Unix History, Episode 2 of the series. Previously: The Rise of Unix — and the Seeds of Its Fall, covering 1965–1983. In this episode: the Unix Wars, 1984–1994.

A Babel of Unixes

By 1984 there were something like 100,000 Unix installations worldwide, running on everything from microprocessors to mainframes. Through organic growth and a strange quirk of legal history, Unix had enmeshed itself deep into the computer industry.

The trouble with uncurated organic growth is what it grows into: a sprawling ecosystem of mutually unintelligible operating systems. A Babel of Unixes.

And then its creator, AT&T, decided to leverage this magnificent thing to make its way into the computer business.

Those efforts were chaotic. This is a chronicle of what we call the Unix Wars.

The Breakup

In 1974, the United States government sued AT&T once again — this time over the way it produced telephone equipment for its own telephone monopoly. AT&T’s eventual answer was extraordinary: it proposed to break itself up.

The agreement was signed in 1982 and took effect on 1 January 1984, when the Bell System formally divested its local operating companies. Ma Bell was dead.

Crucially, the settlement replaced the 1956 Consent Decree — the one that had barred AT&T from the computer business. AT&T had long wanted into “enhanced services”: communications offerings that go beyond merely moving bits, that get involved with the content of the data. Now it could.

As the plan percolated, AT&T had already begun turning Unix into a commercial proposition. Version 6, released in 1975, was the first version widely licensed outside Bell Labs, and AT&T began selling commercial licences for it through Western Electric: roughly $20,000 for the first machine, and reportedly around $8,000 for each additional one.

The story goes that the price was set by an AT&T manager who had real work to do and picked a number high enough that nobody would bother him about it again. If so, the plan backfired: people paid. Unix had found a real market.

Fragmentation

Two trends drove that market.

Programmers who had grown up on BSD. Students used the Berkeley environment at university, liked it, and wanted to keep using it after graduation. That put Unix’s centre of gravity squarely in engineering, scientific and technical computing.

Cheap, powerful workstations. Machines like Sun’s put Unix at the heart of the product.

The first companies to service this demand were porting houses — outfits whose business was simply getting Unix running on somebody else’s hardware. UniSoft and the Santa Cruz Operation are the classic examples.

Then the big vendors joined: DEC, Data General, IBM and others shipped Unix for their own machines, each with its own modifications, because differentiation is what hardware vendors do.

With Unix’s own creator largely absent from the market, the result was a mess. Variants shared the overwhelming majority of their code — as much as 90 % by contemporary estimates — and still an application written for one would not simply run on the others. There was no leader. Many variants weren’t even allowed to call themselves Unix, because AT&T owned the trademark, so they were Unix-like things with invented names. Users and the market alike were thoroughly confused.

System V

In 1982, AT&T’s Unix Support Group consolidated its various internal versions into a single product: Unix System III, the first release aimed squarely at outside customers.

A handful of additional features — several borrowed from BSD — turned System III into System V, announced as a commercial product in January 1983, with an installed base of around 45,000 systems. Reports from the time put the commercial source licence at about $43,000, more with add-ons. Unix got its own home inside AT&T, first under AT&T Information Systems and eventually, in 1989–1991, as the subsidiary Unix System Laboratories (USL).

AT&T’s ambition was simple: make System V the standard Unix, adopted by everyone.

The first clash: System V versus BSD

So by 1983 the Unix tree had two great branches. System V, with AT&T’s official backing. And BSD, dominant in universities and the foundation of a long list of commercial systems — that same year, Berkeley’s CSRG shipped 4.2BSD with its TCP/IP networking stack.

Hardware makers and users had to pick a side. IBM (AIX) and Hewlett-Packard (HP-UX) built on System V. Sun extended BSD — the 4.2 line Bill Joy had worked on — for SunOS.

And the two really were different. Early System V had no Internet networking, no Berkeley Fast File System (a substantial improvement over the old one), and none of the virtual memory work that made BSD such a good fit for 32-bit VAX minicomputers.

It was also a clash of cultures: AT&T as evil big business, the Berkeley crowd as ragtag underdog rebels. The famous Mt. Xinu posters are a pretty good artifact of how that felt from the Berkeley side.

