Other articles in the series:
- The History of Relays
- The History of Electronic Computers
- The History of the Transistor
- The History of the Internet
- The Era of Fragmentation
In the 1968 paper "Computer as a Communication Device," written during the development of ARPANET, and stated that the merging of computers would not be limited to creating separate networks. They predicted that these networks would combine into an "intermittent network of networks" that would integrate "various equipment for processing and storing information" into an interconnected whole. Within less than ten years, these initially theoretical discussions garnered immediate practical interest. By the mid-1970s, computer networks began to spread rapidly.
The Spread of Networks
They permeated various media, institutions, and locations. ALOHAnet was one of several new academic networks funded by ARPA in the early 1970s. Among others were PRNET, linking trucks with packet radio, and satellite SATNET. Other countries developed their own research networks based on similar principles, especially Britain and France. Local networks proliferated even faster due to smaller scale and lower costs. Besides Ethernet from Xerox PARC, one could find Octopus at the Lawrence Radiation Laboratory in Berkeley, California; Ring at Cambridge University; and Mark II at the British National Physical Laboratory.
Around the same time, commercial enterprises began offering paid access to private packet networks. This opened a new national market for online computing services. In the 1960s, various companies launched ventures that provided access to specialized databases (legal and financial) or time-shared computers to anyone with their own terminal. However, accessing them across the country via the regular telephone network was prohibitively expensive, making it difficult for these networks to expand beyond local markets. A few larger firms (for example, Tymshare) created their own internal networks, but commercial packet networks lowered the cost of using them to reasonable levels.
The first such network emerged after the departure of experts from ARPANET. In 1972, several employees left Bolt, Beranek, and Newman (BBN), the company responsible for creating and operating ARPANET, and formed Packet Communications, Inc. Although this company ultimately failed, the sudden shock served as a catalyst for BBN to create its own private network, Telenet. Led by Larry Roberts, the architect of ARPANET, Telenet successfully operated for five years before being acquired by GTE.
Given the emergence of such diverse networks, how could Licklider and Taylor have predicted the rise of a unified system? Even if it were organizationally possible to simply integrate all these systems with ARPANETâwhich was not feasibleâthe incompatibility of their protocols made it impossible. Nevertheless, in the end, all these heterogeneous networks (and their descendants) did connect with each other into a universal communication system known to us as the internet. It all began not with any grant or grand plan but with a forgotten research project that a middle manager at ARPA was working on. .
The Problem of Bob Kahn
Kahn obtained his Ph.D. in signal processing at Princeton in 1964 while casually playing golf on the nearby courses. After briefly working as a professor at MIT, he joined BBN, initially intending to take time off to immerse himself in the industry to learn how practical people determined which problems were worth researching. Coincidentally, his work at BBN involved investigating potential behaviors of computer networksâshortly after, BBN was commissioned for ARPANET. Kahn was deeply involved in this project, providing most of the foundational developments regarding network architecture.
Kahn's photo from a 1974 newspaper
His "short vacation" turned into a six-year stint where Kahn served as a network expert at BBN during the development of ARPANET to full operational capacity. By 1972, he was exhausted by the topic, and more importantly, tired of grappling with the constant politicking and infighting among the heads of BBN departments. Thus, he accepted an offer from Larry Roberts (even before Roberts left to create Telenet) and became a program manager at ARPA, leading the development of automatic production technology, with the potential to oversee millions of dollars in investments. He abandoned his work on ARPANET and decided to start anew in a different field.
However, just months after arriving in Washington, D.C., Congress scrapped the automatic production project. Kahn wanted to pack up and return to Cambridge immediately, but Roberts convinced him to stay and help develop new networking projects for ARPA. Stuck in the constraints of his own knowledge, Kahn found himself managing PRNET, a packet radio network intended to provide military operations with the advantages of packet-switched networks.
