Programming Windows: More Roads Not Taken (Premium)

Microsoft chief scientist Nathan Myhrvold opined in an internal email in 1992 that the software giant’s business model of the day amounted to no more than “retail sales supplemented by upgrades.” Microsoft, he said, should pursue new business models. “Upgrades [today] represent the closest thing we have to an annual fee or subscription,” he wrote. “This is a powerful way to draw revenue from the installed base, and to keep them loyal to our product.”

As was so often the case, Microsoft eventually acted on the futurist’s advice: The software giant adopted a subscription sales model when it began selling its offerings to the enterprise, and it has since shifted most of its product lines to subscription sales models. In fact, one might gauge the success of the “new” Microsoft by this core change to its business model.

But in the interim, Microsoft experimented with different ways to normalize its revenues and avoid the uneven ups and downs of its Windows upgrade schedule, which typically amounted to a major new version once every three years. There were various permutations of this evolving strategy. But one you’ve almost certainly not heard of is that Microsoft briefly toyed with the idea of annual releases of Windows that would occur as the firm tried to move its codebase off of MS-DOS and Windows 9x and to NT.

It was February 1998. Microsoft’s first major update to Windows 95, now called Windows 98, had been delayed until May 1998 because of the software giant’s ongoing antitrust issues. Windows NT 4.0 had been in the market for about 18 months, but its successor, Windows NT 5.0, had likewise been delayed thanks to the difficulty Microsoft had bringing a full-featured directory services infrastructure, Active Directory, and its IntelliMirror management technologies to market.

By that point, Cairo, Microsoft’s original plan for the next major release of Windows NT, had already been canceled. Instead, Microsoft had delivered three minor releases, NT 3.5, 3.51, and 4.0, which delivered bug fixes, performance improvements, compatibility with new hardware platforms, and, in the 4.0 release, the same stripped-down user interface refresh that had debuted earlier in Windows 95. But since mid-1996, all had been quiet on the NT front.

Sales of Windows 95, by comparison, had exploded. And thanks to external forces—the rise of the Internet and new Plug and Play hardware advances, most obviously—Microsoft had been forced to quickly update that platform over the intervening couple of years. But instead of refreshing the retail version of Windows 95, Microsoft issued service packs, which fixed bugs and added support for new hardware such as large disks (via the FAT32 file system), USB, IEEE1394 (Firewire), Intel’s MMX processor extensions, and newer Intel processors like Pentium Pro and Pentium III. Microsoft also released OEM Services Releases (OSRs) for PC makers which bundled the features from service packs with the original release of Windows 95. The final OSR release, called OSR 2.5, even added Internet Explorer and its Windows Desktop Update, giving it many of the same features that would later ship as part of Windows 98.

Despite the success of Windows 95, Microsoft hadn’t given up on NT. Far from it: The firm’s “vision” for the PC desktop in February 1998 was to plot the transition from Windows 9x to Windows NT over the next few years. Windows 98, the strategy planners at Microsoft said at the time, would be the final release of a DOS-based Windows version.

That wouldn’t happen for a variety of reasons, and as most readers probably remember, Microsoft actually shipped two more DOS-based platforms after Windows 98, called Windows 98 SE (Second Edition) and Windows ME (Millennium Edition). But one of the less-heralded issues that Microsoft faced in the late 1990s was the emergence of the market for sub-$1000 PCs. This suddenly exploding market was a dual-edge sword for Microsoft. On the one hand, demand for lower-cost PCs, in particular from households, drove Windows license sales higher and higher. But these lower-end PCs could not run Windows NT, necessitating further investment in Windows 9x. And further delays for the expected transition to NT.

This was a problem, as the software giant had also finally figured out a role for NT. Where this advanced system was previously relegated to an ill-defined market for workstations as well as servers, Microsoft in 1998 finally had a marketing message that would make sense to all of its customers: Windows 98 was for consumers, and Windows NT was for businesses.

The change in positioning came about in the wake of Windows NT 4.0, which sold far slower than Microsoft had hoped. Despite its visual similarity to Windows 95 and its dramatically better reliability and security, NT 4.0 had been installed on only 8.4 percent of PCs sold to businesses since its release in August 1996; even the business world had rallied around Windows 95. That was a problem because Microsoft charged PC makers a lot less for Windows 95 than it did for NT 4.

As bad, the OSR strategy—which analysts referred to as “dribbleware”—had confused the market. There was no longer a single version of Windows 95. Instead, there were several versions, each with unique features. And Windows 98, which should have been viewed as a major upgrade, was seen by most as just another OSR release. In short, the OSR strategy had made an actual retail upgrade, Windows 98, less enticing, not more. Not surprisingly, it arrived to far less fanfare than did Windows 95.

