IBM Gets Hybrid with Servers, Talks Up BAO Boxes
Published: May 18, 2009
by Timothy Prickett Morgan
I don't own any IBM stock, and you probably don't either. (Well, we probably do by virtue of the index funds most of us have at least a taste of in our 401(k) and IRA funds. But that doesn't count.) So you might be thinking that the annual IBM Investor Day event didn't really concern those of us who care about IBM's system platforms. On the contrary. The annual event is one of the few times when all of IBM's top brass are present and speaking their minds about the future.
Including Sam Palmisano, Big Blue's president, chief executive officer, and chairman, who doesn't make a lot of public appearances--just like his predecessor and the one who trained him for the top job, Louis Gerstner. But Big Sam was there in New York last Wednesday, and he kicked off the event, which brought together Wall Street analysts, industry gurus, and the IT trade and business press to hear many hours of presentations by all of the general managers of IBM's various hardware, software, services, and research groups.
And while IBM is still the largest systems company in the world, at least when ranked by revenues, I can tell you one thing: IBM is focused a lot more on cash flow and profits than it is on being a hardware vendor that plays in all markets. Under Palmisano's stewardship, Big Blue has exited PCs, printers, and disk drives, mainly because the margins from these products did not justify the investments. And Palmisano is quite proud of these moves and the $20 billion IBM has spent between 2000 and 2008 to flesh out its services and software portfolio. Palmisano said that the macroeconomic environment of the past eight years created an opportunity for the company, and one that it correctly seized despite much criticism about layoffs and offshoring and divesting of long-held IBM units.
"We divested of commodity businesses that do not recover their capital no matter how well you execute," Palmisano explained, and added that IBM stopped focusing on hardware sales in particular and concentrated more on outcomes for businesses that wanted to use IT to make their businesses better, not as an end in itself. The problem with hardware, Palmisano said, is "that as soon as you are under budget pressure, the focus shifts," and he added that hardware is the second thing that companies cut right after they slash advertising spending when the economy goes into a downturn.
And Palmisano was therefore not apologetic about getting rid of PCs, disks, and printers. "Obviously, the world is different from what it was in 2006," he said. "But I would argue that because we have done all of these things, we're not in bad shape." And Palmisano then went on to slam companies that "buy commodity stuff and integrate piece parts." That's a bit silly, of course, because a mainframe or a Power Systems server is, in fact, a bunch of commodity stuff and some special IBM hardware (z10 or Power6 chips) and software (z/OS or i, AIX, and Linux) all integrated in Big Blue's factories. (System x machines are built by partners, not IBM.)
IBM's chairman only danced close to talking about the rumored acquisition of rival Sun Microsystems, a deal that according to documents filed by Sun as part of Oracle's $5.6 billion buy of Sun, was originally started by Palmisano himself when he contacted Jonathan Schwartz on November 6 last year to talk about some sort of tie up between the companies as Sun's finances were plummeting and so was its share price. (The SEC document, a preliminary merger proxy statement, identifies the initiator of the talks as Party A, but I am certain that this is IBM; and I am guessing that the Party B identified in the document is Hewlett-Packard, but it could have been Fujitsu or Dell. I went through all the dances Sun did with its three suitors over at The Reg, which you can read here.)
Despite the downturn, which Palmisano bragged that IBM got ahead of and started to cut costs for before it hit in full force last fall--"call it insight, we did it, it is done"--IBM's chairman said that "it is a perfect time to prudently invest," but countered that you "don't do crazy deals just because they are available."
Palmisano explained that IBM was, like himself, an elder statesman of information technology. "We are not like the other companies in the IT industry," he explained. "We are not crazy kids. We don't throw numbers out just to throw numbers out." By which Palmisano meant that if IBM says it can do at least $9.20 of earnings per share in 2009 and follow that up with $10 to $11 per share in 2010, and even if sales drop by 7 percent this year and stay flat next year, well, my friend, you can just take that to the bank. And one reason why IBM is so confident in its numbers is that tired old saw, the annuity-like revenue stream that most of IBM's customers are contributing to. "We don't have the dependency on transactional deals that other companies have." About half of IBM's revenues and two-thirds of its profits come from these revenue sources, things like monthly licenses for mainframe software, software and hardware support, leasing contracts, long-term and short-term services engagements, outsourcing deals, and so forth.
