Interview with the Father of Ethernet" - Bob Metcalfe
Three decades ago, the medium that would one day earn the moniker "information superhighway" had such light traffic that there was little need for lanes, speed limits or highway patrols.
The odds were slim enough that electronic messages sent in a fraction of a second between a small pool of academic and military researchers would collide with one another. But scientists at Xerox's famed Palo Alto Research Center knew that the situation was destined to change.
The task of devising a way to connect multiple computers to one another and to exchange messages over increasingly busy networks fell to Xerox PARC researcher Bob Metcalfe, who in a 1973 memo described the technology that would evolve into today's ubiquitous Ethernet protocol. Metcalfe's Ethernet wasn't the first of such network protocols--some preceded it, and many more followed. But it won out and became the the dominant local area networking technology for businesses, en route to its place in the Internet, besting a long series of what Metcalfe today calls networking "Godzillas."
Metcalfe, a 57-year-old native of Brooklyn, N.Y., earned from the Massachusetts Institute of Technology bachelor's degrees in both electrical engineering and management, and a master's degree in applied mathematics, before getting a doctorate in computer science from Harvard University. He completed his dissertation, "Packet Communication"--a study, now available in hardcover, of the Arpanet and the Aloha Network--the same year that he wrote the Ethernet memo. Now a resident of Boston with his wife and two teenage children, Metcalfe spoke to CNET News.com from LagunaNiguel, Calif., where he was attending IDG Forums' Vortex computer networking conference andpreparing for Thursday's 30th birthday observances for Ethernet in Palo Alto, Calif.
Tell me first of all how today's Ethernet has evolved from what you
described 30 years ago.
It's evolved tremendously. Today's Ethernet bears very little resemblance to
what David Boggs and I built in 1974, and it gives rise to the question, what is
Ethernet exactly? I have a bunch of answers to that. It's gotten faster, it's
been hubbed and switched unlike the original Ethernet. The original had packet
collisions, but the latest versions have very few of those.
Today's Ethernet technology is extremely diverse and has very little in common with what appeared in '74. The good news is that they still call it Ethernet, and that's my word.
What did the word mean to you then?
My good fortune was to be given a problem that no one had ever had before--how
would you interconnect several computers with one at every desk? I was certainly
early if not first in trying to solve that problem. We were building the first
laser printer at that time. How could you connect the computers to each other
and connect them to the printer and then to the early Internet, the Arpanet?
Ethernet was based on packets. Data was to be delivered in packets, and the Ethernet was to be decentralized so there could be nothing in the middle that could break or be unscalable. It lay within a hierarchy of protocols, so it only had to do what it needed to do, not things that would be handled elsewhere in the protocol stack, which was a relatively new idea at the time. It was so simple, and that's one of its advantages. Another advantage was randomized retransmissions. That was based on the Aloha Network built at the University of Hawaii by Norm Abramson, a forerunner of 802.11 that had randomized retransmissions.
What's a randomized retransmission?
That means transmissions would be tried again later if they overlapped in time
and interfered with each other, which we called a collision.
Arpanet introduced packet switching in about 1969, and the Aloha brought packet switching to radio in about 1970. So two big things happened in '73: One was the invention of the Ethernet, and the other was the beginning of the development of TCP/IP (Transmission Control Protocol/Internet Protocol), which started at Stanford in the summer of '73. The Internet and Ethernet have developed in lockstep since that time. Today, most Internet packets, 99.99 percent of them, are carried on the Ethernet.
But the Internet, or the Arpanet, preceded your description of Ethernet by
a few years, and the first actual implementation of Ethernet technology by even
longer. So my question is, how were messages transmitted before Ethernet?
The Internet had a preceding set of protocols, an earlier generation designed
for a much smaller network. It was worldwide, but had a smaller number of hosts
and was not designed to be internetworked, connecting networks. So TCP/IP
greatly expanded the network space and provided for building a network of
networks, lots of LANs connected by lots of WANs.