The PRNET project, launched under the auspices of the Stanford Research Institute (SRI), aimed to extend the basic technical core of ALOHANET, which dealt with packet transmission, to support repeaters and work with multiple stations, including moving vans. However, Kahn quickly realized that such a network would yield little benefit, as it was a computer network with virtually no computers. When it became operational in 1975, it had one computer from SRI and four relays located along the San Francisco Bay. Mobile field stations could not feasibly operate with the size and power requirements of 1970s mainframes. All significant computing resources were within ARPANET, which utilized an entirely different set of protocols and was unable to interpret messages received from PRNET. He became curious about how this nascent network could be linked with its far more mature counterpart.
Kahn reached out to an old acquaintance from the early days of ARPANET for help with the answer. became interested in computers while studying mathematics at Stanford and decided to return to graduate school in computer science at the University of California, Los Angeles (UCLA), after working for several years at IBM. He arrived in 1967, and together with his high school friend Steve Crocker, joined the network measurement center led by Leonard Kleinrock, which was a part of ARPANET at UCLA. There, he and Crocker became experts in protocol development and key members of the working group that developed both the Network Control Program (NCP) for sending messages over ARPANET and high-level file transfer and remote login protocols.
A photo of Cerf from a 1974 newspaper
Cerf met Kahn in the early 1970s when Kahn arrived at UCLA from BBN to test the network under load. He created congestion on the network using software developed by Cerf that generated artificial traffic. As Kahn expected, the network could not handle the load, and he recommended making changes to improve congestion management. In the following years, Cerf continued his promising academic career. Around the same time that Kahn moved from BBN to Washington, Cerf went to the other coast to take a position as an adjunct professor at Stanford.
Kahn knew a lot about computer networks but had no experience in protocol development â he was involved in signal processing rather than computer science. He knew that Cerf would perfectly complement his skills, and this would be critically important for any attempt to connect ARPANET with PRNET. Kahn reached out to him regarding interoperability, and in 1973 they met several times before retreating to a hotel in Palo Alto to produce their fruitful work 'A Protocol for Internetwork Packet Communication,' published in May 1974 in IEEE Transactions on Communications. There, the design of the 'Transmission Control Program' (TCP) (soon the P became 'protocol') was introduced â the cornerstone for the software of the modern internet.
External influence
No pair of people or moment is more closely linked to the invention of the internet than Cerf and Kahn and their 1974 paper. Yet the creation of the internet was not an event that occurred at a specific point in timeâit was a process unfolding over many years of development. The original protocol described by Cerf and Kahn in their 1974 paper has been modified and adjusted countless times in the years that followed. The first inter-network connection was tested only in 1977; the protocol was divided into two layersâthe ubiquitous TCP and IPâonly in 1978; ARPANET began using it for its purposes only in 1982 (this timeline of the internet's emergence can be extended to 1995, when the U.S. government removed the firewall between government-funded academic internet and commercial networks). The list of participants in this invention process expanded far beyond these two names. In the early years, an organization called the International Networking Working Group (INWG) served as the main body for collaboration.
ARPANET entered the broader technical world in October 1972 at the first international computer communications conference held at the Washington Hilton hotel with its modernist curves. In addition to Americans like Cerf and Kahn, several prominent networking experts from Europe were present, including from France and Donald Davies from Britain. At the urging of Larry Roberts, they decided to form an international working group to discuss packet-switching systems and protocols, similar to the working group on networks that established protocols for ARPANET. Cerf, who had recently become a professor at Stanford, agreed to serve as the chair. One of their first topics was the issue of inter-networking.
Among the important early participants in this discussion was Robert Metcalfe, whom we have already met as the architect of Ethernet at Xerox PARC. Although Metcalfe could not tell this to his colleagues, by the time Cerf and Kahn's paper was published, he had long been developing his own internet protocol, the PARC Universal Packet, or PUP.
The demand for the internet at Xerox surged as soon as the successful Ethernet network was established at Alto. PARC also had another local network of Data General Nova minicomputers, and of course, there was still ARPANET. The leaders at PARC looked to the future and realized that each Xerox base should have its own Ethernet and that they needed to interconnect somehow (possibly through their internal equivalent of ARPANET for Xerox). To masquerade as a regular message, the PUP packet was encapsulated within other packets of any network it traversed â letâs say, PARC Ethernet. When the packet reached the gateway computer between Ethernet and another network (like ARPANET), that computer would unwrap the PUP packet, read its address, and then re-wrap it in an ARPANET packet with the appropriate headers, sending it to the specified address.
