People say "web" the way they say "phone," which is to say they stopped noticing the word. I have not, quite. I was around when the wires were simpler, and I still wonder how a thing that began as one machine talking to one other machine came to be named for something a spider builds in a corner.
The name is not about the spider. In March 1989, at CERN, Tim Berners-Lee wrote a proposal for a way to keep notes that could point at one another. The proposal described a system called "Mesh." In it he wrote, "This is why a 'web' of notes with links (like references) between them is far more useful than a fixed hierarchical system." The internet was already there. He was talking about documents. The next year, while writing the code, he picked "World Wide Web" over "Information Mesh" and "Mine of Information." We escaped being introduced, for the rest of our lives, as mine designers. It was a near thing.
Before the Web, there was the leash
If you had drawn many of the business connections I remember from that earlier time on a pad, you would not have drawn a web. You would have drawn a leash. One terminal, one mainframe. One PC, one server. IBM's Systems Network Architecture, announced in 1974, was a hierarchy with the mainframe at the top and everyone else asking permission. Bisync, from 1967, was a conversation on a wire in which you spoke when spoken to. Async was the Teletype habit, one character after another, with no architecture in it at all. Serial, in the old RS-232 sense, was only a way of getting the bits out of the box. None of these, sketched honestly, looked like anything a spider would claim.
“You would have drawn a leash.”
A few designs did put several machines on the same wire, and those were a different animal. Datapoint's ARCnet appeared in the 1970s. Ethernet came from Robert Metcalfe and David Boggs at Xerox in 1973, and Metcalfe later founded 3Com to sell it. IBM's Token Ring arrived as a product in 1985. Those were shared streets, not private driveways.
Novell's NetWare made a file server the center of the office. Banyan VINES, which sat on Unix and kept a directory called StreetTalk, tried to give every resource a name a person could find. Microsoft, with 3Com, came along with LAN Manager.
NetBIOS was not Microsoft's invention. Sytek built it for the IBM PC Network, and Microsoft adopted it the way a man adopts a useful coat. Wang was mostly terminals, plus its own WangNet. I used to file QNX in that same drawer. It does not belong there. QNX is a small, quick operating system, created by Dan Dodge and Gordon Bell, who founded their company in 1980, the sort of thing you put in a machine that is not allowed to woolgather. It turns up again later, where it belongs.
The packet only needed the house number
The leash had a weakness you could see on the pad. If a computer can speak only to the particular machine on the other end of its wire, cut the wire and the conversation is over. Paul Baran, at RAND in the early sixties, drew a network that could lose pieces and still carry a military order. That part of the legend is true. It is not the same story as the ARPANET. The ARPANET, from 1969, was built so researchers could use one another's computers. Sharing, not a bomb shelter. The two ideas met, which is how good ideas often meet, sideways. In 1974 Vint Cerf and Bob Kahn published the paper that became TCP/IP, a way for a computer on one kind of network to reach a computer on another, by an address, without caring what the local wires looked like. Jon Postel edited the standards that arrived in 1981. On the first of January 1983, the ARPANET switched over. After that, a packet did not have to know the road. It had to know the house number.
The Internet’s history, told by its pioneers ↗
Bridges joined pieces of a local street. Routers, which people in those days still called gateways, sent the packet toward another town. Cisco, founded in 1984, sold the boxes that did the second job. It did not invent the job, and it did not invent the mail.
A good love story. The wrong origin story.
I heard the rumor everyone heard. Two lovers, at different universities around the Bay, built a box so they could write to each other, and out of that came email. It is a fine story, and it is not the one that happened. Leonard Bosack and Sandy Lerner were married. Both were at Stanford, in departments whose networks would not speak. They set out to sell a router that would translate. The software owed a great deal to a router already written there by William Yeager. Network mail is older. Ray Tomlinson sent it at BBN in 1971. Postel's SMTP specification, RFC 821, dates to August 1982, two years before the company. The love-letter version is the sort of story a business acquires after it has become a business.
The trouble with “simple”
The academics who wrote the early protocols had a fondness for the word "simple," and for once the fondness was earned. Simple Mail Transfer Protocol. Simple Network Management Protocol. Even a Simple File Transfer Protocol, which is not the file transfer people mean now. I had SMS in that list for years. It is not in the family. SMS is the Short Message Service, a telephone invention, and "simple" was never part of the name. A man ought to say so when he has been filing a thing in the wrong drawer.
Postel’s August 1982 SMTP specification ↗
On top of the protocol came the tools, and they did not all arrive on the same afternoon. Telnet, for logging into a machine that was not in the room, was on the ARPANET by 1969. FTP, for copying a file, is Abhay Bhushan's, from 1971, before TCP existed in the form we use. Secure Shell, which replaced those trusting old logins, is Tatu Ylönen's, from 1995. Useful instruments. None of them is the network. They are what you do once the network has agreed to carry you.
Draw the picture now and it does look like a web, in the vulgar spider sense. Packets can travel over alternative routes when the network provides them. That is the network underneath; the Web itself is the linked information carried over it. We still say "web designer," "web developer," "web specialist," mostly because a man at CERN preferred that word to "Mine of Information." We still say "internet," which is only the inter-network, the thing Cerf and Kahn were writing down. A government research office paid for a great deal of the work. Researchers and engineers did the inventing, with government funding helping make the work possible. The client wanted the machines to share.
Meanwhile, at a laboratory tired of operating systems
The other half of this story is not a network. It is an operating system, and it matters because both kinds of web ended up running on it.
