Guild Companies

Rumors about an impending convergence of IBM's various products are as old as the distinct tabulating, punch card, and computer lines that IBM has been selling for more than 80 years. There is a tendency in natural and manufactured systems toward specialization. Nature fills niches, and people like the right tool for the job. With computers, increasing specialization is mitigated by the desire to make as few unique products as possible so research, development, and manufacturing costs can be minimized, which increases profits.

There is always talk about convergence of server product lines at IBM, and this is especially true in the mature server market. The System/38 predecessor to the AS/400 was built in part from the remains of the failed "Fort Knox" attempt to create a unified System/3X, 9370, and Series/1 product line. Talk is cheap, but converging product lines while preserving the distinct character of them is not. Luckily, computer code is more malleable than genetic code, and it has been possible for IBM to merge various products under the covers. This is a good thing, since it allows IBM to keep supporting legacy applications. But, for some reason, the idea of a penultimate or final eServer convergence at IBM, which may be slated for the near future, is unnerving some people. The rumors come around every few months, and I've just heard the latest rumor about the so-called Project ECLipz, which is a pun on consolidating the iSeries, pSeries, and zSeries Power-based RISC platforms into a single chassis. There are those who think IBM can and should go further, putting even X86-style processors into a Power chassis to create what is a real, unified eServer.

IBM has been seriously on the road to convergence in the server business ever since it announced the PowerPC chip alliance with Apple and Motorola in October 1991. The PowerPC family of chips was designed to run OS/400, Windows, Unix, and MacOS from the get-go. Motorola somehow got the job of creating the 64-bit version of the PowerPC, called the PowerPC 630, and that chip saw the light of day for about an hour before it was killed because it was a terrible chip. That chip was supposed to be at the heart of IBM's Unix and OS/400 product lines, and luckily for Big Blue, the engineers in the Rochester, Minnesota, labs created a 64-bit version of the PowerPC chip that supported special OS/400 instructions. This chip, the "Cobra" PowerPC, eventually was extended to the Apache, Northstar, Pulsar, I-Star, and S-Star processors. With the Apache generation of machines, launched in August 1997, the high-end central electronics complexes in the AS/400 and RS/6000 lines were identical. Over the years, the AS/400's SPD bus (SPD is short for Systems Products Division, the name IBM used to have for its proprietary midrange server line) has been replaced with High Speed Link (which is probably nothing more than a Fibre Channel loop with some special OS/400 sauce) and peripheral cards are now PCI or PCI-X. Now the central electronics complexes and the I/O are the same. I would also bet Sam Palmisano's bonus check that the memory cards and many of the feature cards in the current iSeries and pSeries lines are identical, even if IBM pretends otherwise. Much of the packaging used in the pSeries and iSeries lines are the same, especially at the high end, although why they (including rack-mounted iSeries machines) have not been made identical is a mystery to me.

With the Power5 "Squadron" generation of machines next year, as was reported in our sister publication, The Four Hundred two years ago, IBM lays an abstraction layer on top of the Power5 hardware, called a hypervisor. This hypervisor is essentially an abstraction of a SMP server that can be partitioned and can load multiple instances of various, incompatible operating systems that run on the Power architecture. I've heard that the Power5 servers will be able to support 256 logical partitions. If IBM can match the partitioning capabilities it has for Linux on the iSeries, it should be able to support 640 partitions on a 64-way machine. Technically speaking, OS/400, AIX, and Linux will be able to be loaded into those partitions. The hypervisor layer will be under the control of a Linux-based service processor that is under the Squadron skins. IBM could, if it wanted to, sell a Squadron machine just running Linux. And if it saw a need for it, the company could also work with Microsoft to port Windows 2003 to the Power architecture and support Windows partitions. This would be a risky but bold move, much more brash than putting 32- and 64-bit X86 chips from Intel or Advanced Micro Devices under the skins of a Squadron box.

With all this talk of convergence that goes on year in and year out, I would remind you that just because IBM can do something does not mean that it will. There is no question in my mind that IBM could create a unified Power line of servers that supports Linux, Windows, AIX, OS/400, and even z/OS mainframe workloads if it wanted to. IBM could even put in adjunct Pentium, Itanium, or Opteron coprocessors inside the Squadron server pretty easily, since it has been doing on the AS/400 and iSeries for years. IBM could paint all of its machines red and sell them at the same price, too, regardless of operating system platform. I could also win the lottery and be hit by lightning.

