Who Killed OS/2?
How Big Blue Blew It, Though Redmond Helped
by
September 1997 To Boldly Go . . .
Lee Reiswig stood in the wing, stage right, about to make his Broadway debut.
Onstage was Kate Mulgrew, the star of one of the new Star Trek spin-offs but today the leadoff cheerleader for a new IBM software product called "OS/2 Warp."
It was a cute idea, really. The software developers in Boca Raton had taken to giving Star Trek-based codenames to prerelease "beta" versions of OS/2 and related products. One of these was "warp," which on the television show meant the speed of light or something. The product being introduced this October morning in 1994 was a graphical operating system for personal computers; its chief competition was the notoriously slow Microsoft Windows. Associating the new IBM offering with the speed of light would stick it to Microsoft a little.
There was the other Star Trek connection, the irony. The original television show had been doomed to cancellation after just one season. But viewers, more like cultists or adherents than mere fans, bombarded NBC with letters, calls, and other demonstrations of their devotion, and Star Trek remained alive for a couple of seasons more.
Likewise, OS/2, an advanced operating system for IBM-standard personal computers, had after early incarnations fallen into some disfavor within the company. But for the loud campaign of IBM employees -- many of whom were not involved in its development or marketing, but who simply liked it -- and a few true believers on the outside who were somehow privy to goings on within Big Blue, the whole thing might have been dropped or at least taken in a direction that would interest few home and small office computer users. This vociferous fanclub called itself "Team OS/2." Their fervor would have given the most devout Trekkie a run for his money, but for the fact that a lot of Team OS/2 members were Trekkies in much the same way that most Ross Perot supporters believe in UFOs.
Now, at the Richard Rodgers Theatre in New York City, hundreds and hundreds of fans were assembled for what could have been either a Star Trek convention or an OS/2 revival meeting. No one was recorded as having worn the pointed plastic ears of the Vulcan, Mr. Spock, but if anyone had it wouldn't necessarily have been noted. These people had already been using a version of OS/2 that had been available for more than two years, and they loved it. More than that, they were deeply suspicious of anyone who didn't love it.
Also present -- though they did not make their witness in the articles they would write later -- were members of the computer press. They had uniformly opined that OS/2 was both superior to anything else available and doomed to failure. The Warp Trekkies were at minimum highly distrustful of and sometimes openly hostile toward the computer press, which in turn looked upon the greater body of OS/2 users as potentially dangerous zealots.
The audience, then, was divided into two camps, each contemptuous of the other. Meanwhile, loudspeakers blared "Time Is On My Side," "Time Has Come Today," and every other once-popular song in which the word "time" somewhere appeared. The idea was to drive home the fact that Microsoft Corporation kept promising delivery of a new, gee-whiz version of Windows that performed all manner of miracles (miracles, it happened, that OS/2 had been performing for years), and kept moving back the release date -- it was already more than a year overdue. OS/2 was here, now, or so the music was supposed to announce. At stage center was a huge hourglass, the very symbol of the slowness of Windows in performing tasks and of Microsoft's slowness in delivering "Chicago," which would be released the following year as Windows 95.
Reiswig, head of IBM's Personal Software Products division, had been with the company since 1966, when he was a 20-year-old trainee. The IBM in which he had grown up was the company of legend, of white shirts, of refusal to comment on products under development, of dealing with corporate computer departments instead of individual users. Conducting a product rollout before a crowd like this one, with an opening act like this one, at a location like this one was a deviation from -- no, it was flinging down and dancing upon -- the dignity of Mr. Tom Watson's computer company. But never mind. Maybe it would work.
To thunderous cheers from the OS/2 Trekkies and polite applause from the computer press, Reiswig walked out onto the stage. He did not seem comfortable, not as he always did when he would drop in from time to time on meetings of local user groups just to talk about what was going on with OS/2. At those times, he was simply amazing, relaxed and confident, the boss of a big piece of a very big corporation who just happened by in Topsiders and khakis and a sweater to bring the few dozen people there up to date. Today, in suit and tie and IBM-white shirt, he seemed almost at the point of cultural meltdown. He had talked happily with OS/2 users before, and he had delivered the kind of insomnia-curing talks that frequently take place in the computer industry. He seemed unsure which this was supposed to be, and at IBM such things are noticed. IBM, after all, was not Apple, where a person was said to be well dressed if he wore shoes to work.