Although Metcalfe couldn't directly share what was happening at Xerox, the practical experience he gained inevitably seeped into discussions at INWG. Evidence of his influence is seen in the fact that in the 1974 work, Cerf and Kahn acknowledge his contribution, and later Metcalfe would feel somewhat slighted for not insisting on co-authorship. PUP likely influenced the architecture of the modern internet again in the 1970s when advocated for the decision to split the protocol into TCP and IP, so that the complex TCP protocol would not need to be processed at gateways between networks. IP (Internet Protocol) was a simplified version of the addressing protocol, stripped of all the complex TCP logic that ensured the delivery of every bit. The Xerox networking protocol â by then known as Xerox Network Systems (XNS) â had already reached a similar division.
Another source of influence on early internet protocols emerged in Europe, specifically within a network developed in the early 1970s as a result of the implementation of Plan Calcul â a program initiated by to foster its own French computing industry. De Gaulle was long concerned about the growing political, commercial, financial, and cultural dominance of the USA in Western Europe. He aimed to re-establish France as an independent global leader, rather than a pawn in the Cold War between the USA and the USSR. Regarding the computer industry in the 1960s, two particularly strong threats to this independence emerged. Firstly, the USA refused to grant export licenses for its most powerful computers, which France wanted to use in the development of its own atomic bombs. Secondly, the American company General Electric became the main owner of the only French computer manufacturer Compagnie des Machines Bull â soon after which it closed several key product lines of Bull (the company was founded in 1919 by a Norwegian named Bull to produce machines that worked with punch cards â just like IBM. It moved to France in the 1930s, after the founder's death). This led to the birth of Plan Calcul, aimed at ensuring that France could independently secure its computing power.
To oversee the implementation of Plan Calcul, de Gaulle established the dĂ©lĂ©gation Ă lâinformatique (something like the 'delegation for computer science'), reporting directly to his prime minister. In early 1971, this delegation appointed engineer Louis Pouzin to lead the creation of a French version of ARPANET. The delegation believed that packet-switching networks would play a critically important role in computing in the coming years, and thus technical knowledge in this area would be essential for the success of Plan Calcul.
Pouzin at a conference in 1976
Pouzin, a graduate of the Ăcole Polytechnique in Paris, France's premier engineering school, initially worked for a French telephone equipment manufacturer before joining Bull. There, he convinced his employers that they needed to learn more about the advanced developments in the USA. So, while at Bull, he helped develop a compatible time-sharing system (Compatible Time-Sharing System, CTSS) at MIT for two and a half years, from 1963 to 1965. This experience made him the leading expert on interactive time-sharing computing throughout France â and likely all of Europe.
Cyclades Network Architecture
Puzena named the network he was asked to create Cyclades, in honor of the group of Greek islands in the Aegean Sea. According to the name, each computer in this network was essentially a separate island. The main contribution of Cyclades to networking technology was the concept of â the simplest form of packet-switched communication. The idea consisted of two complementary parts:
- Datagrams are independent: unlike data in a phone call or ARPANET message, each datagram can be processed independently. It does not rely on previous messages, their order, or a connection establishment protocol (such as dialing a phone number).
- Datagrams are transmitted from host to host â the entire responsibility for reliably delivering the message lies with the sender and the receiver, not the network, which in this case serves simply as a 'pipe'.
The concept of a datagram seemed heretical to Puzena's colleagues at the French organization dealing with mail, telephone, and telegraph (PTT), who in the 1970s were creating their own network based on connections similar to those of telephones and terminal-to-computer connections (rather than computer-to-computer). This was overseen by another Polytechnic School graduate, RĂ©mi Depre. The idea of abandoning reliability of transmissions within the network was repulsive to PTT, as decades of experience led it to make telephone and telegraph as reliable as possible. Moreover, from an economic and political viewpoint, transferring control over all applications and services to host computers on the periphery of the network threatened to turn PTT into something entirely non-unique and replaceable. However, nothing strengthens an opinion more than firm opposition to it, so the concept of from PTT only helped Puzena to be more convinced of the correctness of his datagram â an approach to designing protocols that work for communication from one host to another.
