Mad Dog 21/21: Motherboarding
Published: January 7, 2008
by Hesh Wiener
In 1995, IBM shifted its proprietary midrange systems from CISC to RISC processor technology. Today, the System i is to a considerable extent a variation of the System p--or visa versa, depending on how you want to look at it. Last year, IBM sent a processor guru out to talk about a chip called z6, a sibling of the new Power6 chip that executes mainframe instructions. IBM will soon announce z6 mainframes that share a lot more than just some processor technology with the System p. They could easily be more than 50 percent System p, by weight.
The z6 mainframe will, of course, be a victory. It will quickly become the best-selling mainframe in the IBM line, if for no other reason than it will soon be the only one available.
Well, maybe not. What if customers aren't impressed? What if they don't order z6 machines and instead ask IBM to sell them a z9 or to turn on more engines in their installed z9 (or older) machines? Will IBM sweeten the z6 deal, making it pretty much irresistible? IBM can phase out the sale of whole z9 boxes pretty quickly, but it cannot easily halt selling upgrades of installed machines, particularly z9 models. That new motherboard had better be the mother lode, or IBM will have a tough time.
The victory, which we are certain IBM will declare no matter how events unfold, might turn out to be one like that of Pyrrhus of Epirus at Asculum. In 279 BC, Pyrrhus fought his second major Italian conflict in as many years. Pyrrhus had a great army, a disciplined cavalry, and at least a dozen war elephants. He was a very skilled general. So it was hardly surprising that he won his battle. But the victory came at a huge cost, and, according to the best accounts available, Pyrrhus said of the battle, "one more such victory would utterly undo him."
Pyrrhus of Epirus: A great general, and smart enough to know that one can win a battle and lose a war.
It took IBM a few processor generations to develop CMOS mainframes with engines as fast as those in their bipolar predecessors. Along the way, a lot of big users stuck with bipolar hardware. But during that transition period IBM was able to knock out its mainframe rivals, Amdahl (controlled by Fujitsu) and Hitachi. Amdahl got out of the mainframe game and Hitachi stayed in for a while as part of a deal that had it building IBM motherboards that were used in both vendors' machines. At the same time, IBM was able to crush the third party mainframe leasing and trading business, which had at its peak been able to affect IBM's pricing power in enterprise systems.
As it stands right now, IBM doesn't have much competition in mainframes. There is only one manufacturer, Platform Solutions, which IBM has frustrated with litigation, and an additional reseller, T3 Technologies, which offers smaller mainframe boxes that use Platform Solutions' technology and the same Itanium chips. There are also a few outfits trading used hardware that can at times offer better deals than IBM, particularly for use in disaster recovery applications. But it really doesn't add up to much.
If IBM is going to be selling mainframes against anything, it will be selling against installed machines a customer doesn't have to replace . . . and against alternative platforms, including its own System p and System i machines.
By making a lot of noise about server hardware consolidation--everything from IBM except X64 servers will become part of the Power box line--IBM will draw attention to something that it could just as easily downplay. The something IBM might well have tried to dance around is that the System p has overtaken the mainframe as IBM's flagship server, just as it previously overshadowed the System i in the midrange. The mainframe, like the System i before it, will retain unique characteristics, proprietary systems software and middleware and a cadre of third parties providing software, support, and services.
IBM wants to keep selling machines in all three server lines; it does not want all its mainframe and proprietary midrange customers to think about moving to System p machines. But it is nevertheless doing a lot of things that will tempt some users to consider migration, and, as IBM knows too well, when they do they might not restrict their thinking to IBM products. They might consider other Unix or Linux machines, or take a fresh look at Windows systems.
Still, the die is cast. IBM does not have another "real" mainframe up its R&D sleeve. It has the z6 or nothing. It's only a matter of when, not if.
Pyrrhus had more chances to regroup. After his battles in Italy, Pyrrhus was able to return to the Balkan Peninsula, where he rebuilt his army and reinforced his reputation as not only a great general but also an effective king. He even became strong enough to consider a return to Italy, or more accurately Sicily, which at the time was Greek in culture but under the political control of Carthage.
He won control of the island but before long he wore out his welcome by promulgating unpopular policies. He ended up with an insurgency that his army could not put down. By this point in his life he was running into more and more resistance no matter what he decided to do, as a frustrating campaign on the Italian mainland showed him. In the end it was back to Macedonia, where in the past he had at various times been a king and also persona non grata. This time Pyrrhus was on an upswing, once again headed for leadership. But with each successive phase in his career, Pyrrhus had to adapt to new conditions. With age and mixed experience, this apparently became more difficult.
IBM System p, System z, System i: Nobody cares if the skins are the same and hardly anyone cares if their innards are the same. . . but IBM is making a big deal about it.
IBM might be wondering how well it can adapt, too.
In the emerging generation of IBM hardware, Big Blue's large servers for the z line will end up in the same racks as the ones used for the System i and System p. The three lines will include as many common components as possible, which means that they will use parts designed primarily for the System p but built with z and i applications in mind.