A series of presentations then unfolded, more or less in keeping with the kind of show-biz extravaganza this was apparently supposed to be. David Barnes, OS/2's very best showman, appeared by satellite link and on tape at various locations around the world, apparently demonstrating that OS/2 would work no matter what a country's standard electrical current was. Other IBMers made short, unmemorable speeches about all the computer companies that would be sending out their new machines with OS/2 on them, and all the fine applications for OS/2 that were just weeks away from release.
Then, finally, the surprise. A frenzied, even confused multimedia show, on screens, with flashing colored lights and slides and booming sound, then the orgasmic conclusion: The giant hourglass exploded, showering the audience with little Mylar rectangles that had pro-OS/2 sayings printed on them, fortune-cookie style. The crowd, at least the ones who knew who Captain Kirk was, went wild. Windows is dead! Long live OS/2!
Lunch followed at the nearby Marriott Marquis. Vendors who promised new programs for OS/2 were there. The audience from the Richard Rodgers Theatre was there. Press briefings were held there through the afternoon.
Maybe this would do it. Maybe the world would finally notice IBM's unquestionably superior operating system. Maybe this would work. Nothing else had.
The first review was troubling: The New York Times did mention the release of the new version of OS/2 -- in a three-inch, one-column wire story on page 5 of the business section the following day. This would be in stark contrast to the coverage ten months later when Microsoft finally delivered its long-awaited new version of Windows, by then called Windows 95. Four weeks before the Windows release, the Times devoted page after page to speculation as to its significance, indulging in an orgy of Microsoft worship that made it seem as if the importance of Creation was that it made Windows 95 possible. (Among the Times's many speculations that day, July 31, 1995, was that IBM would pull the plug on OS/2 the following January which, in fact, did not happen; the correction still has not appeared).
Still, the rollout was seen as a success, if not by the Times then by others, breathing new life into what everyone agreed was a troubled product. Reiswig could catch his breath.
Or so he thought.
The story, which may be apocryphal, has it that the discovery was made at Reiswig's own home, as he upgraded a computer there to the new release. The installation seemed at first to go perfectly smoothly. OS/2's weakest link in earlier versions had been its overly complicated installation procedure, but that had all been fixed. New users would have no problem and old users would be able to upgrade right over the existing program, keeping whatever customizations they had made. Because OS/2 is tremendously reliant on a file called "config.sys," smart users always made a copy of that file, usually calling it "config.bak." Should config.sys get corrupted, it was a small matter to rename config.bak to config.sys. All would then be well.
Unless one were installing the new version of OS/2 Warp that had been released that very day. The installation procedure somehow latched on to any file (in the root directory of the boot drive, which is where config.sys lived) named "config.bak" and mixed it in with config.sys, resulting in a computer that wouldn't even boot all the way. According to the story passed around in IBM and OS/2 circles at the time, this discovery was made that night by Reiswig himself when the computer halted mid-boot.
Whether the discovery took place in Reiswig's home or not, the bug was real. It meant that every single copy of the product, including those proudly handed out that day to members of the computer press (the Trekkies would have to buy their own), would have to be recalled.
It was a monstrous screwup. To this day no one seems to know how something so obvious could have gone unnoticed over the months of testing that so vast a program undergoes before it is offered for sale.
However it happened, what should have been OS/2 Warp's day of triumph was followed by days of frantic action, embarrassment, and not wholly successful attempts at damage control.
Disaster by design
Almost from the beginning, OS/2 had had terrible luck. If no one inside IBM could be found to mishandle the product at some crucial juncture, the operating system had plenty of enemies on the outside willing to fill the gap.
"I was always a fan of OS/2, once I got acquainted with its capabilities," says John C. Dvorak, longtime <i>PC Magazine</i> columnist, "and so it was something of a disappointment that as good as the product was in its heyday, it didn't catch on for a number of reasons. It turns out to be a web of intrigue. The reasons for its failure are not singular, but a complex matrix, and I would put Microsoft -- and IBM -- at the top of the list."
