Development of the Internet and the World Wide Web
The recent growth of the Internet and the World Wide Web makes it appear that the world is witnessing the arrival of a completely new technology. In fact, the Web—now considered to be a major driver of the way society accesses and views information—is the result of numerous projects in computer networking, mostly funded by the federal government, carried out over the last 40 years. The projects produced communications protocols that define the format of network messages, prototype networks, and application programs such as browsers. This research capitalized on the ubiquity of the nation's telephone network, which provided the underlying physical infrastructure upon which the Internet was built.
This chapter traces the development of the Internet,1 one aspect of the broader field of data networking. The chapter is not intended to be comprehensive; rather, it focuses on the federal role in both funding research and supporting the deployment of networking infrastructure. This history is divided into four distinct periods. Before 1970, individual researchers developed the underlying technologies, including queuing theory, packet switching, and routing. During the 1970s, experimental networks, notably the ARPANET, were constructed. These networks were primarily research tools, not service providers. Most were federally funded, because, with a few exceptions, industry had not yet realized the potential of the technology. During the 1980s, networks were widely deployed, initially to support scientific research. As their potential to improve personal communications and collaboration became apparent, additional academic disciplines and industry began to use the technol-
ogy. In this era, the National Science Foundation (NSF) was the major supporter of networking, primarily through the NSFNET, which evolved into the Internet. Most recently, in the early 1990s, the invention of the Web made it much easier for users to publish and access information, thereby setting off the rapid growth of the Internet. The final section of the chapter summarizes the lessons to be learned from history.
By focusing on the Internet, this chapter does not address the full scope of computer networking activities that were under way between 1960 and 1995. It specifically ignores other networking activities of a more proprietary nature. In the mid-1980s, for example, hundreds of thousands of workers at IBM were using electronic networks (such as the VNET) for worldwide e-mail and file transfers; banks were performing electronic funds transfer; Compuserve had a worldwide network; Digital Equipment Corporation (DEC) had value-added networking services; and a VNET-based academic network known as BITNET had been established. These were proprietary systems that, for the most part, owed little to academic research, and indeed were to a large extent invisible to the academic computer networking community. By the late 1980s, IBM's proprietary SNA data networking business unit already had several billions of dollars of annual revenue for networking hardware, software, and services. The success of such networks in many ways limited the interest of companies like IBM and Compuserve in the Internet. The success of the Internet can therefore, in many ways, be seen as the success of an open system and open architecture in the face of proprietary competition.
Early Steps: 1960-1970
Approximately 15 years after the first computers became operational, researchers began to realize that an interconnected network of computers could provide services that transcended the capabilities of a single system. At this time, computers were becoming increasingly powerful, and a number of scientists were beginning to consider applications that went far beyond simple numerical calculation. Perhaps the most compelling early description of these opportunities was presented by J.C.R. Licklider (1960), who argued that, within a few years, computers would become sufficiently powerful to cooperate with humans in solving scientific and technical problems. Licklider, a psychologist at the Massachusetts Institute of Technology (MIT), would begin realizing his vision when he became director of the Information Processing Techniques Office (IPTO) at the Advanced Research Projects Agency (ARPA) in 1962. Licklider remained at ARPA until 1964 (and returned for a second tour in 1974-1975), and he convinced his successors, Ivan Sutherland and Robert Taylor, of the importance of attacking difficult, long-term problems.
Taylor, who became IPTO director in 1966, worried about the duplication of expensive computing resources at the various sites with ARPA contracts. He proposed a networking experiment in which users at one site accessed computers at another site, and he co-authored, with Licklider, a paper describing both how this might be done and some of the potential consequences (Licklider and Taylor, 1968). Taylor was a psychologist, not a computer scientist, and so he recruited Larry Roberts of MIT's Lincoln Laboratory to move to ARPA and oversee the development of the new network. As a result of these efforts, ARPA became the primary supporter of projects in networking during this period.
