collaborative experience: We have found hat, in the process of connecting machines and operat ing ems togethers great Seal of apport ha ben sas between see 14 Chap? sonnel at he various network node ses The reg inte oes Ferre dunreement, and resusons fas been highly refreshing and Ceca i evolved. and me regard the uan neacton a aable ten fthe um elo ear Crocs, and Cet 1970, pp, 529-800) Te fay, the group's very Ick of frm blueprint for its actions gave Paci needed to balance technical and orgaizatonl sues TRE NWG developed its own scl mechanisms to cae the challenges Traced. Acting ona suggestion by Eimer Shapiro, Crocker proposed seta! prponls and minutes of meetings be debuted 38 2 set dccunes called Requests for Comments (RFC). Another ‘Gekardent Jon Pose, took on the job of eng these documents Tae Rice were specially designed to promot informal commun roe ra he shring fen in the absence of technical certainty of SRetaued autorin, The NWC's “Documentation Conventions” sated “The content of NWG note maybe any toogt upgeion ela 2 Beer aed npr te newer Faria pont HOS iar ar specs specie vgn or implementation vie croc ortacagound expan ond eit ‘cc empl anor ae a scp, Thee Sanda aor acco ex for to revo Ft here 1nd CY Ce ey ect fc ute aa Boe Pome tevin ai camo of conceal tes than ahora Hes eae lhe 0 push ometingwnphe,an ope eae nibdon (Cocker 196) ‘uring their fet few years, the RECS were, of necessity, disributed apie, however once the network wat funcona, dhe RFCS were Tepe cine a the Stanford Research Instat’s Network Information Caner and were acesed through the ARFANET. Members of the Nensrk Working Group would post new RFCs concurring with, it Qing or elaborating on ideas presented in earer RFCs, and an caplng dncusion developed. Eventually fer members had debated hvac through RFCe and at NWG meeting, 2 consensus would merge on protocols and procedure, and this consensus was generally ‘SErepnd by ARPA as oficial poiy forthe network. RFCs enabled the [WE to evolve formal standards informally Sheping the Pltical Ensironment ‘One potential source of tension that doesnot sem to ave arisen Sahat the ARPANET community was the involvement of wniverity Duling the ARPANET 75 rsearcher-—manyof them student itary poe ur ‘athe Veta Wr pd hte nck eh wat tetinhereny deste ad oo sete dene pn tba. espe the southon ofthe sademsomtary ier ‘erly reflect the elaclecion of resaches wo et xen tha thindon Howes ka uve ha TFT manages were abet ante crm fir ther contac tempat roe ‘her than lly ajecvs. “Toa lge event ARPA managers eral wo held hi esech ros fm sana pol, hh somes ocd withthe cacy ove prin ARPA upper managemen bese ep at Drege pen arches fom compen ry and feo demands tat ny rode exp ny utenti wort Inthe ine 1908 ad the 100 here were a momber of US presents hotel tat dene mnay hold be epet ny cn projet cna} ed woalary mason (Morey sd ©" 986 poy they fe hr he Depremet of Doers we eoming oo tole nding bane race epecalniw othe sr Provided by cla genic In 1908, 2 pret of US government Rinding for unveraiy tence ame fon the Dpartnent of Dees ty 18 percem fom the Natonal Scene Foundation, in 168 Siw tg one va lethal hat ofthe Natal Sees Frmdaon (hms 1972. 85). Durng that years Senate hea irgon the dene badger, Sensor Mie Manel of Montana al tamed Joho 8 Foner the Dieaor ef Dears Reranch. and Engines to explin why te Deparment of Defense shuld pend finch neo breach hn the NSE Sear Mansa the nner pray de othe poy hat esr tcp mony fr rsarch and devlpient inthe Department of Deese tbaniin any ober department othe Covetomen” Rater No, sir Lev the renton is deeper aleve ie deeper eve tthe reason ster ered pro or ty Te ac oy agin oe ane ote cag ary fu, as Robert Taylor (1989, p. 27) privately died, the National SSience Foundation's budget requents recived clover sratiny from Congress than ARPA, since “he research pices of the Department Defense at compared tothe development piece ofthe Department (Defense were mnuacl, whereas the National Science Foundation stint a research orgciaion™ wee a 76 Char 2 ee eee pi eerteedleenteh td Se Geach atten athens Se eee anne ee as be oe aman ice oamscetsnen mee oe eee iocee teapot cate Oh Sgr a eee Serena et teat ce cere guerre iat Faas coney ant i a ar ea faormmetecnc it ia san ne ee co em see ete ay geese Sueenraneceanoe peice Poy temansen eats sees okra ag ft fae weno aged era Se rit yeti Go ie eyo Soe a eect a ee Seep auratus center mer attain were sea ase ee one ee oe se a at ea er ened ee Sn er terete sear ng nr ened espa es eto Soa eenlerautemaecm oats Seca en abana ttcarenty a “Taken a face val, this statement might have surprised IPTO's aca demic contractors sine the agency wart the same time assuring them Dring he ARPANET 77 of iu support for basi research and graduate education. Many of TOs