Big Blue Should Do Power Windows, Too
October 4, 2004 Timothy Prickett Morgan
For many years, there has been a lot of talk about whether IBM and Microsoft should create a 64-bit variant of the Windows server environment for IBM’s Power family of RISC processors. With the advent of the second-generation of Microsoft’s Xbox game machine, code-named “Xenon,” the issue is very nearly a moot one from a technical standpoint. A lot of the work on the hardware abstraction layer to make Windows work–again–on the Power architecture will be done.
I say “again” because many people forget that IBM and Microsoft worked like crazy in the mid-1990s to fulfill one of the original promises of the PowerPC architecture espoused by IBM and its PowerPC partners, Apple and Motorola, which was to get OS/400, AIX, MacOS, and Windows all running on the PowerPC architecture. In 1995, IBM and Microsoft accomplished this when IBM announced 32-bit PowerPC-based workstations (called Power Personals), running the Windows NT 3.51 operating system. Unfortunately, for IBM and the people (like myself) who have wanted Power to be a strong alternative to Intel‘s X86 architecture, IBM had a new chairman and CEO during the first Windows port to the Power architecture. That new chairman, Louis Gerstner, was under Wall Street’s gun to cut costs and boost profits, and he thought that operating systems were no longer control points in the IT ecosystem. So just as Microsoft got Windows NT 4.0 out the door, the Power Personals and all future work to get Windows server variants running on Power-based AS/400 and RS/6000 servers were killed.
It is painfully ironic that IBM killed support for Windows on Power just as Microsoft delivered Windows NT Server 4.0 (which ran on the PowerPC platforms), since this iteration of Microsoft’s server platform put Windows solidly into the data center as a real alternative to Unix and proprietary systems. Some may argue that IBM pulled the plug on Power-Windows just about the same time as NEC did on its variants of Silicon Graphics‘ MIPS processors. And it wasn’t much later that struggling Digital Equipment (now part of HP, through its acquisition of Compaq) ended NT support on its AlphaServer line. What makes the irony so painful is that, even though Motorola completed screwed up the 64-bit PowerPC 620 chip, the first of the PowerPC family to be designed specifically for servers, IBM itself managed to create a very powerful line of 64-bit chips, called the PowerPC AS chips, for the AS/400 line. It was touch-and-go in 1996, but by 1997 these chips were enhanced to become the Star family of chips that put IBM into the Unix server game for the first time and kept the AS/400 line alive.
Imagine how different IBM’s server business might have looked if the Apache, Northstar, S-Star, I-Star, Power4, and now Power5 generations of servers had been able to run Windows natively? Remember that Windows is one of two growing operating systems in the world today (Linux is the other one) and accounts for roughly one third of worldwide server sales in any given quarter. IBM might have had a Power-based server business that was two- or three-times larger if it had stuck it out and delivered better Windows support than X86 platforms, which are only now really being given practical 64-bit extensions through the “Nocona” Xeon DPs and future “Potomac” Xeon MPs. Hindsight is easy, of course, but IBM took the easy way out and lost a lot of sales and profits.
The question now is not so much if it is too late to get Windows back on Power. According to the rumor mill, the future Xenon Xbox will have three PowerPC cores on a single chip, running at an incredible 3.5 GHz, with 256 MB or more of main memory, and an integrated graphics card with 10 MB of its own memory. The Xenon Xbox will have vector processors built in as well. It is my guess that Microsoft went with three stripped-down PowerPC 980 cores in the box, but they could be Power5 cores or even a quad-core Power6 chip with one of its cores chopped out. (I like this latter idea.)
In any event, if IBM has the foresight to realize that everything it says is a benefit about running Linux on Power (and, indeed, OS/400 and AIX) is equally true about running Windows on Power. Windows, as Microsoft’s Xenon Xbox shows, is a foregone conclusion on the Power platform. Once IBM and Microsoft have created the hardware abstraction layer (HAL) that allows a stripped-down variant of Windows on the Xbox with PowerPC cores, how hard would it be to create the HAL to allow Windows Server 2006 or whatever Longhorn Server is going to be really called run on IBM’s Power6 platforms? I do not believe this will add that much to the work, and there are big benefits to IBM’s customers.
While I would like for IBM to sell as many OS/400 or AIX platforms as possible to support the Power platform, the fact remains that volume is the biggest determinant of success in this business. And if it takes the support of Linux and Windows to drive up Power server volumes, so be it. Because the fact is that, if IBM doesn’t get enough server volumes, it cannot stay on the price/performance curve that guarantees OS/400 and AIX platforms will get less expensive each year. I want the price of OS/400 platforms in particular to come down–and by a large amount–and the only way to afford that price cut is to boost the volumes of the Power server platforms. With Windows on 32-bit X86 now essentially the dominant single platform in the world, it is idiotic not to try to chase down that market and deliver a 64-bit platform that has better systems management and virtualization features than an X86 server from any of the big suppliers. IBM can make this happen.
The people at IBM I have spoken with have said that there simply isn’t enough development money to do this, and while I agree that IBM has probably not budgeted for a Windows on Power project, I do not believe, given the work IBM is already doing on the Xbox, that this is the case any longer.
Funny enough, the ability to run Windows in a logical partition on a Power-based server would be more useful to the current installed base of RS/6000 and pSeries customers, who have server sprawl, just like OS/400 shops, but do not have the benefits of the Integrated xSeries Server (IxS) co-processors, which run Windows and now Linux, or the Integrated xSeries Adapter (IxA), which allows external xSeries machines to plug into an OS/400 platform and use its disks and file management software. While the IxS and IxA have helped bring better control of Windows platforms at OS/400 shops, companies are still limited by the uniprocessor performance of a single card in the case of the IxS or by the four-way machines that can be attached to an IxA. Even one of IBM’s BladeCenter blade server boxes does not help cut down on server sprawl so much as it helps make it more manageable. For true server consolidation, a native Windows implementation would be ideal on the Power boxes.
While having fine-grained Linux partitions is a great thing, and something that IBM has tried to sell for several years against Windows, companies do not want to find Linux-based alternatives to their Windows workloads. If they had the choice, I am certain that they would pick Linux workloads on their merits and put them into Power server partitions and then move as many Windows applications as makes sense onto Power boxes, to cut down on sprawl.
In the IT industry, a vendor very rarely gets a second chance to do something right. The difficulties in the transition from 32- to 64-bit processing in Windows and the superior qualities (especially virtualization and partitioning) of its Power platform have given IBM a second chance. Windows has such a huge installed base (and one that will continue to grow) that IBM is simply wrong not to chase it directly with its Power platform. It is really that simple.
A final thought. Once this Xbox is out, IBM should hack OS/400 onto it and create a really inexpensive eServer i5, and one with killer graphics!