IBM Systems: The Foundation for Glass Skyscrapers
May 17, 2010 Timothy Prickett Morgan
Although IBM‘s president, chief executive officer, and chairman, Sam Palmisano, did not utter the words server, mainframe, or Power Systems once in his presentation to Wall Street investors last week as he outlined the next five-year plan for revenue and profit growth for Big Blue, fear not. Systems remain at the very core of what IBM does today and what it will do in the future. But don’t be confused. Systems are the foundation, they are not the house, and that house will have many more stories of software and services atop it than we are used to seeing.
(See our coverage of Palmisano’s presentation at the 2010 Investor Meeting on Wall Street last week elsewhere in this issue.)
If the 1960s and 1970s saw IBM create what we now know as a data center, commonly called a glass house, than what IBM is talking about delivering is something that, metaphorically speaking, one might call a glass skyscraper. And a large component of this skyscraper will be broad and deep industry and systems knowledge, brought together to give IBM something to sell to customers who are hesitant to shell out dough for anything that does not provide a quick return on investment.
The days of investing in systems for the sake of having them are long gone, as any AS/400 or mainframe shop will laugh and tell you. And the true irony, of course, is that system utilization has never been higher across all platforms and hardware has never been cheaper–although the gaps between x64 and RISC/Itanium and mainframe platforms remain about what they were a decade ago.
Rod Adkins, who is general manager of IBM’s Systems and Technology Group, did, unlike his boss, talk about Power Systems and mainframes, in his presentation to Wall Street last week. And he talked about the prospects IBM sees for its server biz without getting into the weeds going through technology roadmaps. This was Wall Street, after all, and what IBM wanted to talk about was money, not the chips and systems that make it happen.
Adkins started off bragging a bit about how margins are improving in STG and how IBM has kept to the drumbeat of gaining market share, with IBM gaining 9 points of share overall since 2000 and gaining 21 points in the Unix racket, where it has become the dominant player. He did not brag about how well IBM did in X64 servers over the same term or in disk arrays, and that is no doubt because IBM has lost share there, although Adkins did say IBM gained 2 points of share in X64 iron in 2009 and a single point in disk arrays, too. (These execs all talk about market share and pre-tax income. That is, after all, what they get paid to capture. Unlike at Sun Microsystems, which Oracle got this year because Sun’s top brass didn’t understand they were in business to make money, not computers.)
Adkins, like the rest of the residents of the paneled rooms of Armonk, New York, is focused on growth markets, and said that a stunning 28 percent of hardware sales at IBM came from growth markets–Brazil, Russia, India, and China, and a few dozen other areas where their economies are hotter than in the United States and Europe–and that hardware revenue from these countries is expected to grow in the high single digits between 2009 and 2015. The BRIC countries represented only a little more than $1 billion in hardware sales in 2003 for IBM, but grew at a compound annual growth rate of 9 percent between 2003 and 2009; these four countries accounted for 14 percent of IBM’s 2009 hardware sales, and probably would have hit $3 billion last year had it not been for the economic downturn in 2008 and 2009.
So if you are wondering why IBM doesn’t seem to care as much about what is happening in Peoria, Portsmouth, or Paris, and is more concerned with what’s going on in Peking, Pune, or Parana, wonder no more. IBM is chasing growth, and if your business and data center is growing, then IBM is already talking to you, I am sure.
While IBM did not get into a lot of specifics, Adkins reviewed the midrange Power7 rack and tower servers and the entry Power7 blade servers that have been launched, as well as the Xeon 7500-based System x eX5 blade and rack servers, all of which have been covered in great detail in The Four Hundred. So I am not going over that again. Adkins said that in the second half of 2010, IBM would be announcing high-end and entry Power7 machines in the Power Systems family, as well as high-end System x boxes using the eX5 chipset and the Xeon 7500s from Intel. None of this comes as any surprise to anyone, particularly readers of this newsletter. And he mentioned the new System z mainframes that are coming out in the second half of the year as well.
And then Adkins raised the curtain on them just a little. The new System z, presumably called the System z11, will be an “ultra-fast and massively scalable” mainframe, and is being billed as a “system of systems,” by which Adkins means it will be a system with mainframe engines as well as integrated Power7 and X64 blade servers. The forthcoming System z mainframe will also have what IBM calls “special purpose analytic optimizers,” which will presumably allow mainframe data to be chewed on by Cognos and SPSS software, but not on expensive mainframe engines.