How did the Ethernet prevail over the older protocols, and the others that
cropped up in the future?
The speech I'm going to give on Thursday is called "Ethernet vs.
Godzilla." The idea is that Ethernet has encountered a series of
competitors out trying to kill it. The number is huge. The most famous is the
IBM token ring. It was among the many Godzillas and the most formidable because
it was backed by the dominant monopoly in computing at the time.
How was the token ring different from Ethernet? And what were its
advantages and disadvantages compared with Ethernet?
It was developed by IBM. That was its principle advantage. It was organized
around the passing of a token from station to station, so computers could take
turns transmitting messages. With Ethernet, any station would transmit
immediately and hopefully avoid collisions. The token gave permission to
transmit a message. The advantage was what they called determinism, which I
considered to be a red herring.
What did they mean by it?
It meant they could guarantee that data would be delivered in a certain amount
of time. Without the token, in theory, Ethernet packet transmissions could take
forever. That was smaller than the probability of the Earth blowing up, but they
used to throw determinism at us.
So if the token ring worked with tokens, how does Ethernet work?
Any station wishing to send a packet would form the packet in memory, then
listen to the ether--the cable or the fiber or whatever--to see if there were
already a transmission ongoing. And if there were, it would wait, deferring to
the transmission ongoing, and then it would send its packet. But it was possible
that more than one station would be waiting, so then you would get a collision.
It would notice that there was interference on the cable, which we called
collision detection. As soon as it was detected, the station would abort
transmission and choose a random number and try again a few microseconds later.
The second one would defer and avoid multiple collisions.
And what happened to the token ring Godzilla?
There is a lot of token ring out there, but in the last few weeks IBM announced
a $100 million program to tear out token rings inside of IBM and replace them
with Ethernet. That's a sign that you could declare it dead. Although it's been
dead for some time, that would be the final blow. And the reason that Ethernet
won was that Ethernet was an open standard with many competing companies
providing it.
What strikes you most about what's become of Ethernet since your memo 30
years ago?
What Ethernet is today is more than a packet format or media access
algorithm--it is a business model.
How so?
I was hoping you'd ask that. There are four business models out there today. The
first is the vertical model exemplified in the 1980s by the IBM monopoly. The
second, which dominates today, is the horizontal model dominated by AOL, Cisco,
Intel and Microsoft. They are also monopolies, I might add. The third is the
Linux/open-source business model. And the fourth is the Ethernet business model.
It's based on de jure standards with proprietary implementations of those de jure standards, and it is unlike open source in that competitors don't give their intellectual property away. The competition is fierce, but there is a market ethic that products will be interoperable. And the standard evolves rapidly based on market engagement in such a way to value the installed base. There is a heavy value placed on sustaining and maintaining the installed base. That's the Ethernet business model.
So does that have to do specifically with businesses that work with
Ethernet technologies? Or is it a way of doing business modeled on the way
Ethernet has evolved from an industrial perspective?
TCP/IP protocols follow the same model. That's the business model--standards
compliant, interoperable, evolving so they are backward-compatible. It's the
model that Ethernet uses to kill all of its Godzillas.
You're now working as a venture capitalist.
My business is technology innovation, and my fourth career in that business is
as a VC. First I was an engineer, then an entrepreneur. The third time I was a
pundit, and my fourth career was as a VC. All of that is technology innovation.
I've been a VC since 1/1/1, and I'm still learning the trade. I'm with
What are the companies you've brought into the fold?
Ember is an embedded networking company in Boston using wireless mesh technology
to network the embedded processor. Only 1 or 2 percent of processors are in PCs.
Ember's working on networking the remaining 98 percent. I'm on the board of
Narad, which is trying to deliver 100 (megabytes per second) through the cable
TV system. That's not 100 kilobytes, but 100 megabytes. Sometimes we call that
true broadband, as opposed to what they're now selling. I'm on the board of a
nanotechnology company with no products yet. They're developing pretty basic
technology in hopes of putting out a product. It's Nanosys in Palo Alto, and in
Medford, Mass.
