Origins of the Internet
The first recorded description of the social interactions that could be enabled through networking was a series of memos written by J.C.R. Licklider of MIT in August 1962 discussing his “Galactic Network” concept. He envisioned a globally interconnected set of computers through which everyone could quickly access data and programs from any site. In spirit, the concept was very much like the Internet of today. Licklider was the first head of the computer research program at DARPA,4 starting in October 1962. While at DARPA he convinced his successors at DARPA, Ivan Sutherland, Bob Taylor, and MIT researcher Lawrence G. Roberts, of the importance of this networking concept.
Leonard Kleinrock at MIT published the first paper on packet switching theory in July 1961 and the first book on the subject in 1964. Kleinrock convinced Roberts of the theoretical feasibility of communications using packets rather than circuits, which was a major step along the path towards computer networking. The other key step was to make the computers talk together. To explore this, in 1965 working with Thomas Merrill, Roberts connected the TX-2 computer in Mass. to the Q-32 in California with a low speed dial-up telephone line creating the first (however small) wide-area computer network ever built. The result of this experiment was the realization that the time-shared computers could work well together, running programs and retrieving data as necessary on the remote machine, but that the circuit switched telephone system was totally inadequate for the job. Kleinrock’s conviction of the need for packet switching was confirmed.
In late 1966 Roberts went to DARPA to develop the computer network concept and quickly put together his plan for the “ARPANET”, publishing it in 1967. At the conference where he presented the paper, there was also a paper on a packet network concept from the UK by Donald Davies and Roger Scantlebury of NPL. Scantlebury told Roberts about the NPL work as well as that of Paul Baran and others at RAND. The RAND group had written a paper on packet switching networks for secure voice in the military in 1964. It happened that the work at MIT (1961-1967), at RAND (1962-1965), and at NPL (1964-1967) had all proceeded in parallel without any of the researchers knowing about the other work. The word “packet” was adopted from the work at NPL and the proposed line speed to be used in the ARPANET design was upgraded from 2.4 kbps to 50 kbps. 5
In August 1968, after Roberts and the DARPA funded community had refined the overall structure and specifications for the ARPANET, an RFQ was released by DARPA for the development of one of the key components, the packet switches called Interface Message Processors (IMP’s). The RFQ was won in December 1968 by a group headed by Frank Heart at Bolt Beranek and Newman (BBN). As the BBN team worked on the IMP’s with Bob Kahn playing a major role in the overall ARPANET architectural design, the network topology and economics were designed and optimized by Roberts working with Howard Frank and his team at Network Analysis Corporation, and the network measurement system was prepared by Kleinrock’s team at UCLA. 6
Due to Kleinrock’s early development of packet switching theory and his focus on analysis, design and measurement, his Network Measurement Center at UCLA was selected to be the first node on the ARPANET. All this came together in September 1969 when BBN installed the first IMP at UCLA and the first host computer was connected. Doug Engelbart’s project on “Augmentation of Human Intellect” (which included NLS, an early hypertext system) at Stanford Research Institute (SRI) provided a second node. SRI supported the Network Information Center, led by Elizabeth (Jake) Feinler and including functions such as maintaining tables of host name to address mapping as well as a directory of the RFC’s.
One month later, when SRI was connected to the ARPANET, the first host-to-host message was sent from Kleinrock’s laboratory to SRI. Two more nodes were added at UC Santa Barbara and University of Utah. These last two nodes incorporated application visualization projects, with Glen Culler and Burton Fried at UCSB investigating methods for display of mathematical functions using storage displays to deal with the problem of refresh over the net, and Robert Taylor and Ivan Sutherland at Utah investigating methods of 3-D representations over the net. Thus, by the end of 1969, four host computers were connected together into the initial ARPANET, and the budding Internet was off the ground. Even at this early stage, it should be noted that the networking research incorporated both work on the underlying network and work on how to utilize the network. This tradition continues to this day.
Computers were added quickly to the ARPANET during the following years, and work proceeded on completing a functionally complete Host-to-Host protocol and other network software. In December 1970 the Network Working Group (NWG) working under S. Crocker finished the initial ARPANET Host-to-Host protocol, called the Network Control Protocol (NCP). As the ARPANET sites completed implementing NCP during the period 1971-1972, the network users finally could begin to develop applications.
