#5: Internet: Part-I: ARPANET

The graph below shows growth of the Internet (Source: ITU). It has reached 5 billion users which is 70% of the world population and Its life cycle has followed the s-curve discussed in Article-#3. However, its growth has not declined nor has it plateaued.

Any user who joins the Internet gets locked-in because it is useful. This can be compared with adoption of an essential utility like electricity. The electric-ecosystem has grown from a light-bulb to refrigerators, ovens, computers and many other things likewise the Internet-ecosystem will grow from laptops, mobiles, smartwatches to many other forms and sizes of computer-things, which implies, the growth we have experienced so far is just the tip of the ice-berg.

Technically the Internet is a network for computer-things to communicate. It has originated from the research funded by governments in the US, UK and France after World War-II. Post-war, Dr. Vannevar Bush, who led the scientists for applications in warfare, wrote an article titled ‘As We May Think’ which inspired the work of many researchers working around that time.

After the war, the scientists had to disperse back into public life and the government lost the intellect it had gathered. These scientists were absorbed by think-tanks like RAND Corp. and BBN who were engaged in research across multiple fields and industries. Later, many of these think-tanks along with universities collaborated to make the Internet a reality.

The war was over but there were new problems to solve, what if a military command and control network breaks during a war? Extra telephone lines could reestablish the connection but what if many nodes in the network are damaged. Back then AT&T was a monopoly in the telephone market which made it a single point of failure. These problems had to be de-risked.

The US Department of Defense (DoD) engaged RAND Corp. to find a solution. In 1961, Paul Baran, a researcher working there, proposed a theoretical solution in his paper titled ‘On Distributed Communication’ in which he described use of standard ‘message blocks’ with ‘store and forward’ transmissions along with ‘message routing’. We know this is how the Internet works today, but his theory was rejected by Bell System (who was the research implementation partner) because the existing telephone networks were analog.

In 1965, Donald Davies, who was working for National Physical Laboratory (NPL), in the UK, came up with similar thoughts like Paul Baran but for a computer-network and built the NPL-network. One of the insights that led him to the same design was, computer traffic was ‘bursty’ with periods of silence. This indicated the network was faster than the user and hence under utilized.

The idea was to break the data to be communicated between two computers into small blocks called packets which are labeled to indicate the origin and destination. These packets are then stored and forwarded over the network until all the packets arrive at the destination computer. If packets are lost in transit then the message can be resent by the originator. Such Packet switched networks do not assign a dedicated path for any connection between two computers, instead the whole network is shared for everyone.

This was also a solution to the problem of command-control failure during war. If packet switching technology is used and redundancy is planned then failure of the network due to broken links or nodes can be reduced considerably by rerouting information on a different path thus making the network resilient.

In 1957, the first artificial satellite Sputnik-1 was launched by the Soviet Union, which resulted in the US creating the Defence Advanced Research Project Agency (DARPA), to remain on the frontiers of science and technology. In 1961, DARPA started an office for computer research and named it ‘Command and Control Research’. In 1962, J.C.R. Licklider was made the first director of this office.

When at MIT, Licklider had led the human factors team on the SAGE project. In 1957, he left MIT and joined BBN as a Vice President. During his tenure at BBN the first time-sharing system was built. In 1960, he published a paper titled ‘Man-Computer Symbiosis’ which envisioned a future where computer technology augments humans for effective thinking and decision making. In 1963 he sent a memo to DARPA colleagues describing a computer network concept similar to the Internet.

To realize the vision, Licklider set up computer science departments at major universities. Mainframe computers were funded for research projects. Users who needed computer access were given dumb computer-terminals which could send input to the mainframe and display the output received from it. This was the early idea of a local network where a faster computer's time was shared with slow user terminals.

In 1965, Robert Taylor who joined DARPA as a deputy had three terminals in his office each connected to different networks. He observed that the networks had developed a community of users who were disconnected from each other. He initiated the ARPANET project to connect these networks for resource sharing and appointed Lawrence Roberts (Larry) as the program manager and also engaged Leonard Kleinrock to analyze the packet network technology. The modern Internet originated from the ARPANET project.

Larry decided to use packet switched networks in ARPANET. The contract was given to BBN, where its design was led by Robert Kahn. Minicomputers were used as nodes in the network and were called IMP. Each mainframe (host) was connected to an IMP (router) using a serial interface. The IMPs were networked using telephone lines and they implemented a reliable IMP-to-IMP protocol. A host-to-host protocol called NCP was implemented whose development was led by Steve Crocker of UCLA with his students Jon Postel and Vint Cerf. The network had features like login, file transfer for resource sharing. The IMP when configured as a Terminal Server was called TIP. It was used to connect multiple computer-terminals in the network. In 1972, Bob Kahn demonstrated the ARPANET with usage scenarios at ICCC which built a conviction that packet switching was a real technology.

In 1970, the first international network was formed between ARPANET (US), NPL-Network (UK) and NORSAR (Norway). NORSAR was connected to the ARPANET using a satellite link. UK decided to join the European Union so instead of NPL-Network the research group at University College London (UCL) led by Peter Kirstein was chosen as the second network to connect to the ARPANET. In that same year, Larry proposed that it was possible to use 64 kbps/s satellite link as a medium shared by multiple satellite earth stations within the beam's footprint as a result a new project called SATNET for satellite packet networks was initiated.

