Windows/400: Windows On Power Systems, Take Five
September 19, 2011 Timothy Prickett Morgan
A moment of indignation on behalf of midrange shops that are up to their necks in both IBM i and Windows: Why on earth does the ARM architecture get a version of Windows ahead of the Power server platform? What’s more important: Hundreds of thousands of customers who spend billions of dollars on expensive systems, or tens of millions of consumers who spend billions of dollars on smartphones and tablets that have hardly any profit margin? Why is the latter billion better than the former, BillG?
Way back when 20 years ago, when I still didn’t have very much hair on my head but the tech world looked very different than it does today, the PowerPC chip created by IBM, Motorola, and Apple was supposed to be a common foundation for PCs and servers, spanning multiple operating systems on a single Unix-derived kernel called Workplace OS. That kernel was intended to support AIX, OS/400, OS/2, MacOS, and Solaris and take on the hegemony of Intel and NetWare, Windows, and a few other platforms that have gone the way of all flesh.
When the PowerPC alliance was formed in October 1991, Linus Torvalds had just released the first version of his nearly eponymous Linux operating system, which in August 1991 he announced in a post to the world as a free operating system that was “just a hobby, won’t be big and professional like gnu.”
A few years later, after the PowerPC group had failed miserably in creating a united embedded systems, PC, and server chip platform–the embedded stuff has done pretty well, you can think of a game console as a very specialized PC if you want to be generous, and only IBM sells Power servers unless you count rebranding by Bull and Hitachi, which I don’t–Intel and its server partner, Hewlett-Packard took a stab at creating a unified platform that, in the late 1990s, was going to span from PCs to supercomputers. The chip it was based on was called Itanium, and everyone lined up behind it–Microsoft, IBM, Sun Microsystems (now part of Oracle), Compaq, Hewlett-Packard, Dell, Red Hat, and a slew of others–with those who controlled operating systems, middleware, and application software porting their code to this new architecture, which didn’t pan out quite as planned.
Windows used to run on Power, and Microsoft pulled the plug, only to revamp it for its Xbox game consoles, which use a variant of the 64-bit Power processors that have been at the heart of AS/400, iSeries, System i, and Power Systems-IBM i machines. So, technically, Windows has run on Power all along, even though it only had a brief life back in the days of Windows 3.51 on Power workstations and never formally made it to a server. And seeing Microsoft getting all hot and sweaty with excitement about porting the future Windows 8 client software to the ARM architecture commonly used in smartphones and tablets last week at its Build developer conference in Anaheim got me to thinking again: How come we don’t have Windows Server running on Power?
With Microsoft under threat in the phone and tablet market, the PC market declining, Itanium on the ropes, and the server market recovered but not growing particularly fast, now might be a good time for Microsoft and IBM to re-examine the possibility of running Windows (or at least Windows applications) on Power machines.
Over the years, I have looked at this six ways from Sunday. That’s ranging from the hard and expensive, which means doing a full, native port of the Windows operating system and the key systems and middleware software, such as the SQL Server database and the Exchange Sever groupware server, to IBM’s Power Systems all the way down to running Windows as an emulated application atop the QuickTransit emulator software IBM bought up and sat on in November 2009 to keep it out of enemy hands to the much wimpier running of .NET applications inside of the Mono clone of the .NET Common Language Runtime. On the plane ride back from Intel Developer Forum, which I was working all last week up in San Francisco, I even pondered a sort of back-to-the-future approach, imagining that you could take a bunch of Xboxes and hack Windows Server on them and attach then to Power machines. (To what end? Who would trust enterprise applications on that kludge? And besides, the techies would be using them to play games, as we all know.)
The reason why so many of us return, again and again, to this Windows on Power idea is that we get, perhaps more than enthusiasts for other operating system platforms, that the software is what drives the hardware as the hardware merely supports the software. At this late stage of the systems market, where the X64 server base accounts for something close to 97 percent of shipments and maybe two-thirds of platform revenues worldwide, anything that makes the Power architecture stronger and more widely used helps protect and preserve the AS/400 and its progeny. Just like AIX sales help make it possible for RPG and COBOL applications to live on and remain relevant–some would say that IBM brock the back of the AS/400 market to kill the Unix businesses of Hewlett-Packard and the former Sun Microsystems, and did so by alienating half its customer base which was kinda epically shortsighted (and that some who was doing the saying would be me)–having Windows on Power and arcing into that Windows partner channel could only help preserve your ability to keep coding in RPG all that much longer.