Puzan and his colleagues from the Cyclades project actively participated in INWG and various conferences where ideas underlying TCP were discussed, and they did not hesitate to express their opinions on how the network or networks should operate. Like Melkaf, Puzan and his colleague Hubert Zimmerman deserve mention in TCP work from 1974, and at least one other colleague, engineer Gérard Le Lann, also helped Cerf refine the protocols. Later, Cerf recalled that " using the sliding window method for TCP was taken directly from discussions on the topic with Puzan and his people⊠I remember how Bob Metcalfe, Le Lann, and I are lying on a huge piece of paper in the floor of my living room in Palo Alto, trying to sketch state diagrams for these protocols.
The sliding window refers to the way TCP manages the flow of data between the sender and the receiver. The current window consists of all the packets in the outgoing data stream that the sender can actively send. The right edge of the window shifts to the right when the receiver indicates that space has been freed in the buffer, and the left edge shifts to the right when the receiver reports the receipt of previous packets.
The concept of the diagram perfectly aligned with the behavior of broadcast networks like Ethernet and ALOHANET, which inevitably send their messages into a noisy and indifferent ether (in contrast to the more telephone-like ARPANET, which required sequential message delivery between IMPs over a reliable line from AT&T to function properly). It made sense to tailor the protocols for intra-network transmission to the least reliable networks rather than their more complex relatives, and this is precisely what the TCP protocol by Kahn and Cerf did.
I could continue in this vein, describing the British role in the development of early stages of inter-network interactions, but it is worth avoiding excessive details to not miss the main point â the two names most closely associated with the invention of the internet were not the only ones of significance.
TCP conquers all
What happened to these early ideas of intercontinental cooperation? Why are Cerf and Kahn celebrated everywhere as the fathers of the internet, while Puzhen and Zimmermann remain unheard of? To understand this, we must first delve into the procedural details of the early years of INWG.
Following the spirit of the ARPA network working group and its Requests for Comment (RFC), INWG created its own system of 'common notes'. In this practice, after about a year of collaboration, Kahn and Cerf presented a preliminary version of TCP for INWG in note No. 39 in September 1973. This was essentially the same document they published in IEEE Transactions the following spring. In April 1974, the Cyclades team led by Hubert Zimmermann and Michel Eli published a competing proposal, INWG 61. The difference lay in varying views on distinct engineering compromises, primarily regarding how packets are split and reassembled when crossing networks with smaller packet sizes.
The divide was minimal; however, the need to come to some agreement gained an unexpected urgency due to plans to review network standards announced by the ComitĂ© Consultatif International TĂ©lĂ©phonique et TĂ©lĂ©graphique () [International Consultative Committee for Telephony and Telegraphy]. CCITT, a subdivision of the , responsible for standardization, operated on a four-year cycle of plenary sessions. Proposals to be considered at the 1976 meeting had to be submitted by the fall of 1975, and no changes could be made between that date and 1980. Frantic meetings within INWG led to a final vote, resulting in the new protocol described by representatives from the most important organizations in computer networking worldwide â Cerf from ARPANET, Zimmermann from Cyclades, Roger Scantlebury from the British National Physical Laboratory, and Alex Mackenzie from BBN. The new proposal, INWG 96, chose something in between 39 and 61, seemingly establishing the direction for the development of internetworking in the foreseeable future.
But in reality, the compromise served as the last breath of international cooperation in the field of networking, and this fact was preceded by the ominous absence of Bob Kahn at the INWG vote regarding the new proposal. It turned out that the voting result did not fit within the deadlines set by CCITT; furthermore, Cerf worsened the situation by sending a letter to CCITT describing that this proposal lacked full consensus in INWG. However, any proposal from INWG most likely would not have been accepted, as the dominant telecom leaders in CCITT were not interested in datagram-supported networks conceived by computer researchers. They wanted full control over traffic in the network, rather than delegating that power to local computers, which they could not supervise. They completely ignored the issue of interoperability and agreed to adopt a virtual connection protocol for a separate network, named .