In October 1972, Kahn organized a large, very successful demonstration of the ARPANET at the International Computer Communication Conference (ICCC). This was the first public demonstration of this new network technology to the public. It was also in 1972 that the initial “hot” application, electronic mail, was introduced. In March Ray Tomlinson at BBN wrote the basic email message send and read software, motivated by the need of the ARPANET developers for an easy coordination mechanism. In July, Roberts expanded its utility by writing the first email utility program to list, selectively read, file, forward, and respond to messages. From there email took off as the largest network application for over a decade. This was a harbinger of the kind of activity we see on the World Wide Web today, namely, the enormous growth of all kinds of “people-to-people” traffic.
The Initial Internetting
The original ARPANET grew into the Internet. Internet was based on the idea that there would be multiple independent networks of rather arbitrary design, beginning with the ARPANET as the pioneering packet switching network, but soon to include packet satellite networks, ground-based packet radio networks and other networks. The Internet as we now know it embodies a key underlying technical idea, namely that of open architecture networking. In this approach, the choice of any individual network technology was not dictated by a particular network architecture but rather could be selected freely by a provider and made to interwork with the other networks through a meta-level “Internetworking Architecture”. Up until that time there was only one general method for federating networks. This was the traditional circuit switching method where networks would interconnect at the circuit level, passing individual bits on a synchronous basis along a portion of an end-to-end circuit between a pair of end locations. Recall that Kleinrock had shown in 1961 that packet switching was a more efficient switching method. Along with packet switching, special purpose interconnection arrangements between networks were another possibility. While there were other limited ways to interconnect different networks, they required that one be used as a component of the other, rather than acting as a peer of the other in offering end-to-end service.
In an open-architecture network, the individual networks may be separately designed and developed and each may have its own unique interface which it may offer to users and/or other providers. including other Internet providers. Each network can be designed in accordance with the specific environment and user requirements of that network. There are generally no constraints on the types of network that can be included or on their geographic scope, although certain pragmatic considerations will dictate what makes sense to offer.
The idea of open-architecture networking was first introduced by Kahn shortly after having arrived at DARPA in 1972. This work was originally part of the packet radio program, but subsequently became a separate program in its own right. At the time, the program was called “Internetting”. Key to making the packet radio system work was a reliable end-end protocol that could maintain effective communication in the face of jamming and other radio interference, or withstand intermittent blackout such as caused by being in a tunnel or blocked by the local terrain. Kahn first contemplated developing a protocol local only to the packet radio network, since that would avoid having to deal with the multitude of different operating systems, and continuing to use NCP.
However, NCP did not have the ability to address networks (and machines) further downstream than a destination IMP on the ARPANET and thus some change to NCP would also be required. (The assumption was that the ARPANET was not changeable in this regard). NCP relied on ARPANET to provide end-to-end reliability. If any packets were lost, the protocol (and presumably any applications it supported) would come to a grinding halt. In this model NCP had no end-end host error control, since the ARPANET was to be the only network in existence and it would be so reliable that no error control would be required on the part of the hosts. Thus, Kahn decided to develop a new version of the protocol which could meet the needs of an open-architecture network environment. This protocol would eventually be called the Transmission Control Protocol/Internet Protocol (TCP/IP). While NCP tended to act like a device driver, the new protocol would be more like a communications protocol.
Four ground rules were critical to Kahn’s early thinking:
- Each distinct network would have to stand on its own and no internal changes could be required to any such network to connect it to the Internet.
- Communications would be on a best effort basis. If a packet didn’t make it to the final destination, it would shortly be retransmitted from the source.
- Black boxes would be used to connect the networks; these would later be called gateways and routers. There would be no information retained by the gateways about the individual flows of packets passing through them, thereby keeping them simple and avoiding complicated adaptation and recovery from various failure modes.
- There would be no global control at the operations level.
Other key issues that needed to be addressed were:
- Algorithms to prevent lost packets from permanently disabling communications and enabling them to be successfully retransmitted from the source.
- Providing for host-to-host “pipelining” so that multiple packets could be enroute from source to destination at the discretion of the participating hosts, if the intermediate networks allowed it.
- Gateway functions to allow it to forward packets appropriately. This included interpreting IP headers for routing, handling interfaces, breaking packets into smaller pieces if necessary, etc.
- The need for end-end checksums, reassembly of packets from fragments and detection of duplicates, if any.
- The need for global addressing
- Techniques for host-to-host flow control.
- Interfacing with the various operating systems
- There were also other concerns, such as implementation efficiency, internetwork performance, but these were secondary considerations at first.