An operating system is, in its strictest definition, a program that prepares the computer hardware for the demands that will be placed on it by the applications -- the word processors, databases, and spreadsheets, as well as games, graphics programs, and communications programs -- that will run on it. An operating system has what is called an application programming interface, or API, which is a set of common commands that application programmers must use in order to get in touch with the operating system and, through it, the hardware.
The IBM Personal Computer got its name not because every person was expected to have one but because it represented a considerable departure from the way business computing had been done. Until that time, computer users in businesses used "terminals," which were not computers themselves but which were connected by cables to computers, located in "the computer room" or even off site entirely. A "personal" computer was one whose computing power was right there, on your desk. No cable was needed except one to the electrical outlet. IBM never expected its PC to become a common household appliance. The company's middle name is "Business," and it expected to deal with businesses -- the ones about whom the saying was, "Nobody ever got fired for buying IBM."
"There were some IBM product planners who had a forecast that for the life of the industry there would be 250,000 PCs sold -- during the entire life of the industry," remembers John Soyring, director of worldwide technical projects for IBM Personal Software Products. "Obviously, they got that wrong."
Rather than start from scratch, IBM developers in Boca Raton, Florida, were able to employ off-the-shelf components from outside vendors, thereby greatly reducing development time and resulting in "open architecture," which is to say a design that was not proprietary and that therefore could be copied by others. Introduction of the machine was in sight, but there was still no operating system for it. The story is now legendary about how two guys in Seattle named Bill Gates and Paul Allen, who had made a nice little business for themselves writing high-level BASIC interpreters for microcomputers, snagged the deal. They bought something called "QDOS," for Quick and Dirty Operating System, from a local company, fiddled with it a bit, and licensed it to IBM for use as its PC operating system. They had named their company "Microsoft."
The fact is, PC-DOS -- MS-DOS when licensed to other computer makers -- was never really an operating system. Applications that ran under it dealt directly with memory, with peripheral devices such as printers, video, mice, and modems. A poorly-written application could bring down the whole system. While there were a few programs, called "Terminate and Stay Resident" or TSR programs, that could more or less run while other applications were operating, this was a chancy thing involving considerable user alchemy and no guarantee of success. This was as far as "multitasking" would get, as long as the processor was Intel's 8088 chip, which was not designed to do more than one thing at a time anyway.
But microprocessors were destined to become more complex and powerful, and it was clear that a real operating system, one that provided an absolute buffer between hardware and application software, would be needed.
IBM had a problem here. Its business was built on mainframe computers and their smaller brethren, minicomputers. It did not wish to cut into that business by producing microcomputers -- PCs -- that could do the things that otherwise required more expensive machines. It was to IBM's overall company advantage to keep PCs, well, functionally challenged. But because the PC was "open architecture" and because no other computer company had an interest in maintaining IBM's mainframe computer business, IBM's attempts to put the brakes on PC development slowed only the development of PCs made by IBM. Soon Compaq and others were cleaning IBM's clockspeed, putting the hottest chips that Intel made into new, high-quality computers. It was the first time that IBM faced a major conflict over the PC within its various businesses. It was far from the last.
Under its agreement with IBM, Microsoft could license its software to other computer makers. The company did just that, providing not just MS-DOS but a growing group of "productivity applications" -- programs that do something useful -- led by its word processor, Word. BASIC was made a part of MS-DOS.
In the late summer of 1984, IBM announced its PC-AT, a computer that differed from the original PC in that it was powered by the Intel 80286 processor. The advantage of this chip was that, given the right software, it could run several programs at once. The disadvantage was that the right software didn't exist, and there was no easy way to convert existing programs into 80286 programs. The result was in effect a faster, far more expensive version of the 8088, the original PC chip. Backward compatibility -- the ability for the new computer to run old programs -- was essential. So was a way to make use of the 286's powerful new features.
What followed was a period of high-stakes confusion: A company buyer who purchased hundreds of ATs in anticipation of a solution being found to the software problem risked being stuck if no solution were forthcoming. For the first time, it was possible to be fired for buying IBM, yet buyers took the risk, to the tune of hundreds of thousands of units. In response, IBM promised an operating system that would take advantage of the 286. Microsoft would write it.