computer science projects were proposed bythe researcers {hemseres, or were designed to allow researchers to continue work in areas they had explored independently. Of IPTO, ‘Taylor (1989, P. 10-1) sid thi wee ot cosine find something only eee ff my scemaces:When temic Care Hel ho warned otARER hein at ane ate ARPANET an ic ad ke mney ah ye othr par of ARPA og ng ff rnd he Boe “specifically ask me for a defense rationale. = Even ithe reming technologies eventually became part ofthe mile tary oxmand and control system, the defense rationale might come er she fc. Descriing hs interactions with ARPA in he 1070, Kleinrock acknowledged: “Every time I wrote a proposal T had te show the relevance to the military's applications” But he aimed, “lt was not at all imposed onus": he and his colleagues would ‘come ap with their own ideas and then suggest military applications forthe research” Wesley Clark's view wat that, though {PTO cou tracts alvays specified some deliverable forthe military. “Estendaly they were funding research with fairly loosely defined objectives And the idea was to help them, whenever they needed help, to jai the work rou were doing with respect co their sponsors in turn, the Department of Defense in general." (Clark 1990) Obvious, ARPA contactors did not have absolute intellectual fee- ‘dom. Vinton Ger (who became an IPTO program manager in the mid 1870s) commented in 1900 that, aluhough Principal Investigators at universities acted as buffers beeen their graduate students and. the Department of Defense, chus allowing students to focus on the esearch without necessaily having t confront its military implies tions, ths only diguised and did not negate the fact that muitary imperxives drove the research (Cert 1900, p. $8) Ths was especialy true inthe late 1970s and the 1980s, when ARPA began to intent ‘raphats on defense applications. However, during te peried during which the ARPANET was bul, computer scientists perio! ARPA se tle wo provide research funding with few strings tached, and this Perception made them more wiling to paridpate in ARPA project ‘The ARPA manages’ skill at constructing an acceptable image of the ARFANET and similar projec for Congres ensured a continuation of iberl funding for the project and minimized outside srutny. In see 78 Chapte 2 his way ARPA was able vo generate vuppor rom both ts poll and its reverch constituencies, Launching the Sytem By the end of 1971 mot ofthe infrastructure forthe ARPANET wa in place The Bcen original ser were all connected to the network, Lee cee ween tere hnetnegor hte tat tes on the network were only inimaly nvled in rxoures Sereerinmpercaaa oduct hat hd ten amped, Though te bars nd Se See aero Sanear nom rm cia oe a ea seminars ymafaree Sh fest papa Shoat Saas elas waiving Sead nae ee aretaseaten seat Ryan et icheatatemas aa hc atte thc ea wearer pectic cada eaenatoe romero Fear fomuectstaemmencnraea er as ey ec EL ooastiacmiemaletacee = Tyme cn te wacrema ter mre Sedpsennaa tee vyeemesodast Soriano eis Snr arto cre ce cams mh Sic ata tata eamre cena aromatase ie Soctanumecmmaeanere na arenas aniramass Bunge ARPANET 79 ‘commanity and an increate in ale over the network (Ornstein et 1872) By the time the First International Conference on Computer Communications opened, enough programs were ready to capture the attention ofthe crowd. ‘The thousand oF 90 people who traveled to Washington for the ICO were able to witnes a remarkable technological feat From 3 emorstaion aea containing dozens of computer termina, ate dees were able to use the ARPANET to access computes located hundreds o thousands of miles away there was even a temporary ink to Par Software on these computers allowed participant to try out meteorological model, an air trafic simulator, conferencing stems, 2 mathemati sytem, experimental databats, sytem for displaying Chinese characters, a computerized chess player, Joseph Weizen- ‘Daum pychitist program Elza and a variety of other applications (Roberts and Kahn 1972) The diverse terminals, computers, and rogram, all operating succesfully and responsvely, some across ‘considerable dances, made a powerful impresion. Cet (1090, p25) lucer described visting engineer a having been “jus a excited at Tie Hd, because all these neat things were going on." Another observer recalled: “There was more than one person exclaiming, "Wor! Whats thi thing?” (Lynch and Rose 1993, p10) The trade Journal Electrons (107, p. 36), cing "the great intereat in computer ‘newoths indicated by .- dhe crowds in the Arpanet demonstration oom” dedared networs “dearly the wave ofthe future.” ‘The ARPANET contractors had reported on the progress ofthe evelsing network at various profesional conferences, but the response tothe 1972 demonstration uggents that ther colleagues did not necessarily take these reports seriously uni they saw the network in actbn. "Te was the watershed event that made people suddenly real that packet switching was areal technology” recalled Kahn (1980, p. 3). The sheer complexity of the sytem, Roberts (1978, _. 