"We are comfortable with our model," Palmisano declared emphatically. "We are not exuberant. We are not over-caffeinated."
(Speak for yourself, Big Sam. With my two jobs and taking care of the kids and the house, I sure as hell am. And out here among the 330 million people in the United States, I am not alone. It must be nice to not have to worry about anything, as IBM's top brass seem to not have to do.)
Palmisano talked briefly about one of the areas where IBM will be focusing its attention in the coming years, and one that cuts across all IBM groups and divisions: business analytics and optimization, or BAO for short. In plain English, this means IBM creating software that enables business to do real-time analytics and the big iron boxes that enable this analysis. Palmisano said that IBM was projecting that BAO sales would be as big as sales relating to enterprise resource planning (ERP) software for IBM in the next five years. And the difference this time around, of course, and something that Big Sam didn't point out, is that IBM has the application software driving this market, not SAP, Oracle, or Microsoft, and that software includes DB2, Cognos, FileNet, and a slew of other funky projects. (More on that in a minute.)
Frank Kern, the new general manager of Global Business Services (the $19.6 billion group that has the consulting business of PriceWaterhouseCoopers as its foundation), explained in his presentation at the Investor Day event that the BAO sales--including hardware, software, and services--accounted for about $105 billion in sales in 2008, a little less than a fifth of the overall market for a vague category Kern called "business automation" and which he said accounted for a $566 billion global market. IBM is excited because between 2007 and 2012, it is projecting that the BAO market will grow at a compound annual rate of 8 percent, twice the project growth rate of the IT industry overall. (And that is if the IT industry and the overall economy doesn't get worse, of course.) Inside of IBM's business, Kern said that BAO will grow at a 10 percent rate in 2009, hitting $2 billion in sales, and he added that growth will accelerate from 2010 and later to between 15 percent and 20 percent annually. Given this, within five years, BAO will drive as much sales for IBM as ERP systems do today.
Something is funky in that statement, because if I take $2 billion and grow it by 20 percent per year out to 2013, that is only $5 billion, and ERP systems account for a heck of a lot more than this amount of the company's $106 billion in sales in 2008. Suffice it to say, IBM is excited about BAO and will be giving some more accurate numbers as this market grows.
To give customers an idea of what kinds of systems IBM is talking about when it talks about BAO, Bob Moffatt, general manager of IBM's Systems and Technology Group, talked a lot about the kinds of diverse, hybrid systems that would be required by the "smart world" that IBM is trying to help the governments and companies of the world build out. Moffatt went on about how companies deploy multiple platforms and operating systems, that IBM would be focusing on innovations in systems design, virtualization, and workload management, and that Big Blue's servers and storage and operating systems were the "best positioned" for smart infrastructure and that IBM had the "strongest product line in the industry."
Moffatt gave some presentation time to cloud computing, as all IT vendors do these days, and said that the future of computing was to take "ensembles," collections of pooled servers and storage located in different parts of a data center, or in different data centers, and make them look like a system in the traditional sense of that word. He said that the future was virtualized compute and storage infrastructure linked by high speed interconnect--"we all draw it the same way"--and that the differentiation would be in how cloud infrastructure was implemented, yielding scalability, efficiency, security, and resiliency.
That all sounds pretty far from taking orders inside your RPG application, right?
Rod Adkins, who heads up development at System and Technology, talked about how IBM would be bringing to bear hybrid systems, which move away from IBM's general purpose system designs and integrate components that are highly tuned to run very specific workloads. But don't get the wrong idea. STG is still going to make general purpose machines like the Power Systems that run i 6.1. "Our investment will continue to be at all levels of the stack," said Adkins, from transistors and microprocessors on up through systems and up to operating systems, middleware, and databases. But IBM will mix and match these components and tune for workloads.
The two old examples, which IBM trotted out again, were the Cell-Opteron hybrid supercomputer called "Roadrunner," which broke through the petaflops number-crunching barrier last year, and the hybrid Cell-mainframe setup used by Hoplon International--a so-called "gameframe"--to run its 3D game environment.