In 1971, the French government funded a computer network called CYCLADES. It was led by Louis Pouzin who had worked on a time-sharing system at MIT. He studied ARPANET and found it overly designed. Unlike ARPANET, this network used a connectionless protocol in the packet switch (IMP). Rearranging of packets then became a responsibility of the host. This end-to-end principle became fundamental to the design of the Internet later. The uniqueness of this project was, networks with different networking-hardware were connected using gateways and the resulting network was called catenet (concatenation of networks).

In 1971, a wireless packet network called ALOHAnet was led by Norman Abramson and Franklin Kuo. The goal was to share a central mainframe computer at the university of Oahu with users on the group of Hawaii islands using commercial low-cost radio hardware. The mainframe computer (host) interfaced with a minicomputer (Menehune) as a communication processor (IMP) which was connected to a Ultra High Frequency (UHF) transmitter and receiver using modem. The island users joined the network by connecting their computer-terminal to a special hardware called Terminal Control Unit (TCU). The challenge on this network was simultaneous access of the shared wireless medium by all users which was addressed by implementing a ‘random access’ protocol.

In 1973, inspired from ALOHAnet, DARPA initiated a project called PRNET to explore communication in a wireless ad-hoc network. Its purpose was to tackle hard problems like multi-hop wherein packets have to be moved between different base stations to reach a mobile station. One of the use cases was, there are a series of ground PRNETs, PRNET-1 and PRNET-2 each connected to the Internet. There is a mobile Packet Radio Van which while moving transitions from connectivity with PRNET-1 to PRNET-2.

In 1973, while NPL and NORSAR networks were integrated with ARPANET, projects like SATNET, Secure-Net and PRNET also had to be linked to carry their own set of developments. ARPANET itself was evolving to high network speeds, new IMPs and more complex routing. The early ARPANET had uniform topology but the diverse networks needed changes to the IMP, and BBN who was central to these developments was becoming a bottleneck. All these challenges had to be met without disrupting the existing functional ARPANET.

Drawing on the ideas from CYCLADES network, Rob Kahn and Vint Cerf developed the concept of a gateway (a router) for internetworking the diverse packet networks to be linked with ARPANET. The host-to-host protocol had to be changed for internetworking functions that would deal with different kinds of packets, this resulted in the Internetwork Protocol which is now identified as Internet Protocol (IP). The gateway was designed to be connection-less hence it was unreliable like in the CYCLADES network. Reliability became a responsibility of the new host-to-host internetworking protocol called Transmission Control Protocol (TCP) which was connection oriented.

In 1973, Robert Metcalfe who had joined Xerox PARC was tasked with creating a wired local area network (LAN) which would allow the personal computers that PARC was building to communicate in an office environment. Metcalfe drew upon the ideas of ALOHAnet random access protocol to create a new protocol for wired medium and called it Ethernet. It defined how packets were transmitted and received on a wire.

In the 1970s, besides Ethernet many LAN technologies like Token Ring, ARCNET and FDDI. were developed for wired mediums. The LAN was primarily used for sharing storage and printers in a local-network. Enterprise adoption of PCs triggered the growth of these networks. With different LAN technologies in the market each network chose different hardware, cables, physical medium and network software. A company named Novell came up with Netware, a software solution for IBM Compatible PCs and supported different LAN hardware around that time. Their software required a Certified Novell Engineer (CNE) to set the local network. Later they lost the market to Microsoft Windows OS which had an GUI for easy network configuration.

In 1977, Robert Metacalfe left Xerox and started a company called 3Com to promote Ethernet. He convinced Microsoft and Digital Equipment Corporation (DEC) and Xerox to promote Ethernet as a de facto standard. In 1980, the first ethernet standard was published with 10 Mbits/s speed. In 1981, 3Com started selling Ethernet Network Interface Cards (NIC) for all kinds of computers which became the de facto LAN hardware due to the growth of PCs. Today all computers come with a default Ethernet port with speeds ranging from 10 Mbits/s to 400 Gbits/s and it is the most widespread wired medium protocol.

Different protocols during the evolution of the Internet gave rise to Protocol wars where each implementation was pushing for its own design, so standardization was required. One of the efforts of standardization was the OSI Model which recommended the OSI Protocols. There was also another standard body called Internet Engineering Task Force (IETF) which was funded by the US government (now under Internet Society). The TCP/IP protocol of the Internet Protocol Suite, which was standardised by IETF, won this protocol war because it was rapidly adopted by many networks.

The adoption of TCP/IP started in 1982 when DoD declared it as a standard for adoption by ARPANET. In 1983, the first international networks NORSAR and UCL migrated. In 1985, the Internet Architecture Board (IAB), a committee of IETF, held a workshop for 250 vendors to promote the protocol. In the same year Daniel Lynch started an interoperability conference called Interop. IBM, AT&T, DEC were the first to adopt TCP/IP though they had competing proprietary protocols. There were protocol stack implementations from small companies and individual programmers for different operating systems. In 1989, University of Berkley open sourced their TCP/IP implementation which was ported to many platforms.

By 1984, the ARPANET which had grown to 113 nodes was reduced by 68 nodes that became a part of the Military-Network (MILNET). After the introduction of a new network backbone called NSFNET in 1985, the ARPANET project was formally decommissioned in 1990.

This article started with a graph showing the growth of Internet Users from 2005. Would the number of ‘individual Internet users’ be the right parameter to measure its growth? Technically the Internet is a network of computers, so the count of computer-things would be the right parameter to measure its growth. Today many users own two (laptop, mobile) or more (laptop, mobile, tablet, smartwatch, ...) computer-things so the total growth of the Internet must be more than 5 billion.

Will reserve the number crunching for another article, there are more facets to growth of the Internet.

Stay tuned!