Kahn began work on a communications-oriented set of operating system principles while at BBN and documented some of his early thoughts in an internal BBN memorandum entitled “Communications Principles for Operating Systems“. At this point he realized it would be necessary to learn the implementation details of each operating system to have a chance to embed any new protocols in an efficient way. Thus, in the spring of 1973, after starting the internetting effort, he asked Vint Cerf (then at Stanford) to work with him on the detailed design of the protocol. Cerf had been intimately involved in the original NCP design and development and already had the knowledge about interfacing to existing operating systems. So armed with Kahn’s architectural approach to the communications side and with Cerf’s NCP experience, they teamed up to spell out the details of what became TCP/IP.
The give and take was highly productive and the first written version7 of the resulting approach was distributed at a special meeting of the International Network Working Group (INWG) which had been set up at a conference at Sussex University in September 1973. Cerf had been invited to chair this group and used the occasion to hold a meeting of INWG members who were heavily represented at the Sussex Conference.
Some basic approaches emerged from this collaboration between Kahn and Cerf:
- Communication between two processes would logically consist of a very long stream of bytes (they called them octets). The position of any octet in the stream would be used to identify it.
- Flow control would be done by using sliding windows and acknowledgments (acks). The destination could select when to acknowledge and each ack returned would be cumulative for all packets received to that point.
- It was left open as to exactly how the source and destination would agree on the parameters of the windowing to be used. Defaults were used initially.
- Although Ethernet was under development at Xerox PARC at that time, the proliferation of LANs were not envisioned at the time, much less PCs and workstations. The original model was national level networks like ARPANET of which only a relatively small number were expected to exist. Thus a 32 bit IP address was used of which the first 8 bits signified the network and the remaining 24 bits designated the host on that network. This assumption, that 256 networks would be sufficient for the foreseeable future, was clearly in need of reconsideration when LANs began to appear in the late 1970s.
The original Cerf/Kahn paper on the Internet described one protocol, called TCP, which provided all the transport and forwarding services in the Internet. Kahn had intended that the TCP protocol support a range of transport services, from the totally reliable sequenced delivery of data (virtual circuit model) to a datagram service in which the application made direct use of the underlying network service, which might imply occasional lost, corrupted or reordered packets. However, the initial effort to implement TCP resulted in a version that only allowed for virtual circuits. This model worked fine for file transfer and remote login applications, but some of the early work on advanced network applications, in particular packet voice in the 1970s, made clear that in some cases packet losses should not be corrected by TCP, but should be left to the application to deal with. This led to a reorganization of the original TCP into two protocols, the simple IP which provided only for addressing and forwarding of individual packets, and the separate TCP, which was concerned with service features such as flow control and recovery from lost packets. For those applications that did not want the services of TCP, an alternative called the User Datagram Protocol (UDP) was added in order to provide direct access to the basic service of IP.
A major initial motivation for both the ARPANET and the Internet was resource sharing – for example allowing users on the packet radio networks to access the time sharing systems attached to the ARPANET. Connecting the two together was far more economical that duplicating these very expensive computers. However, while file transfer and remote login (Telnet) were very important applications, electronic mail has probably had the most significant impact of the innovations from that era. Email provided a new model of how people could communicate with each other, and changed the nature of collaboration, first in the building of the Internet itself (as is discussed below) and later for much of society.
There were other applications proposed in the early days of the Internet, including packet based voice communication (the precursor of Internet telephony), various models of file and disk sharing, and early “worm” programs that showed the concept of agents (and, of course, viruses). A key concept of the Internet is that it was not designed for just one application, but as a general infrastructure on which new applications could be conceived, as illustrated later by the emergence of the World Wide Web. It is the general purpose nature of the service provided by TCP and IP that makes this possible.
Concepts Proving the Ideas
DARPA let three contracts to Stanford (Cerf), BBN (Ray Tomlinson) and UCL (Peter Kirstein) to implement TCP/IP (it was simply called TCP in the Cerf/Kahn paper but contained both components). The Stanford team, led by Cerf, produced the detailed specification and within about a year there were three independent implementations of TCP that could interoperate.
This was the beginning of long term experimentation and development to evolve and mature the Internet concepts and technology. Beginning with the first three networks (ARPANET, Packet Radio, and Packet Satellite) and their initial research communities, the experimental environment has grown to incorporate essentially every form of network and a very broad-based research and development community. [REK78] With each expansion has come new challenges.