Microsoft’s plan, what it called its “silver bullet,” was centered on Windows NT 5.0, the next major release of its premier OS platform. NT 5 would bring the directory services and management advances that Microsoft had been promising for years and it would significantly close the gap with Windows 9x from a hardware compatibility perspective. Too, its user interface would be nearly identical to that of Windows 98, giving users a more seamless experience if they moved between the two environments.

NT 5, in short, would be a major release. And it would form the foundation for a three-year plan to transition the user base from Windows 9x to NT using a minor-major-minor schedule. Microsoft would itself move to the NT codebase after Windows 98 had hit “RTM” (release to manufacturing), though it would also deliver a “very limited” Windows 98 service pack that included bug fixes, Internet Explorer component updates, and a very [emphasis Microsoft’s] small number of [new] features. Then, it would move forward to Windows NT 5.0 and to three upgrades to NT 5.0 that would each arrive one year apart, with NT 5 arriving first in fiscal year 2000 (which started in July 1999).

The two minor updates—scheduled for FY2001 and FY2002, respectively—were not to be service packs. Instead, Microsoft also created a separate team to work on service packs for each release, and these service packs would be “kept small to avoid impact on [the] annual releases.” They would also be slipstreamable, a new capability that allowed system administrators, PC makers, and even end-users to combine service packs with an OS install image so that the service packs wouldn’t have to be manually installed, one at a time, later.  (Today, these kinds of updates are cumulative, eliminating the problem once and for all.)

The three upgrades to Windows NT 5.0 were called NT “Asteroid,” NT “NepTune,” and NT “Triton,” respectively. (And yes, Microsoft’s internal documents originally referred to that middle release as NepTune, I assume to highlight its connection to NT. But I will refer to it as Neptune from here on out.)

NT Asteroid was a minor upgrade to Windows NT 5.0, and it was expected to appear in Q4FY99, or the second calendar quarter of 2000. It was a “point release” that would accelerate NT 5 deployments thanks to fixes and improved deployment tools. For consumers, Asteroid would “remove barriers for NT on high-end consumer systems.” And it would be made available for sale at both retail and via PC makers.

(Fun aside: Wikipedia incorrectly claims that Asteroid was the codename for Windows NT 5.0 Service Pack 1. As noted, none of these releases were designed as service packs, and NT 5.0 SP1 had no codename.)

NT Neptune was a major release set for release in Q4FY00, or the second calendar quarter of 2001. It would feature major new features like Active Directory 2.0, Zero Administration for Windows (ZAW), and 64-bit support, initially only on the Intel Merced (Itanium) platform. Consumers would get a “significantly simplified end-user experience.” And developers would get major platform advances such as COM+, Win+ (the predecessor to the new user interface functionality that Microsoft would later transition to Longhorn), and a new storage infrastructure.

(Fun aside: Exactly one external build of Neptune leaked, presenting an early look at Microsoft’s take on a consumer-oriented version of Windows NT.)

NT Triton was another minor release, and it was scheduled for Q4FY01, or the second calendar quarter of 2002. Since this release was so far out, there wasn’t much information about it in Microsoft’s 1998 planning documentation. The only item of note is “new hardware support for CY01 holiday.”

For the NT 5.0 active lifespan—roughly 1998-1999—Microsoft expected all low-end and mid-range PCs to ship with Windows 98. As the schedule transitioned to Neptune in 2000 and 2001, however, entry-level and mid-range PCs would be running Neptune and then Triton, and not Windows 98. They were to have transitioned to NT.

This is all very interesting. But none of it ever happened.

Windows NT 5.0 did eventually ship, of course. But it was renamed to Windows 2000 and didn’t arrive until February 2000, over three and a half years after its predecessor. In the interim, Microsoft also shipped Windows 98 SE, and then Windows Millennium Edition shipped in September 2000.

As for Asteroid, Neptune, and Triton, those astronomy-themed projects gave way to Whistler, which shipped as Windows XP in 2001, and to Longhorn and Blackcomb, neither of which ever shipped either. Longhorn was originally designed as a minor upgrade to Whistler, whereas Blackcomb would be a major release. (Longhorn was later recast as a major release but was later canceled.) But Microsoft had abandoned more than just the codenames. It also gave up on its annual release strategy: For a variety of reasons, Windows XP stayed in the market for several years as a host of other external forces—from antitrust problems to security vulnerabilities and more—triggered further changes in Microsoft’s strategy.

We’ll get to all that eventually. But let’s get back to 1998 first, as there’s still plenty to discuss.