Robert Melancton Metcalfe

Robert M. Metcalfe was born in Brooklyn and grew up in a suburban ranch house in Bay Shore on New York’s Long Island. His father was a unionized technician in an aerospace company whose work focused on testing gyroscopes. Metcalfe inherited his father’s love of technology, playing with toy trains and building machines from relays. He excelled in mathematics and science, going on to study at MIT where he “enjoyed every minute of it” despite the pressure and late nights.[1]

To support himself during college Metcalfe worked at Raytheon to program a submarine targeting computer in assembly language. Having initially focused on number theory and then physics, he eventually graduated with two undergraduate degrees, one from the electrical engineering department (then home to MIT’s computer scientists) and one in management. Among the professors he worked with were Jay Forrester and Marvin Minsky.

ARPANET

For graduate study Metcalfe headed across town to Harvard’s program in applied mathematics, home to its computing specialists, which he “hated from the very first day.” Before long he was also working back at MIT for Project MAC, a group funded by the Defense Department’s Advanced Projects Research Agency (ARPA) to explore new modes of interactive computing. Like other ARPA-funded research centers M IT was planning for ARPANET, the first network based on the packet switching approach that underlies today’s wired and wireless networks.

Traditional networks, like those used to set up telephone calls, opened electrical circuits between the two ends of the call and maintained them until it finished. Packet switching broke messages into multiple data packets. Each packet was tagged with address information and routed separately across the network. ARPA provided each network site with an IMP, a small computer programmed to exchange data with other IMPs. Each data packet was copied from IMP to IMP over leased telephone lines until it finally arrived at its destination.

Metcalfe “had no idea what a network was” when he started work but his undergraduate courses had included digital design and computer engineering.[2] He designed interface hardware and protocols to join the group’s main computer, a Digital Equipment Corporation (DEC) PDP-6, to the local IMP. Metcalfe made several important contributions to the small group setting up the ARPANET. For example, he contributed to the initial proposal for the Telnet protocol that let users of one computer log into another system and use its resources over the network.[3] Most memorably, he was a host and organizer of a 1972 demonstration of the network and its applications. Held at the Hilton Hotel in Washington, DC in conjunction with a major communications conference, the event brought widespread attention to the ARPANET for the first time.

Metcalfe wrote up a Ph.D. thesis based on his ARPANET work. It turned out that Harvard didn’t like him either. At what was supposed to be the grand conclusion to his studies, recalls Metcalfe, he walked into a “room full of people who I vaguely recognized and promptly failed my defense.” His thesis just wasn’t mathematical enough. Salvation came from his exposure to the AlohaNET system being created by the University of Hawaii to shuffle data packets between computers and terminals located on different islands. All network stations shared a common frequency, meaning only one machine at a time could transmit data. When two devices tried and failed to talk at the same time, each waited for a randomly determined interval before trying again. As the network got busier these collisions of packets got more frequent, throughput dropped and wait times rose. The element of chance made the degradation of network performance under increasing load into an interesting statistical problem, one that Metcalfe correctly believed would add enough mathematical rigor to his revised thesis for it to pass muster.

Xerox PARC

By the time Harvard finally accepted his thesis had been in California for a year, working at Xerox’s Palo Alto Research Center (PARC). PARC’s computing group was set up by Robert Taylor, who was recruited away from ARPA and wanted Metcalfe for his talents in networking. The accomplishments of the PARC team over the next few years have been recognized with several other Turing awards: Chuck Thacker for designing the Alto workstations that anticipated by fifteen years the capabilities of the most powerful personal computers of the late-1980s; Butler Lampson for his work on the hardware and software of the Alto; and Alan Kay for the work he did with Adele Goldstine to develop object oriented programming and graphical user interfaces.

The ARPANET joined together big computers, each with many simultaneous users, across large distances using low speed connections well suited for transmission of email or the display of screenfuls of text. PARC needed something different: high speed local networking. It was putting a powerful, single user computer under the desk of each researcher. The Altos needed to be connected to each other, to the group’s shared minicomputers, and from there onward to the ARPANET. Xerox was also pioneering laser printing, a good potential fit with its core office copier business. The printers were very expensive, and so had to be shared by many computers. Because they printed graphics at high resolution this meant moving huge amounts of data quickly from workstation to printer.

