[Top] [Prev] [Next]

Packet Switching and the Internet


Original need for survivable command and control systems in the cold war era. Early contributors were drawn from the ranks of defense contractors, federally funded think tanks, and universities: the RAND Corporation, Lincoln Laboratories, MIT, UCLA, and Bolt Beranek and Newman (BBN), under Defense Department funding.

In the wake of the Soviet Union's success in the launch of the Sputnik, the Advanced Research Projects Agency (ARPA) of the Department of Defense established in 1958 to make sure that in the future the U. S. would avoid technological surprises. Under the early leadership of J. Licklider, Robert Taylor, Larry Roberts, ARPA's Information Processing Technology Office provided substantial funding for communications networking research.

Licklider Taylor

1960s--Taylor and Roberts oversaw original development of the ARPANet; 1970s--Robert Kahn led the development of internetworking and extended the packet switching concept to radio and satellite networks. 1980s--shift from Defense-based sponsorship of the network, to the National Science Foundation and its use to interconnect university-based researchers, and finally the transition to private sector management of the Internet.

Store and Forward Networks

The concept was well known by the 1960s. Telegraphy systems: telegrams forwarded from one site to the next in the network, where messages were transcribed and retransmitted to the next site down the line in the direction of the final destination. This does not fully capture the nature of a packet switching network however.

Baran and Davies

early 1960s: Paul Baran, working at RAND Corporation, postulated many of the key concepts of packet switching networks that were ultimately to be implemented in the ARPANet. Motivation was to use radically new approaches to build survivable communications systems. An engineer working for the British government, Donald Davies, had similar ideas around the same time, and is credited with coining the term "packet".

Baran Davies

Telephone system is centralized switching architecture; rigid concept of connection or "circuit" that must be established between the parties of a communications. If a pathway or switch is broken (or destroyed) during a connection, the path will be broken and the communications will fail. Unacceptable in a survivable system. Replace centralized switches with large numbers of distributed routers, each with multiple connections to adjacent routers. Messages would be divided into parts (blocks or packets), routed independently, on a packet by packet basis. Some advantages when dealing with unpredictable traffic demands.

Dynamic packet-at-a-time routing, which Baran called "hot potato routing," allows traffic to be statistically multiplexed over available communications paths, gracefully adapting to the traffic demands and better utilizing existing link capacity without the need for preallocated link bandwidth. Local information used to choose the next hop on the routing path; no centralized point of control in the system. This is the essence of a packet switching network.

Development of the ARPANet

Middle 1960s: First efforts to interconnect geographically distributed computers. Tom Marill, Computer Corporation of America, coined term "protocol" to describe the rules of exchange of data between machines. Under ARPA sponsorship, his team and additional engineers at MIT's Lincoln Labs experimented with the first cross country interconnection of computer via telephone modems.

1967: ARPA proposed ambitious program to connect many host computers at key research sites across the country, using point-to-point telephone lines. Wes Clark proposed idea of using separate switching computers, rather than the hosts, to serve as the routing elements of network, thereby offloading this function from the timesharing hosts. Bolt Beranek and Newman got contract to build the interface message processors (IMPS). Frank Heart, Bob Kahn, David Waldin, Willy Crowther, and other engineers at BBN develop necessary Host-to-IMP and IMP-to-IMP protocols, the original flow control algorithms, and the congestion control algorithms. The ARPANet is born with it first four nodes at UCLA, UCSB, SRI, Utah.

IMP Heart

The BBN Team

1969: UCLA becomes first node of the ARPAnet. Second IMP placed at SRI a few months later. The IMP was later followed by the TIP, or Terminal IMP, to allow dial-up access to the ARPANet without the need for a host computer login.

The Aloha Net

1969: Norm Abramson launched Aloha Project at University of Hawaii. Motivated by the poor telephone lines in the Hawaiian Islands, funded by ARPA to investigate how to build a packet switched network based on fixed site radio links. They developed significant new technology for contention-based media access protocols, the so-called "Aloha Protocols," and applied these techniques to satellites as well as radio systems. Their work was built upon by Robert Metcalfe at Xerox PARC, leading to the development of the Ethernet protocols.

The Need for Internetworking

1972: Kahn left BBN for ARPA. Roberts and Kahn launch ARPA Packet Radio Program: packet switching techniques on the mobile battlefield. ARPA also created a packet-switched experimental satellite network (SATNet), with work done by Comsat, Linkabit, and BBN. Motivates need for technology to link these independent networks together in a true "network of networks."

1973: Kahn and Vint Cerf (Professor at Stanford), developed concept of a network gateway as well as the initial specifications for the Transmission Control Protocol (TCP). Breakthrough concept: shift transmission reliability from network to end hosts, thus allowing protocol to operate no matter how unreliable the underlying links. "End-to-end" argument: the underlying network is only as strong as its weakest link; improving reliability of a single link or even an entire subnetwork has little effect on end-to-end reliability. The best place to implement these functions are at the ends of connection, not in the network.

mid 1970s: ARPA began to transition the network. Operational responsibility transferred to Defense Communications Agency. Cerf joined ARPA to complete design of Internet Protocol Suite, oversee separation of the routing portions of the protocols (IP) from the transport layer issues (TCP), and transition the new protocols into the ARPANet. Switch over took place in 1983.

late 1970s: Berkeley UNIX played important role in moving the TCP/IP implementation onto a platform independent operating system and away from the BBN proprietary IMPs. This formed the model for later engineering workstation and personal computer support for TCP/IP.

early 1980s: ARPAnet technology began to move beyond the limited number of members of the ARPA research community. The CSNet project eventually led to the NSFNet under National Science Foundation sponsorship, dramatically increasing the number of nodes on the network.

early 1990s: NSF privatized the network, leading to the explosion in Internet service providers we see today.

The Internet and Convergence

[Top] [Prev] [Next]

Last updated 2 April 1997, Randy H. Katz, randy@cs.Berkeley.edu