“This is really more than a mainframe,” Adkins explained, and I am going to quote him in full for reasons that will become obvious in a minute for students of the AS/400 architecture. “We will deliver on the traditional mainframe promise of value in terms of the improved performance and the virtualization and the management capabilities that we provide. But this system is a new type of system in terms of its ability to integrate across the data center. We’re calling it a ‘system of systems’ because when you think about how we are extending the quality of service capabilities of the mainframe–you think about things like availability, resiliency, security, management–we are extending that to an integrated, multi-architecture environment. I think this is going to be something unique in our industry because this really takes things to the next level in terms of how you get better consolidation across the data center.”
Uh-huh. Something new. Like the hybrid ClearPath mainframes from Unisys? Which have been around for a decade now, supporting MCP, OS 2200, Windows, and eventually Linux all on the same system with different types of CPUs plugging in, and which even run MCP in emulation mode atop Windows on the X64 side of the machine if you want. (An emulated version of OS 2200 running on X64 iron and atop a Linux kernel is now available, too.)
Something new, like the even older File Serving IOP co-processors that IBM slapped into the AS/400 back in 1994, when the 66 MHz 80486 chip from Intel was the hot thing and when having OS/2 LAN Server and NetWare meant it was mainstream. Or how about the Integrated PC Server or the Integrated xSeries Server co-processor that followed the original FSIOP, supporting Windows and Linux. IBM doesn’t do integrated co-processors in the Power Systems any more, but prefers to link external System x and BladeCenter boxes to the machines using iSCSI over Ethernet. But this is not co-processing so much as using the Power Systems machine as a very expensive disk drive for a Windows or Linux server.
I get the impression that IBM is going to be more tightly coupling the mainframe, Power, and X64 processors in the future System z11, if it is indeed called that, than just using a faster Ethernet link. Something more akin to the tight coupling that IBM originally did with the FSIOP, IPCS, and IxS co-processors before Big Blue got lazy.
The reason I bring this up is two fold. First, the minute you bring incompatible co-processors into a computing architecture, you are admitting to your customers and to the world that your platform cannot do the work for the same price–whatever that work is–and that you are therefore not going to run that work natively. Co-processors are the next best thing. If it is not conceding defeat, it is at least calling an armistice, a cessation of fighting. I happen to think that the Power Systems and mainframe machines should be loaded with co-processors and should have long since converged with X64 machines, so the technology could be shared as much as possible.
I have argued, in fact, that IBM should get Power and mainframe engines into the same sockets as X64 processors–it can be done–and come up with some clever BIOS software to let them all run inside the same physical box. If IBM had done that, it would have truly been something new to the industry. But having a legacy machine with a sturdy operating system take control over commodity X64 engines? This is not a new idea, but rather one taken from a machine architecture that Adkins didn’t even pay homage to.
That’s because this is what IBM is really focused on:
That, in one simple picture, is how IBM sees the server racket for the next five years. Transaction and database servers represent a $24 billion market opportunity with about a 1 percent per year growth rate and maybe $2.4 billion in profits. Analytics is a smaller market, at $8 billion in market opportunity in 2009, with only about $600 million in profits, but is growing twice as fast at a meager 2 percent per year. Web, collaboration, and infrastructure represented a $36 billion opportunity last year, according to IBM’s estimates, with $2.2 billion in profits and growing at about 3 percent–three times faster than transaction systems–and where Big Blue is number two, not number one. So expect lots of emphasis on System x and BladeCenter products. And hosting business applications, the mainstay of the AS/400 family, represented a $12 billion opportunity with about $1 billion in profits, but is only growing at 1 percent per year.
Adkins did not lay out how he expected Systems and Technology Group’s revenues to grow over the next five years, but he sure did talk about profits, as you can see here:
As you can see, IBM expects that it can take operating pre-tax income from $1.3 billion in 2009 up to $2 billion by 2015 (that’s about 6 to 8 percent growth per year). About 35 percent of that comes from market share gains in servers and storage (assuming a relatively flat but still anemically growing server and storage biz). Adkins said that IBM expected to gain 4 points of revenue share in servers and 6 points in storage in the next five years. That’s ambitious, but attainable.
Another 65 percent of the additional $700 million coming from “innovation and operating leverage,” whatever in heaven’s name that means. I think it means outsourcing a lot of engineering, manufacturing, and support to China and maybe even a new place, like Africa. I hear there’s real cheap labor there. After that, Mars.