So what you call the Ethernet business model has helped Ethernet prevail
against old competitors. What about new ones?
There are three new Godzillas. Ethernet is busily battling SONET and Fibre
Channel in storage area networking business. Now 802.11 and 802.16 and 802.20
are going after the cellular telephone market using all the ideas of
Ethernet--packet switching, redistributions. That set of ideas is now in the
marketplace battling SONET, the established standard for telephone backbone
communication. Instead of using SONET, which is optimized for carrying 64kbps
telephone conversations, they're using Ethernet, which is optimized for Internet
packets. So we're seeing Ethernet proliferate.
In the storage area networking business, it's been the practice to use specialized networking Fibre Channel, and Ethernet is encroaching in that area.
Why are people opting for Ethernet in these areas?
It's cheaper, faster, more network- and Internet-compatible. And it's more
reliable.
SONET and Fibre Channel can be made to carry Internet packets, but they're not built for that. Ethernet is. Meanwhile, cell phones and 802.11 are beginning a many-year battle for how they are going to complement or compete with each other. And since 802.11 is optimized for carrying Internet traffic in an Ethernet sort of way, I'm guessing it's going to likely win in that battle.
You started 3Com here in Silicon Valley and then moved to Massachusetts.
Now 3Com is following you to Massachusetts.
It had nothing to do with me! Maybe it's part of a trend--Silicon Valley is
passe; and everyone is moving east.
Any thoughts on how things have ended up at 3Com?
"Ended up"--3Com is still a billion-dollar company and it's still
evolving and I'm proud of the company. I'm still emotionally attached to it,
though I haven't worked there since 1990. But I am rooting for them. It has good
DNA if I do say so myself, and they're working hard to restore
profitability--and they're not alone in being unprofitable. It will be fun
having 3Com headquarters in my back yard.
You were CEO of IDG's InfoWorld Publishing Company from 1992 to 1995, and
you wrote a regular column for eight years. Do you ever miss being a journalist?
Occasionally I will develop a strong opinion about something, and then I will
just yell about it to people.
What's the status of your four patents related to Ethernet?
I share four Ethernet patents--there were other people involved. The interesting
thing is that they've all expired by now. They only last 17 years.
Who's most responsible for the development of Ethernet technologies since
your initial description 30 years ago?
That's really hard. In the last 30 years, everyone's gotten involved: Cisco,
3Com, Intel, National Semiconductor, DEC, HP, Compaq, Xerox, Cabletron, Bay
Networks, Nortel--the list goes on. That's an impossible question to answer.
OK, then who's doing the most interesting Ethernet work right now?
I'd have to do some research before I answered that, and if I did answer it I'd
get in trouble. It's a pretty widespread thing. That's why Ethernet is so
successful, because it's in such widely diverse environments with so many fierce
competitors.
I asked you what the name Ethernet meant when you first described the
idea, but I didn't ask where you got the word.
In the history of physics, in the 19th century, they needed to explain how light
got from the sun to the Earth. So they theorized that there was a medium that
filled all of outer space that served as a passive medium for the propagation of
electromagnetic waves. (Albert) Michelson, (Edward) Morley, Einstein, et al,
proved there was no ether. So in 1973, on May 22, instead of referring to
putting packets into Ethernet's cable, I needed a word that was
medium-independent, and settled on the word "ether" because the ether
would be omnipresent, passive and it would serve as a medium for the propagation
of electromagnetic waves.
Some of my critics have said it was appropriate that I named Ethernet after a discredited physical argument, which was a weak counterargument. Ethernet used to be a trademark of Xerox, but in the process of standardizing it through the IEEE, Xerox gave up the right to the trademark. I've appointed myself as being in charge of deciding what's Ethernet and what's not. If they want to call 802.11 wireless "Ethernet," I'm all for it, especially because it's reminiscent of the Aloha network from which 802.11 is derived.