The irony is that the X.25 protocol was supported by Kahn's former boss, Larry Roberts. He was once a leader in advanced networking research, but his new interests as a business leader brought him to CCITT to sanction the protocols already used by his company, Telenet.
Europeans, mainly led by Zimmerman, made another attempt by turning to another standards organization where the dominance of telecom leadership was not as strong â the International Organization for Standardization, . The resulting Open Systems Interconnection standard () had some advantages over TCP/IP. For instance, it did not have such a limited hierarchical addressing system as IP, whose limitations required several cheap hacks to cope with the explosive growth of the internet in the 1990s (in the 2010s networks are finally starting to transition to IP protocol, which was intended to address address space limitation issues). However, this process was prolonged for many reasons, dragging on indefinitely without resulting in a functional software solution. In particular, the ISO procedures, which were well-suited for approving established technical practices, were not applicable to emerging technologies. When the TCP/IP-based internet began to evolve in the 1990s, OSI lost its relevance.
Let's move from the battle for standards to the practical aspects of building networks on the ground. Europeans diligently took on the implementation of INWG 96 to unite Cyclades and the national physical laboratory in the creation of a European information network. But Kan and other leaders of the ARPA internet project were not about to derail the TCP train for international cooperation. Kan had already allocated funds for the implementation of TCP in ARPANET and PRNET and did not want to start from scratch. Cerf tried to promote in the US support for the compromise he had developed for INWG, but ultimately gave up. He also decided to step away from the stresses of being an adjunct professor and, following Kan's example, became a program manager at ARPA, moving away from active participation in INWG.
So why did the Europeans' desire to establish a unified front and official international standard yield so little? Mainly, it was due to differing positions among the heads of American and European telecoms. Europeans faced constant pressure on the datagram model from their postal and telecom leaders (PTT), who operated as administrative departments of their respective national governments. This gave them more motivation to find consensus in official standard-setting processes. The rapid decline of Cyclades, which lost political interest in 1975 and all funding by 1978, provides material for studying the power of PTT. Puzin was blamed for its demise by the administration. . d'Estaing came to power in 1974 and assembled a government from representatives of the National School of Administration (), despised by Pouzin: if the Polytechnic School can be compared to MIT, then the ENA can be likened to Harvard Business School. The d'Estens administration built its IT policy on the idea of 'national champions', and such a computer network required the support of PTT. The Cyclades project would never have received that support; instead, Pouzin's rival, Despre, led the creation of a network with virtual connections based on X.25 called Transpac.
In the US, everything was different. AT&T had no political influence like its foreign counterparts; it was not part of the US administration. On the contrary, the government was heavily restricting and weakening the company at that time, forbidding it from interfering in the development of computer networks and services, and soon it was completely dismantled. ARPA was able to freely develop its internet program under the protective umbrella of the powerful Department of Defense, without any political pressure. It funded the implementation of TCP on various computers and used its influence to force all hosts in ARPANET to switch to the new protocol in 1983. Thus, the most influential computer network in the world, with many nodes from the most significant computing organizations globally, became the breeding ground for TCP/IP.
Thus, TCP/IP became the cornerstone of the internet, and not just the internet, thanks to the relative political and financial freedom of ARPA compared to any other organization involved in computer networking. Despite OSI, ARPA became the dog wagging the outraged tail of the networking research community. From the perspective of 1974, one could note the multitude of lines of influence leading to Cerf's and Kahn's work on TCP, and many potential avenues for international cooperation that could arise from it. However, by 1995, all roads led to a single key moment, a single American organization, and two celebrated names.
Further reading
- Janet Abbate, Inventing the Internet (1999)
- John Day, âThe Clamor Outside as INWG Debated,â IEEE Annals of the History of Computing (2016)
- Andrew L. Russell, Open Standards and the Digital Age (2014)
- Andrew L. Russell and ValĂ©rie Schafer, âIn the Shadow of ARPANET and Internet: Louis Pouzin and the Cyclades Network in the 1970s,â Technology and Culture (2014)
Source: habr.com