In contrast to the NSF, which awarded grants to individual researchers, ARPA issued research contracts. The IPTO program managers, typically recruited from academia for 2-year tours, had considerable latitude in defining projects and identifying academic and industrial groups to carry them out. In many cases, they worked closely with the researchers they sponsored, providing intellectual leadership as well as financial support. A strength of the ARPA style was that it not only produced artifacts that furthered its missions but also built and trained a community of researchers. In addition to holding regular meetings of principal investigators, Taylor started the "ARPA games," meetings that brought together the graduate students involved in programs. This innovation helped build the community that would lead the expansion of the field and growth of the Internet during the 1980s.
During the 1960s, a number of researchers began to investigate the technologies that would form the basis for computer networking. Most of this early networking research concentrated on packet switching, a technique of breaking up a conversation into small, independent units, each of which carries the address of its destination and is routed through the network independently. Specialized computers at the branching points in the network can vary the route taken by packets on a moment-to-moment basis in response to network congestion or link failure.
One of the earliest pioneers of packet switching was Paul Baran of the RAND Corporation, who was interested in methods of organizing networks to withstand nuclear attack. (His research interest is the likely source of a widespread myth concerning the ARPANET's original purpose [Hafner and Lyon, 1996]). Baran proposed a richly interconnected set of network nodes, with no centralized control system—both properties of today's Internet. Similar work was under way in the United Kingdom, where Donald Davies and Roger Scantlebury of the National Physical Laboratory (NPL) coined the term "packet."
Of course, the United States already had an extensive communications network, the public switched telephone network (PSTN), in which digital switches and transmission lines were deployed as early as 1962.
But the telephone network did not figure prominently in early computer networking. Computer scientists working to interconnect their systems spoke a different language than did the engineers and scientists working in traditional voice telecommunications. They read different journals, attended different conferences, and used different terminology. Moreover, data traffic was (and is) substantially different from voice traffic. In the PSTN, a continuous connection, or circuit, is set up at the beginning of a call and maintained for the duration. Computers, on the other hand, communicate in bursts, and unless a number of "calls" can be combined on a single transmission path, line and switching capacity is wasted. Telecommunications engineers were primarily interested in improving the voice network and were skeptical of alternative technologies. As a result, although telephone lines were used to provide point-to-point communication in the ARPANET, the switching infrastructure of the PSTN was not used. According to Taylor, some Bell Laboratories engineers stated flatly in 1967 that "packet switching wouldn't work."2
At the first Association for Computing Machinery (ACM) Symposium on Operating System Principles in 1967, Lawrence Roberts, then an IPTO program manager, presented an initial design for the packet-switched network that was to become the ARPANET (Davies et al., 1967). In addition, Roger Scantlebury presented the NPL work (Roberts, 1967), citing Baran's earlier RAND report. The reaction was positive, and Roberts issued a request for quotation (RFQ) for the construction of a four-node network.
From the more than 100 respondents to the RFQ, Roberts selected Bolt, Beranek, and Newman (BBN) of Cambridge, Massachusetts; familiar names such as IBM Corporation and Control Data Corporation chose not to bid. The contract to produce the hardware and software was issued in December 1968. The BBN group was led by Frank Heart, and many of the scientists and engineers who would make major contributions to networking in future years participated. Robert Kahn, who with Vinton Cerf would later develop the Transmission Control Protocol/Internet Protocol (TCP/IP) suite used to control the transmission of packets in the network, helped develop the network architecture. The network hardware consisted of a rugged military version of a Honeywell Corporation minicomputer that connected a site's computers to the communication lines. These interface message processors (IMPs)—each the size of a large refrigerator and painted battleship gray—were highly sought after by DARPA-sponsored researchers, who viewed possession of an IMP as evidence they had joined the inner circle of networking research.
The first ARPANET node was installed in September 1969 at Leonard Kleinrock's Network Measurement Center at the University of California at Los Angeles (UCLA). Kleinrock (1964) had published some of the
earliest theoretical work on packet switching, and so this site was an appropriate choice. The second node was installed a month later at Stanford Research Institute (SRI) in Menlo Park, California, using Douglas Engelbart's On Line System (known as NLS) as the host. SRI also operated the Network Information Center (NIC), which maintained operational and standards information for the network. Two more nodes were soon installed at the University of California at Santa Barbara, where Glen Culler and Burton Fried had developed an interactive system for mathematics education, and the University of Utah, which had one of the first computer graphics groups.