To make things worse, Intel came out less than a year later with the 80386, which offered a far more elegant way of running more than one program at once. If an operating system were properly written, it would even be possible with a 386 to run multiple DOS applications at once, no modification necessary. Each program would be led by the operating system to believe that it was running by itself on its own computer. There still was no such operating system, but the architecture of the 80386 was such that it seemed unlikely that new versions of it would be different in any important way. The road ahead was a little straighter than it had been before.
In retrospection, the answer seems obvious: Develop for the 386 and forget the 286 and the 8088. But there were now millions of 8088s and 80286s out there, with the IBM logo on their front and with owners who wanted assurance that they hadn't bought into a dead end. (They had, in fact, done just that. IBM, to its credit, remained loyal to the companies and buyers who had been loyal to it, and indeed remains so, even now from time to time introducing new and more powerful versions of plain old DOS.)
Microsoft found it all but impossible to develop a useful multitasking operating system for the 286. This was not Microsoft's fault -- the design of the chip simply wouldn't allow much useful to be done with it. Indeed, had there never been an 80286, the computing world would be a different place today, and the leading players would probably be different. As it is, the 286, in its only meaningful act of multitasking, at once tripped up IBM and catapulted Microsoft into a whole new world of opportunity.
In 1987, a full year after Compaq had begun shipping desktop machines using the 80386 chip, IBM announced its long-awaited line of PS/2 computers and the new operating system. The line inexplicably included 8088 and 80286 models as well as a few high-end, very expensive 386s. It also included a new system architecture, called Micro Channel. Peripheral cards -- modems, video adapters, sound cards, network cards -- that worked in systems using the PC architecture would not fit in PS/2s. IBM hoped to license the new, more efficient design to other computer makers. Few were buying.
A few months later, IBM finally released the OS/2 it had announced at the rollout of the PS/2 machines. The name was deliberately intended to tie the new operating system to the new line of computers, even though it would not run at all on 8088 machines and ran poorly on 286s. The name suggested that the new operating system would run only on PS/2 machines -- and, unfortunately, it was true that a lot of 286- and 386-based machines not built by IBM simply refused to run the new system.
OS/2 1.0 was not impressive. It wouldn't support hard drives larger than 32 megabytes. It had no graphical interface nor support for a mouse. While many software companies hoped to develop applications for it, few had. It was expensive. And it was very nearly killed at the outset by, of all people, Ronald Reagan.
Reagan believed that Japanese memory chip makers were dumping their product on the U.S. market. In hope of stimulating U.S. chip production, the president threatened heavy tariffs; the Japanese in response "voluntarily" raised their memory prices. A megabyte of RAM that had cost $50 was now $200. High-end computers typically were shipped with one megabyte of RAM. IBM had announced, correctly, that OS/2 required at least four megabytes to run at all well.
Buyers were now given the opportunity to spend at least $600 per machine to upgrade their computers so that they could run an expensive operating system for which there were few applications and that ran DOS applications poorly. Few did so. It was now not just possible to get fired for buying IBM but likely.
Microsoft, meanwhile, had been at work on something called "Windows." In its early incarnations it was laughed off in the computer press, and with good reason: It wasn't much good. Still, it was a graphical user interface, it did support a mouse, and though it was slow and required high-end hardware, it did more or less work at least some of the time. It ran atop DOS, so it brought with it many of DOS's limitations. Attempts to overcome those limitations were flimsy. But it was a useful exercise. Microsoft was learning.
OS/2's graphical mask, called "Presentation Manager," was finally released on Halloween 1988 as part of OS/2 1.1. And Microsoft, at least for a time, believed in OS/2. The second version of Windows was called "Windows Presentation Manager 2.0."
The new OS/2 was instantly crippled by all the existing problems plus a new batch: It included neither communications nor network support, nor did it support databases. For those features, buyers would have to await something called "OS/2 1.1 Extended Edition."
