|HOME||PUBLIC LIBRARY||ANOTHER PERSPECTIVE||INFOPERSPECTIVES||CONTACT|
IBM once tried to combine all of its midrange computers into one. The single machine that would replace all IBM midrange products of the early 1980s was called Fort Knox. Fort Knox, had it been successfully built, would have absorbed System/3 and its children, System/32 and System/36. It would have supplanted System/38 and its eventual descendants, the AS/400 and iSeries.
It would have picked up users of the 8100 and Series/1 minicomputers, which gave way to the RS/6000 and pSeries. And it would have replaced the low end of the mainframe line, which, at the time, was called the 4300, and which was in turn succeeded by the 9221, parts of the 9121 family, the small p/390 and Integrated Server, the Multiprise 2000 and 3000, the mainframe-on-Intel emulated by Flex-ES and the z900.
It could be argued that part of the Fort Knox plan did succeed, at least to an extent. The AS/400 initially emulated and later digested the System/3X family. More recently, a modest variation of the pSeries hardware has replaced unique machinery as the host platform for OS/400.
But what about the rest of the job? What about the mainframe zSeries running on the same hardware that supports the pSeries and iSeries?
What about a true successor to the concept that was once called Fort Knox and today might be called Fort Apache?
Fort Knox, Kentucky, is where the U.S. Treasury keeps a pile of gold reserves, although the bullion depository is not actually part of the Fort Knox military reservation. Uncle Sam also keeps lots of gold in other places, including the Denver Mint, the Philadelphia Mint, West Point, and the San Francisco Assay Office.
But naming a server after a place where gold is kept might not seem as apt today as it did a generation ago. With the server business in decline and the punishing costs of developing chips and whole computers looming large in the minds of IBM's bean counters, a host platform named Fort Apache seems more in tune with the times.
Fort Apache, students of history might recall, is a mythical police station in the Bronx, a nearly ruined oasis of dwindling authority surrounded by very hostile citizens, whose most prominent Irish cop was Paul Newman. It was brought to life in a film made in 1981, the very year the Fort Knox idea was rolling around IBM's subarctic development facilities in Minnesota.
In 1981, nobody inside or outside IBM could have predicted that the company might eventually need an all-in-one server platform in order to survive, but nobody thought Minnesota would get Jesse Ventura as governor, either.
As for the real Fort Apache — for there is one, in Arizona — it might have been situated in occasionally hostile territory, but it never suffered continual attacks from the locals the way the fictional one did. But I digress.
Speculation about a future computer using Power family chips that could support IBM's mainframe environments surfaces now and again. Although there is no solid evidence that IBM is getting ready to bring out a Power mainframe, Big Blue did take one real step way from manufacturing mainframes when it farmed out its most recent generation of big iron, the z800, to Hitachi. IBM makes the chips, while Hitachi makes the computers.
IBM still makes the older, but still more powerful, z900 mainframes. It can wait and see how things will work out with the z800 before it decides whether to build or buy complete machines when it comes time to field the next generation of its flagship processors. When it is ready to produce new high-end mainframes, it could produce the machines using unique chips like the circuits it currently uses in zSeries boxes. Alternatively, it could build its next big mainframes around the same Power5 or Power6 engines that will lie at the heart of future pSeries and iSeries machines. The technical aspects of IBM's future plans will be important — or at least seem so to some customers — but, for IBM, business considerations will be decisive.
Basically, IBM has a little problem. It might not be able to afford to build a separate bunch of machines to host its mainframe environments any more than it can afford to create hardware from scratch for the iSeries. To put some numbers on it, we turned to (or, more accurately, clicked to) a bit of research published by Merrill Lynch analyst Steven Milunovich soon after IBM made public its distressing financial results for the first quarter, which was also IBM's first quarter with Sam Palmisano as CEO. Merrill thinks IBM stock is a long-term buy.
Merrill's expert predicts IBM's Enterprise Systems Division (servers, storage subsystems, and networking apparatus), with reported revenue off 21 percent during the first quarter, will end 2002 with a 10 percent decline in sales.
Milunovich believes that only xSeries Intel-based machines will bring in more money this year than last ($2.5 billion), which was up by only 1 percent from the previous year. Sales of Unix pSeries iron will fall 6 percent to $3 billion, compared with a 5 percent fall the year before. Sales of iSeries machines will fall 30 percent to $1.5 billion, compared with an increase of 10 percent in the prior year. Sales of zSeries mainframes will fall 15 percent to $2.7 billion, compared with an increase of 10 percent in 2001. When it's all added up, server sales for 2002 will range between $9.5 billion and $10 billion, and the non-Intel portion of those sales will total between $7 billion and $7.5 billion.
That's a lot of money, but not necessarily enough to support two server hardware platforms, and certainly not enough, based on what IBM has done in the past, to support three. IBM can't just build servers when it pleases. It must also sell them, and it faces hard-working, fast-moving competitors across the board. So it's hardly surprising that industry observers think that, for mainframe hardware, the moment of truth is fast approaching. This would be IBM's moment of truth; the industry pundits don't have a moment for truth, and never did.