In 1973 Metcalfe and his colleague Dave Boggs began to develop network technology to meet those needs. They called it Ethernet, after ether, an imaginary substance that early scientists had hypothesized as the invisible medium that carried light waves through apparently empty space Ethernet was both fast and simple. The initial version had a theoretical capacity of almost three megabits per second, fast enough to keep a hungry laser printer fed with data. The simplicity lay in the architecture: a single thick coaxial cable up to a mile in length snaked through the office, into which taps were sunk to connect workstations.

Ethernet drew from Metcalfe’s work on AlohaNET but with a coaxial cable rather than a radio frequency as the common medium. Only one computer could transmit at a time. Others listened, waiting for a packet addressed to them. As with AlohaNET, if two tried to transmit at once each would wait for a random interval before trying again. The term CSMA/CD for “carrier-sense multiple access with collision detection” was later introduced to encapsulate this technique.

By 1975 around a hundred Ethernet circuit boards had been fitted to PARC’s Alto systems and to its MAXC minicomputer which served as an onward gateway to the ARPANET. Turning Ethernet into a useful network able to handle tasks such as file transfer and email required software as well as hardware, specifically a set of protocols known as PUP (for PARC Universal Packet). Because it routed data between networks, PUP had to solve some of the same problems confronting the ARPANET team as it began the internetworking project that culminated with the TCP/IP protocol suite that defined the Internet. Vint Cerf, co-designer of TCP/IP, has acknowledged PUP as a significant influence, though because Xerox lawyers then saw PUP as proprietary Metcalfe’s colleagues were unable to share their work directly with the team working on TCP. Instead Xerox employees gave strikingly well-informed feedback on the draft design in a way that shared their experience indirectly.

Metcalfe briefly left Xerox for an advanced technology group at Citibank. He returned in 1976 to help commercialize Alto technology in the innovative but commercially unsuccessful Star series of networked graphical workstations. Metcalfe and his collaborators sped up Ethernet, to an unheard of ten million bits per second, and further developed Xerox’s networking protocols for commercial office use.

3Com

Ethernet worked splendidly, but by 1978 Metcalfe was getting frustrated. The Star workstation was still years from being ready for sale and Xerox wanted to hold Ethernet back until it was ready. Metcalfe saw a much bigger potential market for Ethernet than for the Star. He left Xerox to promote Ethernet as a consultant. DEC, second only to IBM in the entire computer industry and the leading producer of minicomputers, agreed to adopt Ethernet. Intel offered to make chips to support it. Xerox agreed to share its trademark and patents for Ethernet.

The effort by DEC, Intel and Xerox to promote Ethernet was boosted, but also complicated, by the creation in 1980 of the 802 standards committee under the auspices of the IEEE. The committee’s designated goal was the standardization of network equipment able to transmit millions of bits a second over a one-mile range. Ethernet was already working and had some powerful supporters but other computer companies promoted rival technologies so in the end the committee endorsed not just Ethernet (formalized in 1983 as 802.3) but also IBM’s Token Ring (802.5) and another protocol. By then more than twenty firms were already building Ethernet equipment.

One of them was 3Com, cofounded in 1979 by Metcalfe and Gregory L. Shaw. It was named for computer, communication, and compatibility. The other key figures in the 3Com’s early years were cofounder Howard Charney, a lawyer and former fraternity brother of Metcalfe, and Bill Krause who was hired as the firm’s president in 1981. After the arrival of professional management and venture capital investments, Metcalfe focused primarily on sales and marketing.

