How did ARPAnet and ALOHAnet lead to Ethernet and go on to form the basis of the Internet we rely on today? Fifty years on, the man responsible, Robert (Bob) Metcalfe is named the recipient of the 2022 Turing Award.

The Internet is an indispensable part of 21st century life and so it deserves to be, but much of the hardware that supports the Internet is based on an invention that we tend to ignore.  Long before the Internet became a part of the popular culture there was still a need to create Local Area Networks. Small groups of computers, usually in the same building, needed to transfer data, share files and printers and a low cost method of connecting them together just had to be invented. This sounds an easy problem, but it isn't. To make it practical a whole new way of sharing a communications channel had to be invented. The result is sometimes called CSMA/CD but most of us know it as Ethernet. The name and the method were invented by Bob Metcalfe.

Robert Melancton Metcalfe, born April 7, 1946

Bob Metcalfe was born in New York in a Norwegian ghetto! His mother was Norwegian and his father was Irish - not an easy combination back in the 1940s. His father was a technician working in the space industry building gyroscopic platforms. He encouraged Bob to build model trains in the basement including the electrical switching needed to control them. In high school he built a very simple, by today's standards, computer. It did nothing more than add binary numbers together but in 1959 there weren't even pocket calculators and so it was a real achievement.

ARPAnet

With his particular interests in electronics it was natural for Metcalfe to enrol as an MIT student of electrical engineering. After graduating in 1969 he went to Harvard to study mathematics but still with a practical angle - the new subject of networking. While he was at Harvard he took a full time job with the Advanced Research Projects Agency (ARPA) on the big wide area network they were building - ARPAnet. Of course we now know that ARPAnet grew into the Internet but at the time it was just a way of connecting a small number of large computers. It did, however, introduce Metcalfe to the principles of packet switching. He wrote programs and helped with the design of the hardware.

It is important to know at this point that while ARPAnet was a packet switched network, i.e. one in which a data packet could be routed from one machine to another until it reached its final destination, there were hardwired long distance connections between pairs of machines. That is, the Wide Area Network (WAN) that was ARPAnet wasn't suitable as a model for building a Local Area Network (LAN) connecting many machines within the same building. You simply can't afford to run a cable between every machine in the building, the number of connections would be too great. You can't even afford to make individual connections between enough machines to make packet switching work.

In 1972 he completed his PhD work on packet switching networks and submitted his thesis. He had a job lined up at Xerox Parc - the now famous research facility that Xerox had set up to study the effects of personal computers and, more to the point, to work out what might happen to a photocopier company in the age of the paperless office! Unfortunately, Harvard decided that his thesis wasn't theoretical enough - after all it was supposed to be a mathematics PhD. Fortunately, Xerox decided to take him anyway and he became the resident networking expert in 1972. At first the focus of his work was on connecting the new machines being built at Parc to the ARPAnet but it became clear that something different was needed. For the first time ever there was an environment where there was a computer on every desk. Outside of Parc computers were still one to a building or one to a company and the ARPAnet methods of connecting things together was suitable. Inside Parc was the new world of the personal computer and such huge engineering solutions were not applicable.

Radio Packet Switching

Almost by accident Metcalfe discovered a paper by Norman Abramson of the University of Hawaii on the problems they were having connecting computer terminals on remote islands -The Birth Of Ethernet. The solution that they had invented was to use radio packet switching. The big difference with this method is that all the radios shared the same communications channel back to the main computer. It wasn't possible at the time to make "point to point" connections using different transmitter frequencies. When a terminal was ready to send some data it just did so, complete with an identification number indicating which terminal had sent the data. The remote computer, an IBM 360, would transmit a data packet on a frequency that all of the terminals were receiving but only the one that it was addressed to would accept the packet. Of course the problem was that two or more terminals might decide to transmit at the same time and so cause mutual interference. The solution was to have the terminals detect this interference by not picking up an acknowledgement from the main computer. Each terminal involved in the interference would then "back off" for a random period before trying again.  This method of networking was called the "ALOHAnet" and in the paper it was shown by statistical analysis that it all worked fine as long as the channel wasn't loaded to more than 17% of its capacity. Beyond this loading the interference and subsequent re transmissions needed simply clogged the system.

ALOHAnet sounded like a solution that would work for the Hawaiian islands, but 17% was a very low efficiency to have to swallow. Metcalfe, however, thought that the analysis was pessimistic because it assumed that terminal users would just carry on typing even when they hadn't received a response from the main computer. Metcalfe decided to do more complex but realistic analysis and Xerox sent him to Hawaii to work with Abramson. Finally his analysis was completed but it too showed that even under realistic conditions ALOHAnet wasn't stable under all load conditions. He did, however, show that if there was some way of stopping stations transmitting when there was congestion then it could be made stable.