Initially, the ARPANET was primarily a vehicle for experimentation rather than a service, because the protocols for host-to-host communication were still being developed. The first such protocol, the Network Control Protocol (NCP), was completed by the Network Working Group (NWG) led by Stephen Crocker in December 1970 and remained in use until 1983, when it was replaced by TCP/IP.
Expansion of the Arpanet: 1970-1980
Initially conceived as a means of sharing expensive computing resources among ARPA research contractors, the ARPANET evolved in a number of unanticipated directions during the 1970s. Although a few experiments in resource sharing were carried out, and the Telnet protocol was developed to allow a user on one machine to log onto another machine over the network, other applications became more popular.
The first of these applications was enabled by the File Transfer Protocol (FTP), developed in 1971 by a group led by Abhay Bhushan of MIT (Bhushan, 1972). This protocol enabled a user on one system to connect to another system for the purpose of either sending or retrieving a particular file. The concept of an anonymous user was quickly added, with constrained access privileges, to allow users to connect to a system and browse the available files. Using Telnet, a user could read the remote files but could not do anything with them. With FTP, users could now move files to their own machines and work with them as local files. This capability spawned several new areas of activity, including distributed client-server computing and network-connected file systems.
Occasionally in computing, a "killer application" appears that becomes far more popular than its developers expected. When personal computers (PCs) became available in the 1980s, the spreadsheet (initially VisiCalc) was the application that accelerated the adoption of the new hardware by businesses. For the newly minted ARPANET, the killer application was electronic mail, or e-mail. The first e-mail program was developed in 1972 by Ray Tomlinson of BBN. Tomlinson had built an
earlier e-mail system for communication between users on BBN's Tenex time-sharing system, and it was a simple exercise to modify this system to work over the network. By combining the immediacy of the telephone with the precision of written communication, e-mail became an instant hit. Tomlinson's syntax (user@domain) remains in use today.
Telnet, FTP, and e-mail were examples of the leverage that research typically provided in early network development. As each new capability was added, the efficiency and speed with which knowledge could be disseminated improved. E-mail and FTP made it possible for geographically distributed researchers to collaborate and share results much more effectively. These programs were also among the first networking applications that were valuable not only to computer scientists, but also to scholars in other disciplines.
From Arpanet to Internet
Although the ARPANET was ARPA's largest networking effort, it was by no means the only one. The agency also supported research on terrestrial packet radio and packet satellite networks. In 1973, Robert Kahn and Vinton Cerf began to consider ways to interconnect these networks, which had quite different bandwidth, delay, and error properties than did the telephone lines of the ARPANET. The result was TCP/IP, first described in 1973 at an International Network Working Group meeting in England. Unlike NCP, which enabled the hosts of a single network to communicate, TCP/IP was designed to interconnect multiple networks to form an Internet. This protocol suite defined the packet format and a flow-control and error-recovery mechanism to allow the hosts to recover gracefully from network errors. It also specified an addressing mechanism that could support an Internet comprising up to 4 billion hosts.
The work necessary to transform TCP/IP from a concept into a useful system was performed under ARPA contract by groups at Stanford University, BBN, and University College London. Although TCP/IP has evolved over the years, it is still in use today as the Internet's basic packet transport protocol.
By 1975, the ARPANET had grown from its original four nodes to nearly 100 nodes. Around this time, two phenomena—the development of local area networks (LANs) and the integration of networking into operating systems—contributed to a rapid increase in the size of the network.
Local Area Networks
While ARPANET researchers were experimenting with dedicated telephone lines for packet transmission, researchers at the University of
Hawaii, led by Norman Abramson, were trying a different approach, also with ARPA funding. Like the ARPANET group, they wanted to provide remote access to their main computer system, but instead of a network of telephone lines, they used a shared radio network. It was shared in the sense that all stations used the same channel to reach the central station. This approach had a potential drawback: if two stations attempted to transmit at the same time, then their transmissions would interfere with each other, and neither one would be received. But such interruptions were unlikely because the data were typed on keyboards, which sent very short pulses to the computer, leaving ample time between pulses during which the channel was clear to receive keystrokes from a different user.