"We believe that we will sustain out leadership position in the data center and in systems design," ended Moffatt, and he didn't say anything about Power6+, Power7, or any future chip roadmaps or the servers that they will go into.
The big news on Investor Day, and one that relates to the BAO effort as well as to system sales, had to do with a special box called System S, which is what IBM calls a streaming server. The System S machine is comprised of a BlueGene/P massively parallel supercomputer that is running a special set of software that IBM Research and Software Group have cooked up called InfoSphere Streams. The stream part is not a revenue stream, but data streams, and in this case, IBM means continuous data streams that are continuously analyzed to help computer systems and the people who access them make decisions.
InfoSphere Streams is a special set of code that can take information in from many different sources--news feeds, stock tickers, ERP systems, data warehouses, video feeds, audio feeds, whatever--and sort through that data for information on a continuous basis to make decisions. The idea behind InfoSphere Streams--and one of the underpinnings of the BAO opportunity that IBM sees--is that instead of running off batch reports or doing a query to ask a question of back office systems, a System S box will be given access to all the relevant data it can get its hands on and run queries continuously. As I had it explained to me by the IBM researchers behind the project, in a normal database query, you might ask to find all of the John Smiths located within a 100-mile radius of a certain location. But with a System S box, you could have it tell you as John Smiths come and go and move around inside of that 100-mile radius. This is a powerful concept.
This is one of the reasons why Toronto Dominion Bank took the first prototype of the System S box and plugged it in as the back end of its options trading system. According to its TD Securities unit, using the prototype, the company was able to create an options trading behemoth that could do 5 million options valuations per second based on a massive amount of real-world data, which was 20 times the number of valuations that had been previously processed by such a machine. And the System S was able to handle 21 times the data flow of the world record holder before now.
Anyway, last week, IBM said the System S machine and its InfoSphere Streams software were moving from prototype to commercial status, and said that customers who wanted to try out the software could get a copy of it from Big Blue on a trial basis. It is unclear if you have to buy a BlueGene/P super to run it, but the code runs on Fedora 8 Linux and presumably can also run on X64 Linux boxes.
The other machine, also a variant of the BlueGene/P, that will play into IBM's BAO plans is code-named "Watson" and it is running what is called a Question and Answer, or QA, system developed by IBM Research that is having its polarity reversed and its databases pumped so sometime next year it can go on the game show Jeopardy and play real human beings. The Watson machine is not a search engine in the strictest sense, and is programmed to listen to questions and give answers rather than all possible information that may be--or may not be--relevant to a question. (I wrote about Watson in detail at The Reg three weeks ago here.) It's an answer engine, I guess, and one that works in natural languages--in this case, English. Watson's software is based on code created by Software Group called Unstructured Information Management Architecture (UIMA), which is a framework for organizing data that is available through the Apache project
The tricky bit of code in the system is going through the mountains of data (mostly textual data in the case of Jeopardy) and scoring relevant bits of data that might be part of a Jeopardy answer (which is a question, not a statement) and then testing out possible answers and then figuring out which one is the answer (or in the case of Jeopardy, the question). And the machine has to do this in three seconds or less if it wants to beat the humans who can do this really well.
This may all sound like magic, and IBM knows it. So John Kelly, the scientist who runs IBM Research, said that Big Blue was not joking around any more than it was in the late 1990s when the "Deep Blue" parallel supercomputer was programmed to play chess at a grand master level and beat out Gary Kasparov.
"It is not a pipe dream," Kelly said. "It is up and running. It beats me, and it will probably beat most of you in this room."
Considering all the questionable financial decisions that Wall Street has made in the past decade, maybe that is not much of a test for Watson.
IBM turns back on server history: To give and to hybrid (The Register)
Beep, Beep: Roadrunner Linux Super Breaks the Petaflops Barrier
IBM Aims for Server Expansion in 2008, Including System i Reincarnation
Brazilian Game Site Chooses Hybrid Mainframe-Cell Platform
IBM's Plan for an Adjacent, Custom Systems Market