But there was a structural catch. BSD was, legally speaking, a set of very helpful extensions to AT&T code. To use BSD you first needed an expensive AT&T source licence. That was a serious obstacle to commercial adoption, and through the 1980s — as System V absorbed BSD’s best features one by one — BSD vendors and users drifted to System V. If you had to buy the AT&T licence regardless, why not?

Collateral damage in the classroom

AT&T’s tightening grip on those licences had consequences nobody at AT&T seems to have weighed. Unix had been a superb teaching tool: professors taught operating systems by walking students through real, working source code. The Version 7 licence of 1979 put an end to that, forbidding classroom use of the source. Courses were purged of Unix internals and became markedly more theoretical.

Textbooks were caught in the same net, because they contained source. The casualty everybody remembers is John Lions’ “Commentary on UNIX 6th Edition” — a genuinely beloved book, and for years an underground photocopied one.

The workaround shaped everything that followed. Andrew Tanenbaum wrote Minix from scratch in 1987 precisely so he could teach and publish a Unix-like system without AT&T’s permission. Hold that thought.

AT&T’s computer business

On 27 March 1984, three months after divestiture took effect, AT&T formally entered the computer business — six models of its 3B line, from desktop multi-user “supermicros” up to a “supermini”, aimed at customers who wrote their own applications. Unix was to be a key asset in the push. As a report from the day put it:

AT&T for the past year has been pushing to have UNIX adopted as an industry standard and today it claimed that it has been “very successful.”

It did not go well.

Ordinary users found AT&T’s Unix PCs hard to use next to MS-DOS on an IBM PC. And application developers found Unix hard to target, precisely because of the variants and their maddening small differences. Why not write for Microsoft, where the ecosystem was larger and far more consistent? By 1985 MS-DOS had roughly five times as many applications as Unix. Without applications, customers stayed away. Chicken, meet egg.

The reckoning came in 1986. AT&T’s computer operations lost somewhere between $700 million and $1 billion that year, and the company took a multi-billion dollar restructuring charge against its fourth quarter, announcing the elimination of some 24,000 jobs — at the time possibly the largest single workforce reduction in American corporate history. Plants closed, part of the PC business went to Olivetti, and the strategy shifted toward selling computer systems to the telecommunications industry.

If Unix was going to save AT&T’s computer business, that sprawling family tree had to be unified. But how?

That is when Sun came knocking.

AT&T and Sun

In 1987, AT&T and Sun joined forces. Sun was by then the largest Unix vendor, with an OS built on 4.2BSD, and both parties believed that together they could unify the platform into something formidable.

By October 1987 they had a deal. AT&T took an option on 20 % of Sun and a board seat. The two would merge their Unix versions — plus Microsoft’s Xenix — into System V Release 4. Sun would ship a new OS based on it, and would get preferential treatment as AT&T built the software out. (SVR4 actually shipped in 1989.)

The two companies pitched this as an open technical standard, available to the whole industry. AT&T, by its own account, wanted Sun’s marketing heft to complement its technical chops.

The Hamilton Group

The rest of the industry did not read it that way.

For DEC, IBM, HP and the others, Unix was a public utility: a common foundation on which each vendor layered its own value-adds. AT&T’s move looked like a plan to sweep those value-adds away and impose a single standard from above.

Worse, that standard would arrive hand-in-hand with a direct competitor. Shortly before the AT&T announcement, Sun had unveiled its own SPARC RISC processor. If Unix became the industry standard, and the standard Unix ran best on SPARC, Sun would have a structural advantage over everybody.

In early January 1988, representatives from Apollo, DEC, Hewlett-Packard, Honeywell-Bull, Gould, MIPS, NCR, Silicon Graphics, UniSoft, Unisys and others met at DEC’s Western Research Laboratory in Palo Alto. The lab sat on Hamilton Avenue, so they became the Hamilton Group.

They first tried talking to the man running AT&T’s computer efforts, Vittorio Cassoni. AT&T, in effect, told them to pound sand.