Abramson's system, known as Aloha, generated considerable interest in using a shared transmission medium, and several projects were initiated to build on the idea. Two of the best-known projects were the Atlantic Packet Satellite Experiment and Ethernet. The packet satellite network demonstrated that the protocols developed in Aloha for handling contention between simultaneous users, combined with more traditional reservation schemes, resulted in efficient use of the available bandwidth. However, the long latency inherent in satellite communications limited the usefulness of this approach.
Ethernet, developed by a group led by Robert Metcalfe at Xerox Corporation's Palo Alto Research Center (PARC), is one of the few examples of a networking technology that was not directly funded by the government. This experiment demonstrated that using coaxial cable as a shared medium resulted in an efficient network. Unlike the Aloha system, in which transmitters could not receive any signals, Ethernet stations could detect that collisions had occurred, stop transmitting immediately, and retry a short time later (at random). This approach improved the efficiency of the Aloha technique and made it practical for actual use. Shared-media LANs became the dominant form of computer-to-computer communication within a building or local area, although variations from IBM (Token Ring) and others also captured part of this emerging market.
Ethernet was initially used to connect a network of approximately 100 of PARC's Alto PCs, using the center's time-sharing system as a gateway to the ARPANET. Initially, many believed that the small size and limited performance of PCs would preclude their use as network hosts, but, with DARPA funding, David Clark's group at MIT, which had received several Altos from PARC, built an efficient TCP implementation for that system and, later, for the IBM PC. The proliferation of PCs connected by LANs in the 1980s dramatically increased the size of the Internet.
Until the 1970s, academic computer science research groups used a variety of computers and operating systems, many of them constructed by the researchers themselves. Most were time-sharing systems that supported a number of simultaneous users. By 1970, many groups had settled on the Digital Equipment Corporation (DEC) PDP-10 computer and the Tenex operating system developed at BBN. This standardization enabled researchers at different sites to share software, including networking software.
By the late 1970s, the Unix operating system, originally developed at Bell Labs, had become the system of choice for researchers, because it ran on DEC's inexpensive (relative to other systems) VAX line of computers. During the late 1970s and early 1980s, an ARPA-funded project at the University of California at Berkeley (UC-Berkeley) produced a version of Unix (the Berkeley System Distribution, or BSD) that included tightly integrated networking capabilities. The BSD was rapidly adopted by the research community because the availability of source code made it a useful experimental tool. In addition, it ran on both VAX machines and the personal workstations provided by the fledgling Sun Microsystems, Inc., several of whose founders came from the Berkeley group. The TCP/IP suite was now available on most of the computing platforms used by the research community.
Standards and Management
Unlike the various telecommunications networks, the Internet has no owner. It is a federation of commercial service providers, local educational networks, and private corporate networks, exchanging packets using TCP/IP and other, more specialized protocols. To become part of the Internet, a user need only connect a computer to a port on a service provider's router, obtain an IP address, and begin communicating. To add an entire network to the Internet is a bit trickier, but not extraordinarily so, as demonstrated by the tens of thousands of networks with tens of millions of hosts that constitute the Internet today.
The primary technical problem in the Internet is the standardization of its protocols. Today, this is accomplished by the Internet Engineering Task Force (IETF), a voluntary group interested in maintaining and expanding the scope of the Internet. Although this group has undergone many changes in name and makeup over the years, it traces its roots directly to Stephen Crocker's NWG, which defined the first ARPANET protocol in 1969. The NWG defined the system of requests for comments (RFCs) that are still used to specify protocols and discuss other engineer-
ing issues. Today's RFCs are still formatted as they were in 1969, eschewing the decorative fonts and styles that pervade today's Web.