At Bell Labs the names are Ken Thompson, Dennis Ritchie, and Brian Kernighan. The story begins in the neighborhood of Multics, the Multiplexed Information and Computing Service, a grand project with MIT and General Electric. It was late and expensive, and in April 1969 Bell Labs left. I have heard it said that the laboratory then passed a rule: never again an operating system. I cannot find the rule. What Ritchie wrote is plainer, and better. "I am quite sure that at that time operating systems were not, for our management, an attractive area in which to support work." They had seen one large operating system. They did not wish to pay for another.
The builders went ahead anyway, which is what builders do when the room has voted. Thompson had a game, Space Travel, that had become too expensive to run on the machine they were supposed to use, something like seventy-five dollars of computer time a game. He and Ritchie moved the work onto a little-used PDP-7 at the laboratory. Out of that small, slightly improper arrangement came Unix.
Kernighan named it in 1970, and his joke is not the one people repeat in bars. "I suggested Unics for Ken's new system, because it was small and had at most one of anything. (Multi and uni are both Latin roots, so it was a very weak pun.) Someone else spelled it with the letter X; no one can remember who." I like the modest joke. One of anything. A world built by men who were officially not building an operating system.
Just a hobby, apparently
I have been lucky in the timing. I began when the first commercial versions of the C language were something you could buy, and I have sat in front of more Unix and Unix-like systems than a sensible man admits to. Xenix, QNX, AIX, HP-UX, SCO, Berkeley, System V in its various releases. Cousins who did not always speak at dinner.
Then, in August 1991, a student in Finland posted a note that deserves to be left as he wrote it. "I'm doing a (free) operating system (just a hobby, won't be big and professional like gnu) for 386(486) AT clones." That was Linus Torvalds. That hobby grew into the Linux kernel: underneath Android, widely used on servers, and central to modern supercomputing.
It is open in the practical sense. You may read it, change it, and pass it on. It is not a gift with no string attached. The kernel is licensed under the GNU General Public License version 2. Distributing a modified kernel brings obligations to provide its corresponding source under that license. To my mind, that bargain mattered as much as any single feature in making it an industry standard. I believe the present wave of machine intelligence would have had a much harder time if the world had not settled on a system anyone could inspect. I do not claim it is the whole internet. I claim it is one of the foundations much of the internet stands on.
Quick and dirty gets a desk
Windows is still here, and it earned the desk. The early editions were a coat of paint on DOS. The consumer line kept its DOS roots through Windows Me. Windows NT, in 1993, was a different building, and the Windows people use now descends from that, not from the paint. The paint's foundation has its own ridiculous name, and this one is not a myth. In 1980 Tim Paterson, at Seattle Computer Products, needed an operating system because Digital Research's CP/M-86 was late. He wrote a stopgap and called it QDOS, the Quick and Dirty Operating System. It became 86-DOS. Microsoft licensed it in 1980, then bought the rights in 1981, and the world received MS-DOS. Quick and dirty. The name was accurate. Its descendants occupied a remarkable number of desks.
The man who deserved more than a footnote
The man I want at the front of that story is Gary Kildall. He is my hero, and not because of the BIOS, though the BIOS is a large part of why any of this works. He kept building the things a small computer was going to need, usually before the room believed a small computer was going to need them.
He wrote PL/M, one of the first high-level languages for the Intel processors, so the serious work could come up out of raw assembly. CP/M began as a chore on that project. He wanted the language to live on the machine itself, which meant the machine had to speak to a disk, and the file system he wrote for the conversation turned out to be an operating system. Control Program for Microcomputers, 1974, in Pacific Grove. With Glenn Ewing he pulled the hardware-dependent pieces into a BIOS, so a designer could fit the same system to a new computer in a couple of hours.
He followed it with MP/M, so more than one person, and more than one program, could use the machine at once. Digital Research went on to GEM, a graphical desktop, while the ordinary office computer still answered with a prompt and a sense of hope. He wrote compilers that were not supposed to fit on a microcomputer, among them a PL/I compiler, and in 1985, through the company that became KnowledgeSet, he set out to put Grolier's encyclopedia on a CD-ROM, when that disc was still a rumor you could hold.
Paterson's stopgap was written to look familiar to anyone who already knew CP/M, which is a compliment even when it is also a taking.
IBM did come to Pacific Grove. Kildall had flown his own plane to Oakland that morning to see a customer, and he was back for the technical talk, as planned. The afternoon ended in a standoff. IBM wanted to buy the system outright. Digital Research wanted a royalty on each copy, so the business it already had would not be signed away. Kildall called back the next week. IBM had stopped answering. What the world was later told, in a line attributed to Bill Gates, was that Gary had been out flying. The cheaper system, built on the bought stopgap, went onto the desks.
Kildall had made the original, and a good deal besides. He received the footnote. I have never been able to see that as a small role.
The IBM meeting narrative follows David Laws’s account for the Computer History Museum. Accounts of the meeting differ. Also explore Gary Kildall in his own words.
A leash becomes a web. A laboratory that does not want another operating system gets Unix, named for having one of whatever the last project had too many of. A hobby kernel becomes the floor under the modern network. A quick and dirty stopgap becomes the thing Windows once stood on. And a man looking for a word chooses "web," and we have been introducing ourselves with it ever since. He was talking about notes. He described the century anyway.
Sources & further wandering
Names in the story link to Wikipedia for background. Historical checks below use original accounts, standards, and institutional collections.
- W3C: the Web’s early timeline and alternative names
- Brian Kernighan on the name Unics
- QNX’s own history
- Internet Society: ARPANET, packet switching, and TCP/IP
- Dennis Ritchie: the evolution of Unix
Photograph and screenshot credits appear with each image. Portraits depict later moments in their subjects’ lives, not reenactments of the events described.