Welcome to Legacy Status, Windows Server
February 25, 2008 Timothy Prickett Morgan
This week, Microsoft will finally get Windows Server 2008, formerly known as “Longhorn Server” and the companion to the desktop Windows Vista operating system, out the door. If you don’t count normal service packs but do include sub-releases, this is the eighth major update to the Windows server platform, and as Windows has come to dominate the server space (unless you count Unix and Linux together, which I do), it has in many ways moved back in time and become a legacy system.
Yes, I know how weird that sounds. But it is nonetheless true.
A lot has changed with the Windows platform since Windows “New Technology” 3.1, the first commercial release of what was supposed to be called OS/2 3.0 until Microsoft broke up with IBM on that jointly developed operating system, was launched with a Windows programming model instead of an OS/2 programming model in July 1993. Microsoft has innovated as fast as it was capable of doing, and was knocked off course first by the Internet in 1995 and by security concerns in 2000. Other issues, such as when Microsoft’s whiteboard plans have from time to time exceeded its programming grasp as it did with the WinFS integrated file system that is supposed to be part of Windows Server 2008, have also delayed Windows server releases, sometimes for years. But the drumbeat has been steady, and no less embarrassing than for other operating system suppliers, if you want to be fair about it. Look at the drumbeat:
Many enterprise customers are perfectly happy to have many years between operating system releases because it is such a huge pain to test and qualify their applications on a new system software stack, regardless of the benefits that software might bring to their organization. That is why Windows NT 4.0 and Windows Server 2000 still persist out there in the data closets and data centers of the world. (Everyone has a few of these old boxes sitting around, and they see very little reason to mess up something that is working as long as they keep these servers safe behind the corporate firewalls where hackers can’t see them.) This is one reason that Microsoft has taken its time getting Windows Server 2003, formerly known by its code-name “Whistler,” and Windows Server 2008 out the door.
But there wasn’t even supposed to be a Longhorn Server if you go back a decade and look ahead on the roadmaps. Microsoft was going to jump from Whistler to a really dramatically different platform codenamed “Blackcombe,” and basically using what we would call the Windows NT 6.0 kernel. Longhorn was a desktop release with an improved NT 5.0 kernel, and because of delays in Blackcombe and Microsoft’s lack of confidence in tackling some of the advanced features it wanted to roll into the platform, Longhorn Server was plunked in the middle of Whistler and Blackcombe. And even then, the company had to yank features out of Longhorn Server to get it out the door. The WinFS file system was excised from Longhorn Server in September 2004 and then removed from the Windows server roadmap in June 2006 completely. (Similarly, the “Viridian” Hyper-V hypervisor is not yet ready for Windows Server 2008, which will launch this week without this key feature.) WinFS was supposed to be a feature of Blackcombe, way back when. No one talks about Blackcombe any more, and the next iteration of Windows, code-named “Vienna” and probably just an update to Windows Vista and Windows Server 2008, is expected in the second half of 2009. There may not be a server edition, in fact.
Windows may account for roughly half of server sales each year and probably around three-quarters of worldwide shipments, but it was supposed to be more than just another legacy platform–even though it was certainly intended to be a proprietary operating system as much as any mainframe or midrange OS was ever intended to me, and more so than any Unix variant or Linux distro.
OS/2 3.0 was intended to run on Intel‘s i860 RISC processor, which was called the N10 apparently, and that is where the NT in the original name comes from. When Microsoft decided to switch from an OS/2 programming API to a Windows32 API with Windows NT, the NT name stuck and the marketeers at Microsoft said it meant “New Technology.” But the idea behind Windows NT and the hardware abstraction layer–which is now missing in Windows Server 2008–was that Windows would be a portable operating system, like Unix tried to be and Linux has become. The original Windows ran on NEC, Alpha, and MIPS RISC processors as well as 32-bit X86 chips, and with Windows 3.51, Windows even ran on 64-bit PowerPC chips from IBM. The idea behind New Technology was really Neutral Technology–Microsoft would liberate the operating system from ties to any particular hardware, and let the best hardware platforms compete. Microsoft would unify instead of dividing, and win no matter who chose Windows.
It was a brilliant idea, and one that Microsoft should have found a way to stick with. (It certainly had and still has enough monopoly money to do this.) Because truth be told, IT managers want a portable operating system that has lots and lots of ISVs supporting different hardware platforms. Monoculture on the network is a disaster in terms of security–a mixed network can’t be compromised so easily–but the idea of Windows everywhere would also mean vendors could pick and choose from a wider variety of hardware, tailored specifically to their performance, price, and power consumption needs. But over time, Microsoft and its chip partners all pulled the plug on Windows support, so the industry is basically stuck today with Windows running on X64 processors from Intel and Advanced Micro Devices and a smattering of servers based on Itanium chips, mostly manufactured by Hewlett-Packard.
Not only has Microsoft limited hardware choice, but the vastness of its Windows installed base–probably 15 million to 20 million machines worldwide, is my guess on a Friday afternoon–obscures the fact that there are a lot of different Windows versions out there, and a lot of them are going to sit right there. It is a lot harder to get people to update their servers than it is to get them to update their desktops–and even that is not necessarily easy at this point. The fact that we beat on our PCs and they get filled up with crufty software after a few years of use keeps us moving ahead, but a server with a steady workload and a job to do doesn’t really ever need to change–and in many cases, the IT department is not even aware that it is doing its job. But with somewhere on the order of 6 million new physical Windows servers shipping each year now, Microsoft can have a pretty good revenue and profit stream from Windows Server even if all that vintage gear running legacy Windows stays put. (The same does not hold true for other legacy bases once they mature to a certain point. IBM needs to sell lots of new System z capacity and System i and System boxes to justify continuing investment in mainframe and Power Systems iron. Ditto for Sun Microsystems and its Sparcs and HP and Intel’s Itaniums. But in their days, these platforms had a lot more dough coming in than was spent to keep them innovative.)
It is with a certain amount of irony that I see that Windows Server 2008 is being modularized, like a proper operating should be, and Unix and Linux have always been. This means you only install the components of the operating system that you need, which means it run leaner and meaner and which makes it easier to lock down and secure. (This is called Server Core and Server Roles in Longhorn speak.) And with Hyper-V, Windows will draw more or less parallel with IBM’s OS/400 operating system from the V4R4 days back in 1998 with logical partitions and Unix platforms that had similar virtualization capabilities as the millennium turned. I also got a chuckle when I learned about PowerShell, the new command line interface for Windows and its related scripting language. This is Windows without the, er, window. Apparently you can still learn a thing or two from mainframe, AS/400, and Unix legacy platforms, eh? Maybe instead of labeling such features as “legacy,” it makes sense to call them what they really are: Grown up.