Building a Legacy
June 27, 2011 Timothy Prickett Morgan
The IBM i platform and its bigger brother, the System z mainframe, take a lot of guff for being a legacy platform. But guess what? Solaris is turning 30 next year, as is HP-UX the year after that. AIX will be 25 this year, Windows server variants are almost 20 years old (remember Windows for Workgroups 3.1?), and Linux pretty much freeze-dried after a hectic 20 years of development. OS/400, of course, just turned 23 last week, but has deeper roots back into CPFon the System/38 in 1978 and SSP on the System/36 in 1983. They are all legacy environments as far as I am concerned.
And that is one of the things that IBM and its Power Systems-IBM i partners should be emphasizing. Perhaps time-tested, stable, well-understood, and predictable are better words to describe what gets slammed as a legacy platform. It’s the truth of the matter and something that drives the executives in charge of Intel‘s Data Center Group, which makes Xeon and Atom server processors and chipsets, to distraction. In Intel’s mind, with over 96 percent of server chip shipments, it should have all of the revenues, too. Ah, but the world doesn’t work that way, as Intel itself showed at its Investor Day briefings in May:
After the relentless double-whammy of the Wintel duopoly and the nibbling at the ankles and then knees from Lintel machines, other platforms–these expensive legacy platforms–have the audacity to persist. As if saving a few bucks on cheaper iron was the only thing that matters.
It is never amazing to me that legacy platforms persist, even as recessions often drive platform transitions and tectonic shifts in computing. The System/38 was on the cutting-edge of the relational database management system revolution, ahead of Oracle on proprietary VAX minis and Unix gear, and even ahead of the IBM System/390 mainframe. The problem was that even though the System/38 was a machine designed to make the job of filling up and tickling transactions and reports out of a relational database easy enough for people who were not computer scientists, it was terribly expensive. Orders of magnitude more expensive per unit of work than an IBM mainframe, in fact. (Chew on that for a second.) The genius of the AS/400 when it launched in 1988 was that it took this capability mainstream at a price that was more akin to the cheapo, flat-file database system in the IBM lineup of the time, the venerable System/36.
Intel looks at that chart above and sees victory. I see defiance. I see customers making the choices to stick with their key platforms, or being unable to reinvent the wheel on a new X64-based platform. Not being able to move and not wanting to move come to the same thing in the end, although how you feel about it is certainly different.
Depending on the year and the state of the economy, I reckon that the IBM mainframe business generates around $4 billion a year in system hardware revenues, numbers that more or less mesh with that coming out of Gartner, IDC, and others as well as company financial reports. For many years, the combined Power-based systems business–the AS/400 and RS/6000, then the iSeries and pSeries, then the System i and System p–generated somewhere around $5 billion in revenues and is now just north of $4 billion. The Itanium-based server business has grown to around $4 billion, and is dominated by Hewlett-Packard with a smattering of other machines. As best I can figure, Oracle is in the process of stabilizing the Sparc-based server business at somewhere just shy of $4 billion.
It looks like $4 billion is the center of gravity for legacy-loving installed bases here in 2011. Maybe 10 years hence it will only be $2 billion for each of these platforms, or maybe only $1 billion. But I am pretty sure that it won’t be zero.
The truly stunning thing about those legacy bases is that we are talking about somewhere between 220,000 and 250,000 machines being shipped per year against a tidal wave of over 9 million servers using Xeon processors from Intel or Opteron processors from Advanced Micro Devices.
The other interesting thing to me is how desperate the Unix players are to crush each other, no matter what the collateral damage. IBM seriously undermined its AS/400 business in the past 15 years as it kept prices artificially high on OS/400 platforms while making them artificially low for AIX customers. Yes, IBM was able to triple revenues for AIX platforms and put the hurt on HP and Sun in the Unix space, taking the number one spot. But in the meantime, it has cut its AS/400 base in half. Best I can figure, this approach took IBM from an estimated 375,000 combined Power-based server customers down to something closer to maybe 225,000. In my book, someone at Big Blue needs to be fired for losing 150,000 customers–about 30 percent of IBM’s enterprise customers in the late 1990s.
Oracle, having shelled out $5.6 billion net of cash to acquire Sun Microsystems in January 2010, is not just hell-bent on stabilizing that Sparc/Solaris server franchise and extracting more profits out of it than Sun itself could. Oracle has gotten into a legal tizzy now with HP over the future of Itanium and therefore of HP-UX, NonStop, and OpenVMS, the three legacy operating systems that HP ported from their three respective platforms to Itanium over the past decade. In late March, Oracle said it would not develop future versions of its database, middleware, and application software on Itanium. As the top seller of Xeon and Opteron servers in the world, you might think that HP wouldn’t care all that much, but undercutting Itanium like that is a death knell for HP-UX and OpenVMS. Why? Oracle is, by far, the dominant supplier of databases for all three big commercial Unixes–AIX, Solaris, and HP-UX–and Oracle controls the RDB database for OpenVMS because Digital sold it off when it was desperate for cash back in 1994.
Databases, as OS/400 and IBM i shops know full well, are far stickier than operating systems. IBM i retains the customers it has precisely because there is one and only one database management system for their databases.
The situation between Oracle and HP has become so poisoned–not the least of which because Oracle co-founder and CEO, Larry Ellison, hired his tennis buddy, former HP CEO Mark Hurd, last fall after he was fired by HP’s board in the wake of an expense report and sexual harassment scandal–that HP sued Oracle in California Supreme Court last week for going back on its commitments to support Itanium processors. And not just in the future, but right now. HP is alleging that Oracle is not providing bug fixes on current Itanium machines running current operating systems from HP, and that it is offering to sell customers shiny new Oracle systems at below cost if they jump.
As you might imagine, both Oracle and IBM must be feeling pretty smug right now. Oracle because of the grief it is causing HP in the systems business to its own advantage (presumably) and IBM because it will now be able to catch an even larger hoard of customers made nervous about the future of the HP-UX platform. But, then again, what is to stop Oracle from saying the same thing about Power Systems? The company has already stopped selling its database on z/OS for the mainframe, although you can get the 11g database on mainframe partitions running Linux. How long before Oracle passes around a memo that says IBM believes the future of Power Systems is Linux and that it is only supporting its software on Linux partition on Power-based boxes? If I were Larry Ellison, I would probably be thinking about that right about now and trying to convey the idea that the only safe place to run Oracle databases is on Xeon or Sparc machines made by Oracle.
Of course, the safest place to run DB2 databases is on an IBM mainframe or a Power Systems box with either AIX or IBM i, depending on your programming preferences. And because of this, I fully expect for z/OS plus DB2, and DB2 for i, and even DB2 for AIX to be around, helping push System z and Power Systems sales, for the foreseeable future. Until Oracle or Microsoft spends a lot of money to skin these DB2 databases and run them virtually atop another database management system–not the stupidest idea I ever thought up–Intel is just going to have to grow its Xeon business by expanding further into storage and switching. It will take at least three more recessions to put big dents in all these legacy systems.