Joining the IETF is a simple matter of asking to be placed on its mailing list, attending thrice-yearly meetings, and participating in the work. This grassroots group is far less formal than organizations such as the International Telecommunications Union, which defines telephony standards through the work of members who are essentially representatives of various governments. The open approach to Internet standards reflects the academic roots of the network.
Closing the Decade
The 1970s were a time of intensive research in networking. Much of the technology used today was developed during this period. Several networks other than ARPANET were assembled, primarily for use by computer scientists in support of their own research. Most of the work was funded by ARPA, although the NSF provided educational support for many researchers and was beginning to consider establishing a large-scale academic network.
During this period, ARPA pursued high-risk research with the potential for high payoffs. Its work was largely ignored by AT&T, and the major computer companies, notably IBM and DEC, began to offer proprietary networking solutions that competed with, rather than applied, the ARPA-developed technologies.3 Yet the technologies developed under ARPA contract ultimately resulted in today's Internet. It is debatable whether a more risk-averse organization lacking the hands-on program management style of ARPA could have produced the same result.
Operation of the ARPANET was transferred to the Defense Communication Agency in 1975. By the end of the decade, the ARPANET had matured sufficiently to provide services. It remained in operation until 1989, when it was superseded by subsequent networks. The stage was now set for the Internet, which was first used by scientists, then by academics in many disciplines, and finally by the world at large.
The NSFNET Years: 1980-1990
During the late 1970s, several networks were constructed to serve the needs of particular research communities. These networks—typically funded by the federal agency that was the primary supporter of the research area—included MFENet, which the Department of Energy established to give its magnetic fusion energy researchers access to supercomputers, and NASA's Space Physics Analysis Network (SPAN). The NSF began supporting network infrastructure with the establishment
of CSNET, which was intended to link university computer science departments with the ARPANET. The CSNET had one notable property that the ARPANET lacked: it was open to all computer science researchers, whereas only ARPA contractors could use the ARPANET. An NSF grant to plan the CSNET was issued to Larry Landweber at the University of Wisconsin in 1980.
The CSNET was used throughout the 1980s, but as it and other regional networks began to demonstrate their usefulness, the NSF launched a much more ambitious effort, the NSFNET. From the start, the NSFNET was designed to be a network of networks—an ''internet''—with a high-speed backbone connecting NSF's five supercomputer centers and the National Center for Atmospheric Research. To oversee the new network, the NSF hired Dennis Jennings from Trinity College, Dublin. In the early 1980s, Jennings had been responsible for the Irish Higher Education Authority network (HEANet), and so he was well-qualified for the task. One of Jennings' first decisions was to select TCP/IP as the primary protocol suite for the NFSNET.
Because the NSFNET was to be an internet (the beginning of today's Internet), specialized computers called routers were needed to pass traffic between networks at the points where the networks met. Today, routers are the primary products of multibillion-dollar companies (e.g., Cisco Systems Incorporated, Bay Networks), but in 1985, few commercial products were available. The NSF chose the "Fuzzball" router designed by David Mills at the University of Delaware (Mills, 1988). Working with ARPA support, Mills improved the protocols used by the routers to communicate the network topology among themselves, a critical function in a large-scale network.
Another technology required for the rapidly growing Internet was the Domain Name Service (DNS). Developed by Paul Mockapetris at the University of Southern California's Information Sciences Institute, the DNS provides for hierarchical naming of hosts. An administrative entity, such as a university department, can assign host names as it wishes. It also has a domain name, issued by the higher-level authority of which it is a part. (Thus, a host named xyz in the computer science department at UC-Berkeley would be named xyz.cs.berkeley.edu. ) Servers located throughout the Internet provide translation between the host names used by human users and the IP addresses used by the Internet protocols. The name-distribution scheme has allowed the Internet to grow much more rapidly than would be possible with centralized administration.
Jennings left the NSF in 1986. He was succeeded by Stephen Wolff, who oversaw the deployment and growth of the NSFNET. During Wolff's tenure, the speed of the backbone, originally 56 kilobits per second, was increased 1,000-fold, and a large number of academic and regional net-
works were connected to the NSFNET. The NSF also began to expand the reach of the NSFNET beyond its supercomputing centers through its Connections program, which targeted the research and education community. In response to the Connections solicitation, the NSF received innovative proposals from what would become two of the major regional networks: SURANET and NYSERNET. These groups proposed to develop regional networks with a single connection to the NSFNET, instead of connecting each institution independently.
