Thursday, July 25, 2013

Ethernet Turns 40


Ethernet Turns 40

Highlights from co-inventor Robert Metcalfe’s IEEE oral history


 15 July 2013



Photo: Robert Metcalfe
















The technology’s genesis dates to 1973, when Xerox PARC, in Palo Alto, Calif., built the Alto personal computer. Robert Metcalfe—who was working there at the time and finishing his Ph.D. dissertation for Harvard—was assigned to design a network for the machine as well as a card that could be plugged into it to enable communication with the Advanced Research Projects Agency Network (ARPANET), the world's first operational packet-switching network and the precursor to the Internet. Xerox PARC was also building a laser printer that could print 500 dots per inch at a speed of one page per minute. The hope was to use it as a central printer for all of Xerox PARC’s personal computers. Hundreds of computers had to be connected.We take for granted how computers in the workplace are all connected together. Sharing files with coworkers, sending documents to a network printer, and accessing data from a networked server are all routine procedures thanks to the invention of Ethernet technology. On Ethernet’s 40th anniversary, the IEEE History Center shares excerpts from the oral history interview it conducted with Ethernet’s co-inventor Robert Metcalfe in February 2004.

“If you do the math on that printer,” Metcalfe explained in his interview with the IEEE History Center, “that was a lot of bits per second…500 times 500, times 8.5, times 11, per minute…that’s a big number. RS-232 was then the standard for interconnecting terminals, and it frequently ran below 19.2 kilobits per second. It was not even close to what was needed.”
To keep the printer busy, the PARC network had to run in megabits, not kilobits, per second. On 22 May 1973, Metcalfe distributed a memo describing the high-speed local network he had in mind. He called it the EtherNet, which was soon rewritten asEthernet. In June, Metcalfe teamed with David Boggs, another PARC employee and an experienced amateur radio operator, to build the network. The similarities between amateur radio—where multiple transmitters use the same frequency and have developed an etiquette for not interfering with each other—and a network of computer terminals communicating over the same wires, made Boggs’s experience particularly valuable.
WHAT MADE IT WORK
Ethernet was not the first attempt to build what would come to be called a local-area network, or LAN. (The acronym LAN would not even come into use for about eight years.) Metcalfe was influenced by the ALOHAnet, which relied on radio to connect computer users in the Hawaiian Islands—it was the first public demonstration of a wireless packet data network. Norm Abramson, professor of electrical engineering and computer science at the University of Hawaii and the director of ALOHAnet, wrote a paper evaluating it. After Metcalfe read Abramson’s paper, he rethought ALOHAnet’s traffic model, which assumed that if two packets of information collided, the users would keep typing and resending packets in the absence of an acknowledgment having been received. The assumption was that if two terminals sent packets at the same time and they interfered with each other, they would each try again but not at the same time.
Metcalfe believed, however, that users who did not receive an acknowledgment were more likely to stop typing and would wait before transmitting again. Thus the packet traffic would decrease, resulting in fewer collisions. With fewer collisions, more of the data would be able to get through. This is the same principle as a traffic jam on a highway: There may be a lot of cars on the road, but no one is getting anywhere. When there are fewer cars on the highway, they are moving, and more of them reach their destinations. With Ethernet, as with any network, the idea is to fill the communication channel with the most traffic that can still move efficiently.
For Ethernet networks, Metcalfe designed a “back-off algorithm” to control the retry time and thus the traffic volume. When the network saw light traffic, it would randomize the retry time over a short interval. But as the traffic load increased, the terminals would take more time before resending. A collision would be evidence that a channel was busier than anticipated, and so the retry time would be lengthened. This insight made Ethernet hugely more efficient. Fewer people were typing, and so a higher percentage of data packets were able to get through. Collision detection made Ethernet different and made it fast.
Metcalfe and his colleagues chose coaxial cable for Ethernet’s physical pathways because they wanted to be able to add or subtract nodes without bringing the network down. They also wanted to be able to insert nodes without cutting wires. David Liddle, who worked down the hall from Metcalfe and Boggs, pointed out something called a cable television tap. This allows tapping into a coaxial cable without cutting it. Ethernet was designed to be media-independent, however. By using the word ether in its name, the PARC people alluded to the possibility that Ethernet could be based on coaxial, twisted-pair or optical fiber wiring and, eventually, on Wi-Fi. Another important operating principle for Ethernet was that it would be vastly distributed; there would be no central control.
“In 1973, the Internet [which at that time was basically the ARPANET] on a good day ran at 50 kb/s,” Metcalfe reflected in his oral history. “Ethernet ran at 2.94 megabits per second.” Over the years, people urged Metcalfe to round the number up to 3. He always resisted, as a matter of emphasis: If one rounds 2.94 Mb to 3 Mb, the rounding error is more than 50 kb/s. “Ethernet’s round-off error was bigger than Internet,” said Metcafe. “That’s how fast Ethernet was running.”
Xerox filed for a patent application for Ethernet on 31 March 1975, listing Robert Metcalfe, David Boggs, Chuck Thacker, and Butler Lampson as its inventors. U.S. Patent No. 4063220, “Multipoint data communication system (with collision detection),” was issued on 13 December 1977.
For his work on Ethernet, Metcalfe was awarded the 1988 IEEE Alexander Graham Bell Medal and was the recipient of the 1996 IEEE Medal of Honor, IEEE’s highest award.
Learn more about the technologies that IEEE 802.3 “Standard for Ethernet” has helped enabled by visiting the IEEE Standards Association website. It includes a video of a conversation with Metcalfe, an “Ask Me Anything” session with him on Reddit, and a list of IEEE 802.3 Ethernet milestones.
http://theinstitute.ieee.org/technology-focus/technology-history/ethernet-turns-40

No comments:

Post a Comment