The problem with all of these scenarios we have been mulling over the years that they did not clearly explain what was in it for Microsoft, what was in it for IBM, and what was in it for you. And as I was pondering the possibilities at the height of my jet lag on Friday afternoon, I had a funny thought:
If you want to get Microsoft to run Windows on Power, maybe you have to port DB2/400 and RPG to Windows and give Windows what amounts an OS/400 command shell. Maybe you have to have IBM and Microsoft be operating system partners again–and I am well aware of how disastrously that worked out for Big Blue with DOS and OS/2. It sure as hell worked for Apple, which has plunked its Mac OS interface on top of BSD Unix to create the Power and Intel versions of Mac OS and used QuickTransit (now controlled by IBM) to run Power executables on Intel iron.
Whenever I am negotiating, I always see this scene in my head, from Excalibur, when Uther Pendragon, the ill-fated father of King Arthur, starts uniting England. Head-strong Uther actually listens to the wisdom of Merlin, however much it pains him:
Merlin: [Whispers to Uther] Show the sword! [Loudly now] Behold! Excalibur, the sword of kings; forged when the world was young and life was but a dream. Speak the words!
Uther Pendragon: One land, one king! That is my peace, Cornwall!
Lord Cornwall: Lord Uther, if I yield to the Sword of Power, what will you yield?
Uther Pendragon: Me yield?!?
Merlin: [Angry now] He has given, now you must give.
Uther Pendragon: The land from here to the sea is yours if you enforce the King’s will!
Lord Cornwall: Done! Come, King Uther, let us go to my castle and feast!
He has given, now you must give. I don’t think I could find truer words in the world if I tried. (Of course, Uther goes and screws it up, quite literally, by stealing Lord Cornwall’s wife. But that’s how we get Arthur and the Round Table and a united England and Camelot, if but for a brief 25 years.)
All the tactics that we have pondered to bring Windows and Power together over the years almost always put IBM and Microsoft in contention, when what I want to do is preserve OS/400 and help make hybrid OS/400-Windows data centers run leaner and meaner. I want there to be a graphical user interface and a native Web environment for OS/400–let’s call it the forthcoming Metro UI that is coming with Windows 8 and Windows Server 8–and I want Windows’ NTFS to be the file system for the whole shebang. I am not sure what the hypervisor layer on the machine should be, but it should look like Hyper-V even if it is actually PowerVM because that is what millions of people in the world will use. Systems Center should be the management tool.
What would such a unified machine give Microsoft? Well, for one thing, there would be something on the order of 125,000 customers in the world with maybe 250,000 machines that would suddenly be using Windows Server Datacenter Edition with unlimited virtualization. Even at the current per socket price, that’s well over $1 billion for the current i5/OS-IBM i installed base for the raw operating system on just the IBM midrange engines. Now, add up all the Wintel engines in these same shops. Is it a factor of 10 larger? A factor of 20? Imagine how many Windows Datacenter Edition licenses you could sell on Power-based servers then. Now we are talking maybe $10 billion to $20 billion. And that doesn’t count the Client Access Licenses for access to Windows, which costs $40 a pop at list price. Call it 12 million end users on the OS/400-IBM i base, and heaven only knows how many users coming in over the Web. The direct users alone would be a half billion dollars of user licenses.
He has given, now you must give.
So what does IBM get out of the deal? Now, instead of there being an installed base of maybe 325,000 cores running OS/400, i5/OS, and IBM i on 250,000 machines, you might be talking about anywhere from 5 million to 10 million aggregate sockets running infrastructure workloads on machines at maybe 10 to 15 percent of CPU utilization that comprise 5 to 10 million sockets. That’s $15 billion to $30 billion dollars of Windows Server Datacenter Edition software, and half that for Enterprise Edition. Not including the Windows CALs. But that doesn’t take into account virtualization compression as efficiency goes up.
How much potential Power Systems iron is that if customers load up on Power 750 or Power 770 machines and create monster Power clusters running at 80 to 90 percent efficiency? It is still arguably many, many billions of dollars of software for Microsoft, and many, many billions of dollars in Power Systems iron for IBM as both Windows and IBM i workloads to scale up and down the iron. And, of course, IBM gets to sell DB2/400, RPG.NET, and COBOL.NET, too, which will be enough money to justify its existence for a long, long time.
And the best thing of all is that the war between Windows and IBM i will be over, since they will be one thing. We could even call it Windows/400 just to be cheeky.