IBM does not want to talk about eServer convergence, even though some of the big guns at the company occasionally do talk about the possibilities of eServer convergence and are misunderstood as saying this will happen. There are political factions within IBM who want total eServer convergence, and those (who like their jobs) who do not. I think it is inevitable, especially now that IBM has a new CEO and chairman and several new executives who don't have strong affiliations with the various eServer brands in charge of the xSeries, iSeries, pSeries, and zSeries units. Even still, IBM doesn't want to talk about convergence, and it doesn't want you to think about it, because then you might question its pricing and marketing practices. As I have said before, IBM is not interested in making any comparisons between members of its eServer product line, because technical comparisons invite economic comparisons that do not always make IBM's profit makers (the iSeries and the zSeries) look as good as its shipment leaders (the pSeries and the xSeries) in the eServer line. An uneducated--or, at least, unconcerned--consumer is IBM's best customer.

IBM has some history with the comparison problem, starting with the Apache AS/400-RS/6000 announcements. It was one thing to compare RS/6000s running 32-bit PowerPC chips to AS/400s running 64-bit Cobra or Muskie PowerPC chips, all with very different architectures. It was quite another to look at the "Raven" 12-way server and know it was exactly the same box, yet the AS/400 version cost a lot more than the RS/6000 version. The latest revamping of the iSeries line brings its pricing more or less in line with the pSeries line for equivalent technology--so long as you don't think about the wicked discounts Unix customers get versus OS/400 customers. I happen to think that any distinction IBM makes for the zSeries mainframes in order to make them different from the Power frames sold as iSeries and pSeries servers is there precisely so customers can't make direct comparisons. Look at it this way. A 32-way "Regatta-H" iSeries or pSeries server costs about $2 million, with about $500,000 of that going for a relational database and an operating system. A base 32-way "T-Rex" mainframe will probably list for $16 million, not including operating system and database fees.

You can see why IBM has not converged zSeries into pSeries and iSeries Power frames yet. But the day when it will not be able to invest the next $1 billion to make the next generation of mainframes is not too distant. I think that will happen with the Power6 generation, due around 2006, in fact. I think Power6 will have an emulated S/390 environment, much as Itanium servers can emulate 32-bit X86 and HP-UX instructions (albeit with very bad efficiency). At the prices IBM charges for mainframe power, if an emulated z/OS environment on a Power6-based machine ran at 50 percent efficiency--even 25 percent efficiency--that emulated z/OS it would feel like the bargain of the century, and it would have several times the processing power of the T-Rex zSeries 990 mainframe IBM will ship later this year!

In off-the-record conversations in the past several years, the top people at IBM that I have talked to have said that IBM would never just sell one eServer that ran all of those environments--OS/400, Unix, Linux, and z/OS. Most of the top server analysts say that they don't think IBM will do it, either. But that was then, and the computer business is in crunch mode. I think IBM will create one server line for these platforms based on the Power architecture and then only sell different operating systems on different brands--and at different prices. You won't be able to buy OS/400 for a pSeries, but you will be able to buy AIX for a partition on the iSeries. This will allow IBM to maximize profits and minimize manufacturing and development costs. That said, it will only be able to do this if customers don't freak out and demand that IBM truly deliver one eServer machine that uses all of the same features at the same prices. I think IBM should do this because I believe that eServer should be a product, not just some marketing. I think a real eServer at an attractive price would be a great benefit to all of IBM's customers.

If IBM doesn't converge its product lines, its competitors are going to do it and gain the financial benefits and customer enthusiasm. Unisys has been re-engineering its Sperry and Burroughs mainframes to plug those proprietary processors into its ES7000 Wintel machines, and it is pretty far along in the work. Hewlett-Packard has already created an Itanium-based server line, the Integrity line, which has one set of machines running HP-UX, Windows, Linux, and OpenVMS; high-end and entry machines have been launched, but midrange Integrity boxes have yet to be launched. Wouldn't you rather buy an eServer that could support multiple operating systems of your choosing than buy one that is stuck with only one personality for life? That's why people buy Intel-based platforms. They can change operating systems as needed.

It would be truly interesting to see a hybrid X86-Power hybrid server for the generic server market from IBM, much as Unisys is delivering in its ClearPath mainframe lines. While not trivial, IBM clearly knows how to make an SMP server that would support Power and Intel processors in the same processor complex. Such a machine would internalize the kind of distributed, hybrid networks that company struggle to manage today, and an integrated eServer machine that does it all is something that a large number of customers would probably want to buy to simplify their system administration.

The thing to remember as you hear all this gibberish about eServer convergence is this (and I will put it bluntly, so you know I mean it): No matter what, there is no way in hell that IBM will not deliver a Windows, Linux, OS/400, AIX, and z/OS platform of some sort. Your applications, no matter what platforms you use, will have a place to run. What it will cost you, well, that's another matter. Now is a good time to write to IBM and tell its people what you expect them to do. You're the customer, and that is your responsibility. You can never get what you don't ask for in this world.