Standardizing Ethernet meant that devices from different companies would be able to exchange data with each other, which in turn made customers less worried about building networks around the products of unknown companies. Other Ethernet startups focused on connecting minicomputers to terminals, but 3Com aimed to sell Ethernet boards to companies that made computers. Its early market was tiny because Ethernet cards were too large and expensive for personal computers. Ethernet also paired well with the increasingly popular Internet protocol TCP/IP, which was becoming a standard feature of UNIX-based minicomputers and workstations around this time. As Metcalfe said later, “Ethernet is basically the low-level plumbing for the Internet.”[4]

3Com’s first major success came with its EtherLink card, launched in 1982 to add networking capabilities to the newly arrived IBM PC. 3Com used highly integrated chips to shrink the card, greatly lowering the cost of Ethernet capabilities. It also introduced cheaper and easier cabling methods. Strong sales for EtherLink paved the way for an initial public offering of 3Com stock in 1984. By 1987 3Com had overtaken its competitors to become the leading producer of Ethernet hardware.

IBM PCs were rarely connected to the Internet. Instead, local area networks were used to share expensive printers and access data files held on server computers. In the mid-1980s Metcalfe headed 3Com’s Distributed Systems Division, focused on servers and computing technologies. 3Com initially tried selling the software and servers to go with its EtherLink cards. Later it offered diskless PC workstations and invested heavily in a partnership with Microsoft on the LAN Manager operating system. These projects eventually lost out to the combination of Novell’s Netware server software with standard PC hardware and 3Com interface cards. Together, these products finally made high speed networking a fixture of corporate offices.

Later Careers

Metcalfe left 3Com in 1989. For a while the firm continued to grow rapidly, making acquisitions of firms producing modems and other communications equipment and diversifying into areas such as the Palm Pilot series of handheld computers. Eventually, though, the integration of networking capabilities into devices destroyed the market for networking cards. After several rounds of downsizing 3Com was acquired by Hewlett Packard in 2009.

During the 1990s Metcalfe focused on journalism. His highest profile role came as a weekly columnist for the computer industry news magazine Infoworld but he also served stints as the magazine’s publisher and as vice president of technology for its parent company, the tech publishing empire IDG. His most influential idea was Metcalfe’s Law, the idea that the economic value of a network accelerates ever faster as it grows (in proportion with the number of users squared). His most notorious column, though, was a 1995 prediction that the Internet would “collapse” within a year because its growth was outstripping the capacity of its core infrastructure. Ever the showman, Metcalfe attracted front page attention by literally eating his words with the help of a blender, making the failed prediction a “brilliant success.”

His next career, from 2001, came as a venture capitalist with the Massachusetts based Polaris Partners. Metcalfe served on, and often chaired, the boards of the companies he invested in. In 2011 he shifted gears again, as a professor of innovation and entrepreneurship in the Cockrell School of Engineering of the University of Texas with a mandate to “lead innovation initiatives.” Metcalfe retired in 2022 but retains affiliations with MIT’s Computer Science and AI Laboratory and the Wolfram Institute.

The Legacy of Ethernet

Ethernet ports remain the standard way of making wired connections between desktop computers, servers, and networking equipment through the standard has been revised many times to accommodate new technology and achieve faster speeds. The most important shift came with Fast Ethernet, standardized in 1994, which ran ten times faster than the original commercial version. During the 1990s Ethernet migrated to cabling that joins each computer directly to a network switch able to route each data packet efficiently to its destination. Today Ethernet ports capable of ten gigabits per second are appearing in consumer hardware, and much faster variants are used within computer data centers. All these technologies have been standardized as extensions to the original IEEE 802.3 standard for Ethernet.

Little of the original technology survived, not even the fundamental approach of having all devices in a local network segment share a common connection. As Metcalfe said in 2006, “the things they call Ethernet these days bear little resemblance to the thing Dave Boggs and I built in 1973.” Other than the Ethernet name, which Metcalfe refers to as a “brand,” he believed the only real continuity was the “business model” of standardization, continued compatibility, and fierce competition between implementors.

The most direct line of descent from Metcalfe’s original technical approach lies not in modern Ethernet but in the WiFi connections used by laptops, phones, and other mobile devices to connect wirelessly to nearby network routers. WiFi is the brand name for IEEE standard 802.11, which relies on many devices coexisting on each radio frequency, just like the wireless AlohaNET system that first inspired Metcalfe.

Author: Thomas Haigh