If IBM's mainframe team were faced with a choice between the company's own Power family chips and 64-bit engines from Intel, it would be inclined to select the proprietary technology. And that's if the decision were made in a vacuum, which it won't be. Other groups in IBM would want the zSeries to go Power, too.
The iSeries and pSeries teams, looking at spreadsheets predicting their likely sales results, are undoubtedly keen to have the zSeries folk using their hardware. They could sock another product group with a chunk of their R&D expense. They might even learn something from their counterparts in mainframe country.
IBM's product line teams have more in common every day. For instance, they all have to get their systems to support Apache. In this case, the Apache is an open source Web server named after the Native American tribe whose moniker was also co-opted by the two Forts Apache. Apache is the most popular Web server in the world, according to Netcraft, which knows about such things. According to Netcraft's most recent figures, there are more than 38 million Apache domains on the Internet, compared with 20 million running Microsoft IIS. By way of comparison, there are about 72,000 domains running Lotus Domino and about 33,000 running IBM HTTP Server (which incorporates Apache technology).
IBM servers also have to run Linux, which, of course, supports Apache. Whether they are running Linux or the operating systems indigenous to IBM, Big Blue's servers also support a version of WebSphere, the IBM application server product that uses, for basic HTTP services and other functions, none other than the same Apache core.
This appreciation of software that comes out of the open source movement is not news to IBM, nor is it surprising. IBM sees services as its greatest opportunity. Its services customers, who this year will pay IBM three times as much as all server customers combined, according to Merrill, pay one bill. They don't particularly care what hardware or software IBM actually uses to process their information; they just want results. IBM cares about results, too. That means Big Blue (unlike many of its server-buying customers who remain quite sentimental about the particular machines they own) will do what gives it the most bang for the buck.
Standardized hardware, whether or not it is actually built by IBM, is one part of the IBM profit equation. Standardized software, which makes it easier for IBM to get the most out of its services group experts, is another. And the more standardized IBM can make its hardware, the better it will be at getting, deploying, and wringing profit out of standardized software.
The whole process is made a bit easier these days because the open software folk seem to have mellowed with maturity, just like Geronimo, the famous Apache chief. The first sign of this was probably the deal IBM made with the Apache Software Foundation, home of the eponymous Web server, when it worked out a way to bring Apache code into its own servers, to add functionality and value without making WebSphere open, and even to use its own name on the resultant products.
Like the software Apaches, Geronimo showed, early on, that he didn't much care what you called him, as long as you spelled it right. Geronimo isn't even a Native American name. Geronimo is Spanish for Jerome, or, if you prefer the Latin, Hieronymus. Geronimo's real name was Goyathlay, and when he was a kid his home territory was part of Mexico. As land changed hands and territory was carved up, the area became part of New Mexico, and some of it, including Fort Apache, eventually became part of Arizona.
When he was a young warrior, Goyathlay's band had a big hit during one Mexican feast of San Geronimo. The Apaches took out a whole garrison of Mexican soldiers and got a nickname for the chief that stuck and reminded the Mexicans of the way San Geronimo blessed their soldiers.
Over the years, Geronimo was captured by the U.S. Army, escaped a couple of times, and in the process developed a hot-and-cold relationship with the fellow who ran Fort Apache, General George Crook. Basically, Geronimo didn't think the U.S.A. lived up to all of its deals, and from time to time he called the general, well, a crook. In the end, however, all was forgiven. In 1905, four years before his death, at 80, Geronimo participated in the inaugural parade for President Teddy Roosevelt.
As for the original Geronimo, the saint, he lived to 80, too, which is a long time for a Dalmatian, most of whom are run over by fire trucks. His longevity alone was quite an achievement for somebody born around 340, more than 200 years before Mohammed. San Geronimo was a translator, an exegete, and a critical theologian, and was a bit of a fighter himself. He spent much of his life on the lam and ended up in Bethlehem, which, at the time, was a lot safer for him than, say, Rome or Dalmatia.
If IBM, like Chief Geronimo, can ride to market victories on the back of Linux and Apache, with its battle cry, "Sure it's crap, but it's not Microsoft crap," nobody in the open source movement will particularly care what any derived products are called. And as long as IBM lets users of its various proprietary operating systems (OS/400, AIX, z/OS, and so forth) coexist with anything new, customers will be satisfied, too.
If IBM continues to bring all the clever engineering that originated in its mainframe hardware group over to the Power hardware people, as it has been doing for a while now, all the better. Partitioning, hot swap components, chipkill memory, and all the other things IBM has come up with for zSeries servers can only make the other IBM server lines better. Similarly, the clever tricks developed by the iSeries and pSeries designers can only help mainframe users get more from their machines — if they are carried over to the zSeries or the zSeries is carried over to Power technology.
So what would customers, particularly mainframe customers, think about their machines if IBM built them on the same frame that was used for iSeries and pSeries servers?
Like iSeries customers using what amounts to pSeries hardware, glass house shops would think about the innards of such machines about as much as they think about the guts of whatever mainframe they use now, which means just about not at all.
They will take a clue from the Native American Chief Goyathlay, whose name in Chiricahua Apache happens to mean "one who yawns."
— Hesh Wiener April 2002