If IBM could figure out how to do it, the company would make server racks that become p, z, or i boxes only near the end of the manufacturing process, when the motherboard was put in place. It would please IBM even more if it didn't even have to do that and instead just make a single line of servers and give each one the required personality with microcode, the way it customizes machines with coded engine and memory configurations now.
Even without going very far towards total unification, the next phase of consolidation will bring immediate benefits to IBM, which will be able to boil down its manufacturing processes and simplify its field support and maintenance activities. But that says nothing about the benefits to mainframe users or to users of IBM's other server platforms, only about the way consolidation helps IBM itself.
It's not that IBM has ignored the desires of its customers. It is just that so far, at least where the z6 is concerned, IBM has only talked about the new chip, not about the ways that chip might be good for customers.
Big Iron, Macedonian style: War elephants, the heavy weapons of pre-Christian Greece and Rome, but in Argos they couldn't get out of their own way.
IBM has done this by putting Charles Webb on the lecture circuit. He is an absolutely first-rate engineer, and as such he has been explaining how he got so many little clowns inside the funny car. This is not the stuff large enterprises really want to hear, except maybe after they have been told about the things that are truly important to them.
It's a bit frightening to see how IBM has managed to miss or ignore what customers think is obvious: Mainframe shops don't want anything exotic or unusual. They simply want better value, more performance, and the ultra high reliability provided by IBM's current machines. This turns out to be more or less what users of IBM's other platforms want, too. And it was what users of IBM's PCs and laptops wanted and, based on the way things turned out, didn't get enough of.
The PC users had an easy time switching over to machines made by Dell and Hewlett-Packard. Server customers generally have greater inertia, although it's less in the X86 and X64 realms than elsewhere, and less in the Unix space than in the proprietary realms of the z and i product lines. (That may be one reason IBM believes it can put z and i users on a forced march to new hardware without taking big risks, whether or not that theory turns out to be correct in the long haul.)
Clearly, IBM knows that its effort to move the mainframe to a hardware platform shaped by the Power6 circuits used in System i and System p machines and their surrounding technology will not automatically win the approval of the big iron shops that pay big bucks for IBM to meet big expectations.
For starters, customers who notice that the System z machines will have a lot in common with their p and i cousins might expect the use of common components to bring about big cuts in mainframe hardware prices, lower maintenance costs, and other benefits. But IBM is usually not so quick to cut prices and Wall Street would react badly if IBM's hardware margins took a big dive due to a pricing error on IBM's part. Overestimating the marginal elasticity of demand in the mainframe business--inaccurately projecting how much sales will increase in response of a price cut of a particular size--would be the kind of mistake investors would not easily forgive.
IBM might, then, include as much specialized hardware in the first z-in-p-box machines as it can, even though in the long run it would want to reduce the number of different components it designs, manufactures, and supports. It would then be able to point to hardware differences as a way to justify price differences.
Still, the hardware issues, which are the things IBM seems to be talking about so far, will turn out to be of secondary importance, or maybe no importance at all. The big issues for mainframe users contemplating new machines will be the same ones the AS/400 users faced when their servers went from CISC to RISC and then endured two rebranding campaigns, first to iSeries and then to System i. These issues came down to:
- What will it take to get software from older systems to work right on newer ones?
- What will use more time, money, and talent to run on the new boxes and what will cost less?
- Is there anything vital I do on my old computer that is impossible or impractical on the new one?
IBM did a pretty good job for AS/400 CISC users who were moving to the evolving RISC platforms. It's probably fair to say that IBM did a better job with CISC-to-RISC migration than it did helping System/36 users jump to the AS/400 when the AS/400 was announced as the successor to both System/36 and System/38 platforms. Part of that success came through a massive increase in the price/performance of the RISC machines compared to their CISC predecessors.
So we are confident IBM will do a very good job helping users of "real" mainframes get their workloads moved to z6 machines, which the company will insist are as real as any mainframe ever made. Some users will agree, some will not. In the end, all will have to face the fact that if a machine is the only one IBM offers, it is real enough.
It's a safe bet that IBM will succeed. It is clear IBM has thought about this quite a bit. But it is still a bet, not a certainty.
After Pyrrhus got frustrated with the results he was getting in Italy and Sicily, he returned to his native Epirus. But he was tempted to meddle in a situation in the Peloponnesian city of Argos. He went into that town with a bunch of soldiers and some elephants. It wasn't long before he found himself in a fight, one that he and his soldiers would normally be expected to win in a jiffy. This time things got confused, elephants got in the way, and a woman watching the battle tossed a roof tile into the melee. The tile landed on Pyrrhus's noggin. Pyrrhus was knocked down and the next thing he knew was the last thing he knew. Somebody came along and sliced off his head.
Post this story to del.icio.us
Post this story to Digg
Post this story to Slashdot