The irony of it all was that OS/2 was a vast improvement on anything that had existed before. Programmers were excited about it. IBM employees were so pleased with it that they would, as hobbies almost, develop little applications for it: Text editors, terminal emulators, a program that would allow it to run old CP/M applications, games, and so on. Among techies, OS/2 was beloved.
Computer users and corporate buyers, meanwhile, decided to sit this dance out. DOS was maturing and worked reliably, and there were many excellent programs for it. Lotus 1-2-3 owned the spreadsheet market. There were excellent database and spreadsheet programs. Networks, though still chiefly "sneaker nets" in which files were transferred by carrying them on floppy disk from one machine to another, were in development and were putting food on the tables of a new industry, computer consultants, who could make networks work -- for a price. Using modems to dial other computers by telephone was a practice enjoyed chiefly by hobbyists, though there were more and more of these. (A recent graduate of the University of Missouri was able to offer a program called "ProComm" as shareware, meaning that people could get it for free, try it out, and if they liked it and wished to have clean consciences they could then register it, paying a few dollars to the program's author. ProComm brought in more than $2 million to the author and his new company, DataStorm, the first year.)
Since its introduction in 1984 the Apple Macintosh had been the machine of choice for graphic designers and others who wished to use computers to make or manipulate pictures. The PC market was wide open for such applications, but it was limited in that PCs running DOS were "character mode" -- they displayed letters and numbers and a few symbols and rudimentary line-drawing and shading characters, but were not in any sense graphical in the way they worked. Companies that wanted to produce graphics programs needed to write huge gobs of code that did nothing but provide a graphical front end for DOS. Microsoft, seeing this need, offered to license to such companies a "runtime" version of Windows. This meant that companies could buy use of a scaled-down Windows that would start when the graphics program started, provide graphical, printer, and mouse services, and shut down when the graphics program did. An added advantage, to Microsoft, was that the graphical programs that used the runtime version of Windows would also run on a full version of Windows. It was an effective way of encouraging software developers to write for the Windows API. Programs such as the very popular CorelDRAW! and Aldus Page Maker were shipped with runtime Windows. It did not seem to matter that the Windows API changed from version to version -- it would be left to the applications developers to produce fixes that would allow old Windows programs to run on new versions of Windows. (Digital Research tried to do the runtime thing with its GEM graphical interface. It never caught on, in part because GEM was awful and because the programs that employed it, led by Xerox Ventura Publisher, were monstrously difficult to use.)
But work on OS/2 didn't end. Version 1.2 was released in 1989, and it did nothing to enhance the reputation of the operating system.
"It was terrible," says IBM's John Soyring. "We were embarrassed to go to customers with it." Those customers had been promised a lean and bulletproof operating system. What they got instead illustrated the growing rift between IBM and its partner in the enterprise, Microsoft.
"We were concentrating on a stable system, technically sound and reliable," says Soyring. "Microsoft was interested in the user interface."
Microsoft's Bill Gates recasts the disagreement, saying that it chiefly dealt with Microsoft's hope that Windows programs could be modified to run as native applications on OS/2 with relative ease, while IBM's demands made this impossible.
It would turn out that both descriptions were correct.
The dispute figured in the companies' decision to go their separate ways when their collaboration agreement expired, which took place in phases until the final split in September 1992. The schism began fairly peaceably about the time the awful 1.2 appeared, but it was not long before it had degenerated to open hostility. Microsoft had very critical things to say about IBM, which complaints were aired on the front page of The Wall Street Journal. IBM worked to calm angry OS/2 customers.
"We had a huge investment by IBM's very best customers, who had bought this thing called 1.2 -- and they weren't happy," says Soyring. "So we had to fix that. We rapidly developed something called version 1.3, which turned out to be an excellent-quality release. We sold over a million copies in just one year of OS/2 extended edition -- and that's selling at a retail price of $695 a copy, so it was a big business all by itself." Version 1.3 was and is a tremendously successful product: it is the operating system used on more than 90 percent of the automated teller machines in use in the world today, as well as specialized airline reservation systems, medical equipment, and even high-end fax machines. It continues to ship as an "embedded system" in such equipment.