Hence, the NSFNET evolved into a three-tiered structure in which individual institutions connected to regional networks that were, in turn, connected to the backbone of the NSFNET. The NSF agreed to provide seed funding for connecting regional networks to the NSFNET, with the expectation that, as a critical mass was reached, the private sector would take over the management and operating costs of the Internet. This decision helped guide the Internet toward self-sufficiency and eventual commercialization (Computer Science and Telecommunications Board, 1994).
As the NSFNET expanded, opportunities for privatization grew. Wolff saw that commercial interests had to participate and provide financial support if the network were to continue to expand and evolve into a large, single internet. The NSF had already (in 1987) contracted with Merit Computer Network Incorporated at the University of Michigan to manage the backbone. Merit later formed a consortium with IBM and MCI Communications Corporation called Advanced Network and Services (ANS) to oversee upgrades to the NSFNET. Instead of reworking the existing backbone, ANS added a new, privately owned backbone for commercial services in 1991.4
Emergence of the Web: 1990 to the Present
By the early 1990s, the Internet was international in scope, and its operation had largely been transferred from the NSF to commercial providers. Public access to the Internet expanded rapidly thanks to the ubiquitous nature of the analog telephone network and the availability of modems for connecting computers to this network. Digital transmission became possible throughout the telephone network with the deployment of optical fiber, and the telephone companies leased their broadband digital facilities for connecting routers and regional networks to the developers of the computer network. In April 1995, all commercialization restrictions on the Internet were lifted. Although still primarily used by academics and businesses, the Internet was growing, with the number of hosts reaching 250,000. Then the invention of the Web catapulted the Internet to mass popularity almost overnight.
The idea for the Web was simple: provide a common format for
documents stored on server computers, and give each document a unique name that can be used by a browser program to locate and retrieve the document. Because the unique names (called universal resource locators, or URLs) are long, including the DNS name of the host on which they are stored, URLs would be represented as shorter hypertext links in other documents. When the user of a browser clicks a mouse on a link, the browser retrieves and displays the document named by the URL.
This idea was implemented by Timothy Berners-Lee and Robert Cailliau at CERN, the high-energy physics laboratory in Geneva, Switzerland, funded by the governments of participating European nations. Berners-Lee and Cailliau proposed to develop a system of links between different sources of information. Certain parts of a file would be made into nodes, which, when called up, would link the user to other, related files. The pair devised a document format called HYpertext Markup Language (HTML), a variant of the Standard Generalized Markup Language used in the publishing industry since the 1950s. It was released at CERN in May 1991. In July 1992, a new Internet protocol, the Hypertext Transfer Protocol (HTTP), was introduced to improve the efficiency of document retrieval. Although the Web was originally intended to improve communications within the physics community at CERN, it—like e-mail 20 years earlier—rapidly became the new killer application for the Internet.
The idea of hypertext was not new. One of the first demonstrations of a hypertext system, in which a user could click a mouse on a highlighted word in a document and immediately access a different part of the document (or, in fact, another document entirely), occurred at the 1967 Fall Joint Computer Conference in San Francisco. At this conference, Douglas Engelbart of SRI gave a stunning demonstration of his NLS (Engelbart, 1986), which provided many of the capabilities of today's Web browsers, albeit limited to a single computer. Engelbart's Augment project was supported by funding from NASA and ARPA. Engelbart was awarded the Association for Computing Machinery's 1997 A. M. Turing Award for this work. Although it never became commercially successful, the mouse-driven user interface inspired researchers at Xerox PARC, who were developing personal computing technology.
Widespread use of the Web, which now accounts for the largest volume of Internet traffic, was accelerated by the development in 1993 of the Mosaic graphical browser. This innovation, by Marc Andreessen at the NSF-funded National Center for Supercomputer Applications, enabled the use of hyperlinks to video, audio, and graphics, as well as text. More important, it provided an effective interface that allowed users to point-and-click on a menu or fill in a blank to search for information.
