Why you'll have a long wait for Microsoft's next OS

The recent buzz over Microsoft’s efforts to build a completely new OS from scratch has led to some wild speculation. The silliness reached its apex last weekend in the New York Times, where San Jose State University business professor Randall Stross argued that “[t]he best solution to the multiple woes of Windows is starting over. Completely. Now.” In a rambling essay filled with factual errors and mistaken assumptions, he mentions Microsoft’s Singularity research project and says “Microsoft should move its researchers into the heart of its systems development team” and begin turning that research project into a replacement for Windows.

That point of view is a popular one. Over the past year, I've read plenty of speculation that Microsoft is planning a "complete rewrite" of Windows. Most are based on wishful thinking rather than anything concrete, and Microsoft has pretty much stomped those rumors for Windows 7. But hope springs eternal for the version after that, which is why Singularity has taken on an almost mystical aura for Windows critics.

I’ve got good news for Prof. Stross: As my colleague Mary Jo Foley has reported, Microsoft already has an all-star team that’s working on a next-generation operating system. It’s called Midori, and Mary Jo’s sources say it’s in “incubation,” which means it’s on a fast track to being turned into a product.

[For another point of view on Microsoft's next-generation OS, see Mary Jo Foley's post, "Might Microsoft's Midori be 'Cairo' revisited?"]

But will Midori replace Windows in the near future? Not a chance. If Microsoft really does turn this project into a commercial product, I believe it will exist alongside Windows for several years, at a bare minimum. To learn why, let’s dust off the Windows history books.

Way back in 1993, Microsoft rolled out Windows NT. It was technically a 1.0 release, and the code base was completely new, built by a team led by Dave Cutler, who had previously worked on VMS. (Cutler reportedly told Steve Ballmer that he didn’t want to build a “toy operating system.” Ballmer says he replied, “Good. We already have a toy operating system.”)

The NT label stood for New Technology, and for almost another decade more Microsoft built the consumer (3.1/9x) and business (NT) lines in parallel. It wasn’t until the introduction of Windows XP at the end of 2001 that the old line was killed off and the “new technology” became mainstream for all Windows users.

During that eight years, Microsoft released five major versions of its “old technology,” and four versions of its NT product, in both server and workstation flavors, with multiple service packs along the way. Many businesses continued using Windows 95 and Windows 98 on desktops for years after the launch of Windows XP and more than 10 years after the introduction of the “new technology.”

So what does this history lesson have to do with Midori? Maybe it will make more sense if we give Midori a new name: Windows NNT (for New New Technology). An operating system built on a completely different kernel would, by definition, be frightening to conservative business customers, and incompatibility issues would be legion, by definition. It took three years before Windows NT was available in a version that was considered acceptable for desktop use, and it took more than six years before Windows 2000 put all the pieces together in a package that achieved wide acceptance. Meanwhile, Microsoft continued to ship tens and eventually hundreds of millions of Windows licenses using its “old technology” OS.

If we look at Midori as Windows NNT, it has the potential to coexist alongside the Windows Vista/Windows 7/Server 2008 line for at least five years. Just as with NT in its early days, there would be plenty of customers willing to kick the tires and even deploy the new OS for specialized reasons. I can think of three situations where a new OS would be welcome:

• Special-purpose consumer devices. Windows Media Center is mature and extremely well supported. It wouldn’t be that difficult to port the Media Center code to a next-generation operating system that could then form the basis for cool, quiet PCs that could form the hub of a household digital media system. In fact, a device like HP’s MediaSmart Server, which currently runs Windows Home Server, could combine Media Center and backup functions into a single box and would probably run better without the unnecessary overhead of Windows components it doesn’t use.
• Virtual servers. Windows Server 2008 includes the Hyper-V virtualization platform, which can host multiple virtual machines on a single physical box. Instead of requiring Windows Server 2008 Core, why not build the virtualization platform on the new OS? Microsoft would continue to sell Windows Server licenses for the VMs themselves, but could improve performance, manageability, and security on the underlying platform.
• High-performance workstations. A small but influential percentage of Windows customers use the platform for graphics, design modeling, and other high-performance tasks where compatibility isn’t an issue. Presumably, AutoCAD and Adobe would be among the first companies to port their software to the new platform.

Meanwhile, it would be business as usual for the rest of the Windows platform. In a world where a significant percentage of businesses are still running on Windows 2000, it’s hard to imagine that an all-new platform would achieve any critical mass until early adopters had pounded on it for years. Likewise, I can’t imagine OEMs being willing to accept the burdens of selling and supporting a completely new platform until it has proved itself in the real world for a generation or two. And just as in the early days of the NT family, hardware manufacturers would continue to focus their priorities on the mainstream OS, meaning that choices would be more limited for early adopters of the new OS.

When I add it all up, I see a product mix that looks remarkably similar to the one Microsoft sold in the 1990s, with a mainstream line (the Windows Vista/Windows 7/Server 2008 family) and a new line that has to prove itself for at least five years before it can take the lead.