Ethernet — a
networking protocol name for the ages
Michelson, Morley, and Metcalfe
In the beginning, Ethernet was optional. When Bob Metcalfe and David Boggs cooked up their network protocol at Xerox PARC in the early 1970s, it was meant to connect the research haven’s now famous Alto machines — but only if researchers felt the need. “Each scientist would get a kind of Alto order form,” Metcalfe remembers, “and you had to check a box if you wanted Ethernet.”
Then, one afternoon, with maybe ten Altos on the desks of ten PARC researchers, someone accidentally disconnected a networking cable. When ten people stood up to ask “What happened?,” Metcalfe realized his fledgling network protocol might be a keeper. “From then on,” he says, “Ethernet was not an option.”
More than thirty years later, Metcalfe went looking for a new Ethernet cable, strolling into an everyday American electronics retailer. “The woman at the cash register took me to a twenty-foot-wide wall filled with cables and said ’What color do you want?’”
After Metcalfe brokered an epic Ethernet group handshake between Intel, Xerox, and DEC in the late 1970s and brought his protocol to market through a new company he called 3Com, Ethernet shed its optional status on a global scale, evolving from LAN hardware into mainstay internet plumbing. In 2008, according to IDC, 350 million Ethernet switch ports were pushed out into the world. And that doesn’t include WiFi gear.
Last year, in honor of his Ethernet exploits, Silicon Valley’s Computer History Museum anointed Metcalfe with its annual Fellow Award, and this week, he turned up at the Mountain View geek mecca to reminisce — and, in that inimitably-opinionated Metcalfe way, trumpet his version of the future.
Harvard’s dark little heart
Metcalfe joined the Palo Alto Research Center in 1972, just after some Harvard profs failed his PhD thesis defense. His research focused on, yes, packet communication. And packet communication wasn’t a place where he and Harvard saw eye-to-eye.
When he first arrived at Harvard, Metcalfe offered to plug the venerable university into the Arpanet, a fledgling research network that would one day evolve into the modern interwebs. But Harvard declined.
“Harvard to this day is ambivalent about engineering, in their dark little heart of hearts,” he tells The Reg. “When I said I wanted to get them on the Arpanet, I was behaving like an engineer in a department of applied mathematics, and they said 'No, we’re going to hire a company do it.’”
That company was Bolt, Beranek, and Newman, a major cog in the rise of the Apranet. But the way Metcalfe tells it, BBN promptly hired another grad student to do all the work he was willing to do for free.
So, he went down the road to MIT, his undergraduate alma mater, offering his networking services to their DEC PDP-10 mainframe. MIT said “Yes,” and Metcalfe spent his Harvard career doing research on MIT’s Arpanet IMP (interface message processor). In the end, his Harvard overlords were not amused.
“I was — I still am — naive about politics,” Metcalfe says. “I didn’t really understand the process. The professors should have noticed this, but they didn’t. I showed up for my thesis defense and they gunned me down.”
But a new job was already in place at PARC — the Xerox research center that would give rise to everything from laser printing to the GUI interface — and PARC agreed to let him finish his PhD there. This, Metcalfe says, is the sort of luck that turns a life into history.
“My first stroke of luck was being born to my parents. My second stroke of luck was being given the networking job in a building full of personal computers. This was a problem that had never before occurred in the history of the world. The solution was not as amazing as the luck of getting that assignment.”
Next page: Into the ether
Actually, my network virginity was broken on Apollo Token Ring (with uni-directional coax cabling, dammit). Worked quite nicely at the time because the response order of the nodes was deterministic. Too long ago for me to remember Apollo's corporate colour. I have only ever decommisioned the IBM thing.
Fair'nuff on tokenless ring. The real point is that very little of Metcalf's thinking is left in the network we use today. Common cable and CSMA-CD just didn't withstand the scale that data networking demanded. I susbscribe to the view that people are making a hero out of someones inflated opinion of himself. Credit him with creation of the name but what else?
I'd guess at three of Jas. Watts ideas being still used in turbines.
Since you ask, are we talking about Cherubim, Seraphim or Thrones on this here pin?
"If memory serves CSMA-CD was under severe strain with the 100Mb standards and was done away with for 1Gb."
I would have said under severe strain with 1GB (but it was in the standard for political reasons) and abolished officially with 10GB. There were tighter restrictions on the number of repeaters you could chain at 100MB/s but that was close to being a moot point.
[Citation needed] "They replaced it with a 'token passing algorithm'."
They replaced it by ceasing to support shared media LANs and requiring full duplex operation 'anyone can talk when they like'. Whether you call it Ethernet without CSMA/CD or Token Ring without the token is an 'angels dancing on the head of a pin' question. I think you'ld have to be looking through (if not rose coloured) then big blue coloured spectacles if you think token ring really had the last laugh.
"at a high technical level as would be required to be able to afford to pay for the research needed for either solar power satellites "
What research are you thinking of? Mfg techniques, including how to use microwave oven grade magnetrons (very cheap but not considered precision frequency sources) were all worked out in the 70s, 80s and 90s.
OTOH if you mean lowering launch costs by 2 orders of magnitude that requires skilled engineers dedicated to the idea with a belief its possible and an adequate but not extravagant budget to try. That's engineering, not science. All know strategies to build an SPS fail on economics. Roughly speaking launch requirements give systems with very long payback. Reasonable payback needs big launch. The Iridum sat network cost c$5Bn to launch. Its total mass (66 sata @c1500lb ea is < 17000Lbs)is nowhere near a relatively small SPS, its too low and will still need replacing.