So in May 1988 they brought in IBM and founded the Open Software Foundation (OSF) — Apollo, DEC, HP, IBM, Groupe Bull, Nixdorf and Siemens. OSF would pool its members’ Unix technology and work toward an AT&T-free Unix.

The alliance was, by the standards of the era, astonishing: it put DEC and IBM on the same side. At the announcement, a Wall Street Journal reporter noted that neither he nor anyone else present could recall DEC co-founder Ken Olsen ever sharing a stage with an IBM executive.

Ken Thompson heard about it while travelling in Australia, turned to Dennis Ritchie and said: “Just think, IBM and DEC in one room. We did it!”

OSF versus Unix International

Sun’s CEO Scott McNealy was less delighted. He renamed OSF the “Oppose Sun Forever” foundation.

Sun and AT&T responded by founding Unix International later in 1988, with Intel, Toshiba, Unisys, Motorola and others signing up to build products around System V. Between them, the two camps eventually enlisted a couple of hundred members, and OSF cheerfully co-opted Sun’s “open” language, complete with appeals to liberty and justice.

OSF’s reality struggled to live up to the rhetoric. Its members were, Apollo aside, largely mainframe companies — and competitors. Herding cats, with politics.

The technology base was a case in point. OSF initially announced it would build on IBM’s AIX, reportedly one of IBM’s conditions for joining: a product designed by committee before a line of it was written. By 1989 the concrete deliverable was a rather good user interface, OSF/Motif, and not much else.

Then economics intervened. The mainframe market was shrinking. HP bought Apollo and circled Bull. DEC, IBM and others were losing money on computers.

OSF did eventually deliver: OSF/1 shipped in January 1992, built not on AIX but on Carnegie Mellon’s Mach kernel with BSD components. Its adoption was narrow, though — mostly DEC, which shipped it on Alpha as Digital UNIX and later Tru64. In 1994, with AT&T out of the Unix business and Unix International disbanded, OSF stopped funding OSF/1 development altogether. In 1996 OSF merged with X/Open to form The Open Group, which is where the Unix trademark lives to this day.

Sun, for its part, did ship: Solaris 2.0, built on SVR4, arrived in 1992 and turned out to be a solid product. Sun even adopted Motif in the end.

BSD splits from AT&T

By the late 1980s BSD was no longer commercially viable as an operating system. But the code still had enormous value — and AT&T’s licence fees kept climbing. A commercial source licence had cost about $20,000 in the 1970s; by 1988 it was in the neighbourhood of $100,000, and by 1993 approaching $200,000.

That created a specific, painful problem. Vendors who wanted to build standalone TCP/IP networking products for the PC market — using code Berkeley had written, not AT&T — still had to pay AT&T per binary. So they asked Berkeley to separate the networking code from anything AT&T owned.

That work shipped in June 1989 as Networking Release 1 (Net/1): the first freely redistributable code out of Berkeley.

Keith Bostic at the CSRG then pushed for the full version: purge all AT&T-derived code. He pioneered mass net-based development, soliciting volunteers to rewrite the Unix utilities from scratch against their published descriptions alone — a clean-room reimplementation. Eighteen months later, in June 1991, Networking Release 2 (Net/2) was done.

They very nearly made it. When the dust settled, exactly six kernel files remained contaminated and too awkward to rewrite trivially.

The lawsuit

Aware they had something interesting, several CSRG people went commercial. Berkeley Software Design, Inc. (BSDi) was founded in 1991 by Rick Adams together with Keith Bostic, Kirk McKusick, Mike Karels, Bill Jolitz and Donn Seeley.

With the university’s blessing, BSDi began selling BSD/386 — Net/2 with the six missing files reimplemented — in January 1992, source and binaries, for $995. (Bill Jolitz left BSDi and, with Lynne Jolitz, released a separate, free Net/2-derived system for PCs called 386BSD in March 1992. The two are constantly confused; only BSDi’s product was ever at issue in court.)

BSDi advertised a 99 % discount on the price of a System V source-plus-binary licence, and told readers this was a product that did substantially the same thing. The phone number in the ad was 1-800-ITS-UNIX.

AT&T did not care for that. It asked BSDi to change the marketing, and BSDi did.