The Battle is Joined
Microsoft, for its part, had been working on its Windows product line, which included different versions for 286 and 386 processors. But when Windows was mentioned outside of Microsoft, it was usually in a complaint or as the butt of a joke. Then the company began demonstrating a product called "Windows 3.0." It showed real promise. When it was rolled out in late May 1990, the word of mouth was so enthusiastic that tens of thousands of copies sold before any of the buyers knew much about it.
Windows 3.0 introduced many things to the PC world, not all of them good. It established the industry standard of ridiculously exaggerating the minimum equipment required to use the product -- the Windows box said it required only an 8088 processor. While it was perhaps technically true that it could be made to run somewhat on such a dinky processor, certainly no one ever did any useful work with such an arrangement.
It was pretty -- excitingly so for the time. The deep blue Microsoft Windows screen came up, the installation usually went more or less smoothly, the program to identify and "migrate" existing DOS applications so that they could be launched by Windows wasn't perfect but was a great first effort -- and it had a tremendous Solitaire game.
But it perched atop DOS and, while it added some operating system features to that glorified program loader, it was subject to many of DOS's shortcomings -- plus some of its own. It was supposed to multitask DOS and Windows applications, and there are recorded instances of its actually having done so, but these were remarkable. Mostly, it crashed a lot, for reasons of architecture and of execution. The architectural problem was that Windows employed what was called "cooperative" multitasking. It sounds nice, doesn't? What could be better than programs on a computer all happily cooperating to get things done? The reality was far different. Cooperative multitasking means that all programs had to cooperate or the system would come crashing down. The execution problem came about because DOS applications -- including those from Microsoft; indeed, especially those from Microsoft -- were accustomed to having the machine all to themselves. They would do something that Windows didn't like, and it was time to reboot, a painful 90 seconds which the hapless user could employ wondering how much work had been lost.
And to make things even worse, Windows 3.0 didn't even run Windows programs very well. The existing Windows applications had been written for earlier versions of the product, and Microsoft had been none too careful in working to assure backward compatibility. The company had to rush out a quick fix for its own "Word for Windows" word processor.
Indeed, Windows 3.0 quickly became known for the initials "UAE." These stood for "Unrecoverable Application Error," which is what the pop-up box proclaimed when Windows was just about to crash. It meant the system had gone haywire due to a lack of cooperation -- even if only one program, typically Word for Windows, was running -- and that any work done since the last time it was saved to disk was now about to be lost. Perhaps the most maddening thing about it was the sheer audacity of the little "OK" box: The computer would sit there and do nothing until the user clicked on this box, whereupon the computer would go ahead and finish crashing, now with the user's permission. There was a cottage industry of hacking the code so that the box would have an icon of an upraised middle finger, instead of the Windows alarm icon, and considerably different wording than that supplied by Microsoft, now along the lines that the user had just been screwed by Windows, which was now going to destroy some data. It has been said, and there's good reason to believe, that it was the Solitaire game that kept Windows alive during the first year or so. Users would launch Windows to play the game, which is addictive and which rewards the winner with a spectacular explosion of cards, then close Windows to do actual work in their far-more-reliable DOS applications.
Still, the idea of a graphical interface was an attractive one, and there really was no choice other than Windows. So people mostly left it on their machines and, one by one but then in a flood, native applications capable of perching atop Windows without breaking any of the glass began to appear. UAEs continued to appear, too, but they were taken as part of the cost of being on what had come to be known as "the bleeding edge."
Though IBM and Microsoft had disagreed as to what it would do and how it would look, there had always been plans for OS/2 2.0. It was to be a 32-bit operating system (Windows was thusfar only 16-bit), allowing the operating system to make more robust use of the 80386 and better processors. While IBM's first priority was to clean up the sorry mess that was the (16-bit) 1.2 version of OS/2, work there proceeded, too, on 2.0. The war turned nasty. In Boca Raton, IBM's master software writers realized that due to the residue of the IBM-Microsoft Joint Development Agreement, IBM had full rights to Windows code. This meant that rather than coming up with a way to port Windows applications to OS/2, they could actually build Windows right in to OS/2 2.0. This meant that OS/2 2.0 would run not just OS/2 applications but Windows applications and, through a very solid emulation, DOS applications as well. Dozens of them, al