The development of the Internet and the World Wide Web has had a tremendous impact on the U.S. economy and society more broadly. By
January 1998, almost 30 million host computers were connected to the Internet (Zakon, 1998), and more than 58 million users in the United States and Canada were estimated to be online (Nielsen Media Research, 1997). Numerous companies now sell Internet products worth billions of dollars. Cisco Systems, a leader in network routing technology, for example, reported sales of $8.5 billion in 1998. Netscape Communications Corporation, which commercialized the Mosaic browser, had sales exceeding $530 million in 1997.5 Microsoft Corporation also entered the market for Web browsers and now competes head-to-head with Netscape. A multitude of other companies offer hardware and software for Internet based systems.
The Internet has also paved the way for a host of services. Companies like Yahoo! and InfoSeek provide portals to the Internet and have attracted considerable attention from Wall Street investors. Other companies, like Amazon.com and Barnes & Noble, have established online stores. Amazon had online sales of almost $150 million for books in 1997.6 Electronic commerce, more broadly, is taking hold in many types of organizations, from PC manufacturers to retailers to travel agencies. Although estimates of the value of these services vary widely, they all reflect a growing sector of the economy that is wholly dependent on the Internet. Internet retailing could reach $7 billion by the year 2000, and online sales of travel services are expected to approach $8 billion around the turn of the century. Forrester Research estimates that businesses will buy and sell $327 billion worth of goods over the Internet by the year 2002 (Blane, 1997).
The Web has been likened to the world's largest library—with the books piled in the middle of the floor. Search engines, which are programs that follow the Web's hypertext links and index the material they discover, have improved the organization somewhat but are difficult to use, frequently deluging the user with irrelevant information. Although developments in computing and networking over the last 40 years have realized some of the potential described by visionaries such as Licklider and Engelbart, the field continues to offer many opportunities for innovation.
Lessons from History
The development of the Internet demonstrates that federal support for research, applied at the right place and right time, can be extremely effective. DARPA's support gave visibility to the work of individual researchers on packet switching and resulted in the development of the first large-scale packet-switched network. Continued support for experimentation led to the development of networking protocols and applications, such as e-mail, that were used on the ARPANET and, subsequently, the Internet.
By bringing together a diverse mix of researchers from different institutions, such federal programs helped the Internet gain widespread acceptance and established it as a dominant mode of internetworking. Government programs such as ARPANET and NSFNET created a large enough base of users to make the Internet more attractive in many applications than proprietary networking systems being offered by a number of vendors. Though a number of companies continue to sell proprietary systems for wide area networking, some of which are based on packet-switched technology, these systems have not achieved the ubiquity of the Internet and are used mainly within private industry.
Research in packet switching evolved in unexpected directions and had unanticipated consequences. It was originally pursued to make more-efficient use of limited computing capabilities and later seen as a means of linking the research and education communities. The most notable result, however, was the Internet, which has dramatically improved communication across society, changing the way people work, play, and shop. Although DARPA and the NSF were successful in creating an expansive packet-switched network to facilitate communication among researchers, it took the invention of the Web and its browsers to make the Internet more broadly accessible and useful to society.
The widespread adoption of Internet technology has created a number of new companies in industries that did not exist 20 years ago, and most companies that did exist 20 years ago are incorporating Internet technology into their business operations. Companies such as Cisco Systems, Netscape Communications, Yahoo!, and Amazon.com are built on Internet technologies and their applications and generate billions of dollars annually in combined sales revenues. Electronic commerce is also maturing into an established means of conducting business.
The complementary missions and operating styles of federal agencies are important to the development and implementation of new technologies. Whereas DARPA supported early research on packet switching and development of the ARPANET, it was not prepared to support an operational network, nor did it expand its network beyond DARPA-supported research institutions. With its charter to support research and education, the NSF both supported an operational network and greatly expanded its reach, effectively building the infrastructure for the Internet.