USL sued anyway, in April 1992, in federal court in New Jersey, seeking to halt distribution on claims of breach of contract, copyright infringement, trademark dilution and trade-secret misappropriation.

As always, it dragged. BSDi’s answer was elegant: fine, the six files are on us — but the other 18,000 files in the system were already being distributed by the University of California, so take it up with them. The judge agreed and told USL to amend its complaint. Six measly files being a thin basis for a landmark case, USL refiled and added the Regents of the University of California, now seeking to stop Net/2 as well.

In December 1992, Judge Dickinson R. Debevoise heard arguments. He denied the preliminary injunction and threw out all but two of the claims — and made an observation that must have been chilling in Basking Ridge: AT&T had probably lost its copyright in 32V altogether, having distributed the system in 1978 without a copyright notice and only registering it in 1992.

The University then counter-sued in California state court, on the grounds that USL had failed to give Berkeley the credit for BSD code in System V that its own licence required. Suggested remedies included apology advertisements in major publications.

The whole affair was covered exhaustively at the time — which, for a technology whose main selling point was neutrality, was not helpful.

The wars end

In December 1992, AT&T agreed to sell USL to Novell, the Utah software company that had pioneered the PC local area network with NetWare; the deal closed in July 1993. Novell wanted Unix integrated with NetWare.

Novell’s CEO Ray Noorda — not a founder of the company, but the executive who made it what it was — said he would rather compete in the marketplace than in the courtroom. USL and the University stepped back, and a settlement was reached in January/February 1994, largely in Berkeley’s favour: of some 18,000 files, three were removed, about 70 gained AT&T copyright notices, and USL agreed not to sue users of the resulting system. 4.4BSD-Lite followed in June 1994.

Selling Unix to Novell put the first damper on the Unix Wars. But the chaos had already cost the industry its window — as I mentioned in the episode on Sun, it let Windows NT invade Unix’s turf in workstations and higher-end computing.

Windows NT 3.1, released on 27 July 1993, was Microsoft’s “business-oriented” operating system: 32-bit, properly multi-user, networked, with file sharing. Its design lineage runs through DEC’s VMS rather than Unix — Dave Cutler and his team came from Digital — and its POSIX subsystem existed largely to satisfy US government procurement checklists. But it did the thing Unix vendors had spent a decade failing to do: offer developers one consistent target with Unix-class capabilities. Then Microsoft pointed its marketing machine at it.

The threat did what a decade of standards diplomacy could not. In March 1993, members of both OSF and Unix International — HP, IBM, SCO, Sun, Univel and USL — announced the Common Open Software Environment (COSE) to build a genuinely common environment across the fractured Unix space. COSE’s “Spec 1170” work was handed to X/Open, and Novell transferred the UNIX trademark and the Single UNIX Specification to X/Open in late 1993, so that conforming systems could finally earn the brand.

The wars were over. There was rather less left to win than there had been in 1987.

GNU/Linux

The other beneficiary of the Unix Wars was Linux — or GNU/Linux; there is a naming controversy and I’m going to leave it aside.

In September 1983, Richard Stallman, a programmer at the MIT AI Lab (not, as often claimed, a professor), announced the GNU project, with development starting in January 1984. GNU’s goal was a complete operating system that was free software; the decision to make it Unix-compatible was pragmatic, since Unix’s interfaces were what programmers already knew. Compatibility with the IEEE’s POSIX standards — the API specifications later adopted by the US government for procurement — became a target once POSIX itself existed, from 1988 onward.

By the early 1990s GNU had a great deal of excellent software and no kernel. Which is a problem, the kernel being the part that manages memory, processes, disk space and all the other unglamorous essentials.

The kernel arrived from an unexpected direction: Linus Torvalds, a Finnish undergraduate at the University of Helsinki, inspired by Minix — the very system Tanenbaum had written to escape AT&T’s teaching restrictions. In August 1991 Torvalds posted about his project in the Minix newsgroup, and contributors came straight out of that ecosystem. Kernel plus GNU userland made a usable system; the first distributions appeared in 1992, and Linux 1.0 followed in 1994.