1308) believed, was enough wo make engineers skeptical unt they ritnesed ifr themselves: 1 wa ica for many experienced profesional that time to asept the fh thn cleo ef compete, eiebaed est and snore {thr noder pect equip taling well overs undred~—cel al Fanaiosogetier slaty, Sa the ARPANET demonseatoa ited for he ign ad cealy gaye i relate peran a pub Cerf (900, pp. 25-26) noted “a major change in attitude” among. “diehard cireie switching. people from the telephone industry one 20 Chap 2 “Though these communications experts had, with some justification, been skeptical of the idea of packet switching, they were able to Zpprecine the significance ofthe ARPANET demonstration, and it trould be only a few years before the telephone companies started planning packet switching network oftheir own, “The 1O0C demonstration marked staring poin inthe use of the [ARPANET: Packet trafic onthe netfork, which had been growing by Gn 2 few perent per month jumped by 67 percent inthe month of the conference and maintained high growth rates aferward {Sehclonka 1976, pp: 5-20. In addon, che enthusiastic response to the demonstration encouraged some ARPANET contactors to start the fre commercial packet svitching networks In 1972 a group of ‘engineers left BBN eo form their own company, Packet Communicr oun They to. mathet an ARPANETiike service. BBN quickly Teaponded to this defection by launching is own network sobsidiany, ‘Telenet Communications Corporation, and Roberts lft ARPA to be ome Telene's president. Telenet was the first network to each the ‘earth. inating service to seven US ties in August 1978. These new ‘neworks began to offer the general publ che kind of reliable, ost ‘ficient data communications thatthe ARPANET had provided for @ telat few. "The triumphant public debut ofthe ARPANET was the culmination “ot several years of intense work in which the 1PTO community devel Gped a sion of what a network should be and worked out the ‘Roniques that would make this vision a reality. Creating the ARPA- INET was a formidable task that presented a wide range of technical bstaces and conflicts of interest. Inthe fae of these challenges, the soeces ofthe projec depended onthe ability ofthe system's builders to foster a collaborative val dynamic among contractors, maintain ‘Banda support rom Congress, and reduce the technical complexity ofthe system through techniques such a layering "These sategice had lating implications. ARPA did not invent the iden of layering: however, dhe ARPANET s sucess popularized layer- ing asa networking technique and made one particular version of layering a prominent model for builders of other networks. The UARPANET alo influenced the design of computers by highlighting the Gifeulies tha exiting machines encountered in a networked envc onment and offering some solations. The ost system programmers hhad demonstrated how to redesign operating systems to incorporate ‘ommonications functions, and. experience with the ARPANET Dating he ARPANET 51 ‘encouraged hardware designers to develop terminals chat could be ‘ed witha variety of ystems rather than just single local computer (Grocher eta. 1972, p. 275; Roberts 1970; Ornstein et al. 1972, 236 “The community that formed around the ARPANET ensured thats techniques would be dicused in profesional forum, aught in con ie tence deprimenty and implemented income yes ncouraged its contactors to publish thes findings and pro- ‘vided ‘unding for them wo present papers at conferences. Deualed sccouna of the ARFANET in the professional computer journal dis: serinited its techniques and legitimized packet switching a reliable and economic atemative for data communications (Roberts 1988, . 149; Leonard Kleinroc’s work became the bade reference in ‘queuing theory for computer networks, and a number of gradu- ‘te moslente that Kleinrock and others had supported based their later careers om expertise they had acquired while working on the [ARPANET” ARPA alo encouraged its contractors to turn their ARPA NET experience w commercial uss, as Lawrence Roberts had done With Telenet. ARPA’s funding of Principal Investigators, its careful Culivaion of graduate students, and its nsstence that al contractors take pit inthe network project ensured that personnel a the major US computing research centers were commited to and experienced with the ARPANET technology, The ARFANET would train a whole {eneraion of American computer scientist to understand, we, and Soca ks new networking techniques. use

You might also like