The early implementations of TCP were done for large time sharing systems such as Tenex and TOPS 20. When desktop computers first appeared, it was thought by some that TCP was too big and complex to run on a personal computer. David Clark and his research group at MIT set out to show that a compact and simple implementation of TCP was possible. They produced an implementation, first for the Xerox Alto (the early personal workstation developed at Xerox PARC) and then for the IBM PC. That implementation was fully interoperable with other TCPs, but was tailored to the application suite and performance objectives of the personal computer, and showed that workstations, as well as large time-sharing systems, could be a part of the Internet. In 1976, Kleinrock published the first book on the ARPANET. It included an emphasis on the complexity of protocols and the pitfalls they often introduce. This book was influential in spreading the lore of packet switching networks to a very wide community.
Widespread development of LANS, PCs and workstations in the 1980s allowed the nascent Internet to flourish. Ethernet technology, developed by Bob Metcalfe at Xerox PARC in 1973, is now probably the dominant network technology in the Internet and PCs and workstations the dominant computers. This change from having a few networks with a modest number of time-shared hosts (the original ARPANET model) to having many networks has resulted in a number of new concepts and changes to the underlying technology. First, it resulted in the definition of three network classes (A, B, and C) to accommodate the range of networks. Class A represented large national scale networks (small number of networks with large numbers of hosts); Class B represented regional scale networks; and Class C represented local area networks (large number of networks with relatively few hosts).
A major shift occurred as a result of the increase in scale of the Internet and its associated management issues. To make it easy for people to use the network, hosts were assigned names, so that it was not necessary to remember the numeric addresses. Originally, there were a fairly limited number of hosts, so it was feasible to maintain a single table of all the hosts and their associated names and addresses. The shift to having a large number of independently managed networks (e.g., LANs) meant that having a single table of hosts was no longer feasible, and the Domain Name System (DNS) was invented by Paul Mockapetris of USC/ISI. The DNS permitted a scalable distributed mechanism for resolving hierarchical host names (e.g. www.acm.org) into an Internet address.
The increase in the size of the Internet also challenged the capabilities of the routers. Originally, there was a single distributed algorithm for routing that was implemented uniformly by all the routers in the Internet. As the number of networks in the Internet exploded, this initial design could not expand as necessary, so it was replaced by a hierarchical model of routing, with an Interior Gateway Protocol (IGP) used inside each region of the Internet, and an Exterior Gateway Protocol (EGP) used to tie the regions together. This design permitted different regions to use a different IGP, so that different requirements for cost, rapid reconfiguration, robustness and scale could be accommodated. Not only the routing algorithm, but the size of the addressing tables, stressed the capacity of the routers. New approaches for address aggregation, in particular classless inter-domain routing (CIDR), have recently been introduced to control the size of router tables.
As the Internet evolved, one of the major challenges was how to propagate the changes to the software, particularly the host software. DARPA supported UC Berkeley to investigate modifications to the Unix operating system, including incorporating TCP/IP developed at BBN. Although Berkeley later rewrote the BBN code to more efficiently fit into the Unix system and kernel, the incorporation of TCP/IP into the Unix BSD system releases proved to be a critical element in dispersion of the protocols to the research community. Much of the CS research community began to use Unix BSD for their day-to-day computing environment. Looking back, the strategy of incorporating Internet protocols into a supported operating system for the research community was one of the key elements in the successful widespread adoption of the Internet.
One of the more interesting challenges was the transition of the ARPANET host protocol from NCP to TCP/IP as of January 1, 1983. This was a “flag-day” style transition, requiring all hosts to convert simultaneously or be left having to communicate via rather ad-hoc mechanisms. This transition was carefully planned within the community over several years before it actually took place and went surprisingly smoothly (but resulted in a distribution of buttons saying “I survived the TCP/IP transition”).
TCP/IP was adopted as a defense standard three years earlier in 1980. This enabled defense to begin sharing in the DARPA Internet technology base and led directly to the eventual partitioning of the military and non- military communities. By 1983, ARPANET was being used by a significant number of defense R&D and operational organizations. The transition of ARPANET from NCP to TCP/IP permitted it to be split into a MILNET supporting operational requirements and an ARPANET supporting research needs.
Thus, by 1985, Internet was already well established as a technology supporting a broad community of researchers and developers, and was beginning to be used by other communities for daily computer communications. Electronic mail was being used broadly across several communities, often with different systems, but interconnection between different mail systems was demonstrating the utility of broad based electronic communications between people.