Linux borrowed Unix’s method — volunteers worldwide working a common code base — with one decisive difference: it had a natural leader. Torvalds acting as curator kept the project from fragmenting the way Unix had, and let it ship fixes and updates fast.

It also settled its licensing question early, and permanently. The very first Linux release actually forbade commercial distribution; Torvalds relicensed the kernel under the GNU GPL in early 1992, and that was that. There would be no repeat of a vendor changing the terms of the deal once the platform got valuable.

Over time Linux overtook Unix as the primary “open” operating system, especially for scientific and technical work — and it made particularly good gains during the two years AT&T and BSDi spent in court, when the free Unix with the clear legal status was the one that wasn’t being sued.

Conclusion

The Unix Wars were wars of conquest, and the conquest failed.

I have some sympathy for AT&T here. They must have felt they’d hit the jackpot: this thing they thought they owned had become one of the foundations of a thriving industry.

But their attempts to monetise it are a fine illustration of a line attributed to Bill Gates:

“A platform is when the economic value of everybody that uses it exceeds the value of the company that creates it. Then it’s a platform.”

Unix the platform created enormous value for the community using it, and AT&T — not unfairly — felt entitled to a larger share of it as the platform’s creator. But collecting that share meant centralising and organising a delicate community of users.

Those users had alternatives. When pressed, they left for greener grass.

Unix is still around, of course, and the branches have famous descendants: Oracle Solaris, HP-UX, AIX, FreeBSD, OpenBSD, NetBSD. And macOS and iOS descend from NeXTSTEP, which was Mach plus 4.3BSD — macOS isn’t merely Unix-like either, it’s a certified UNIX, carrying the very brand AT&T spent the 1980s trying to own.

Seen that way, Unix’s family exceeded everyone’s expectations. But the Unix Wars still leave you wondering what might have been.

Next episode: to be decided — candidates include the rise of Linux in the enterprise, or the long afterlife of the BSD family.

Sources and notes

  • Marshall Kirk McKusick, Twenty Years of Berkeley Unix: From AT&T-Owned to Freely Redistributable — the primary account of Net/1, Net/2, the six encumbered files, BSDi and the lawsuit.
  • UNIX System Laboratories, Inc. v. Berkeley Software Design, Inc. and Judge Debevoise’s ruling on the 32V copyright.
  • The Open Group’s UNIX history timeline for installed-base figures (100,000 installations in 1984; 45,000 at the System V announcement; 1.2 million by SVR4).
  • Klara Systems, UNIX Wars — The Battle for Standards, on the Hamilton Group and the OSF/UI standoff.
  • Contemporary press coverage of AT&T’s 27 March 1984 computer announcement and the 1986 restructuring.

Corrections to the commonly told version of this story:

  • The Bell System breakup was agreed in 1982 but took effect on 1 January 1984, not in 1983.
  • BSDi’s product was BSD/386 (later BSD/OS). 386BSD was the Jolitzes' separate free system and was never party to the lawsuit.
  • OSF/1 did ship — January 1992, on Mach rather than AIX. OSF didn’t abandon the goal so much as deliver a product almost nobody adopted, then stop funding it in 1994.
  • Ray Noorda was not a Novell founder; he became president and CEO in 1983.
  • Richard Stallman was a programmer at the MIT AI Lab, never an MIT professor, and GNU predates POSIX — POSIX conformance became a goal later, not the original specification GNU was written against.
  • Linux was not openly licensed from day one: the original 1991 licence barred commercial distribution, and the GPL relicensing came in early 1992.
  • Windows NT’s design came from DEC’s VMS lineage via Dave Cutler, not from Unix; what NT took from Unix was the market, not the architecture.

Figures I could not corroborate and have hedged in the text accordingly: the $8,000-per-additional-machine Version 6 fee; the anecdote about the $20,000 price being chosen to keep people from asking questions; the $43,000 System V source licence and its announcement at the Unicom conference in San Diego; and the “MS-DOS had five times as many applications” comparison for 1985. AT&T’s 1986 numbers vary by announcement — the widely reported figures are computer-business losses of $700 million to $1 billion, a multi-billion dollar restructuring charge, and 24,000 job cuts (later announcements said 27,400).