But Wait, There's More

Gartner Says Server Revenue Up in 2003

While it's been known for weeks that server shipments climbed steadily in 2003, what was not so obvious was whether the actual money that those server sales generated would grow in the last quarter of 2003 and throughout the full year. As it turns out, at least as far as Gartner analysts can figure, there actually was revenue growth last year. This, perhaps more than any other factor in the IT market, is reason for a small amount (no irrational exuberance, please) of optimism.

According to the latest figures from Gartner, the worldwide server market grew from $44.1 billion in 2002 to $46.1 billion in 2003 across all platform types and sizes of machines. Gartner reckons that IBM put quite a bit of distance between itself and upstart Hewlett-Packard, which bought Compaq to try to get a bigger server business than IBM. So far, IBM is keeping ahead, with $14.8 billion in sales in 2003, up 10 percent over 2002. That gives IBM 32 percent of the server pie, and an increase of two points of market share. There will be plenty of bonuses going around at IBM's Enterprise Systems Group, which has been united with its Technology Group (see "IBM Explains Merged Systems and Technology Groups"). Gartner figures that, for 2003, IBM's sales of Intel-based servers were up 21 percent, to $3.4 billion, and that Unix sales were up 13 percent, to $4.1 billion. Across all platforms, IBM's Linux-related server sales were up 60 percent, to $552 million.

There will be some sighs of relief over at HP, too, thanks to the Gartner numbers. HP managed to grow server revenues in 2003 by 5 percent, to $12.5 billion, matching the market pace that it helps set. A cynic or HP competitor might say that HP kept down the class average, but considering the relatively difficult merger with Compaq, there is no way to look at HP's rebound in sales as proof that it has a handle on that merger. HP gained only two tenths of a point of market share in 2003, rising to 27.2 percent of worldwide sales. The company posted nearly $6.9 billion in sales of Intel-based servers, an increase of 17 percent, giving it a 33.8 percent share of this $20.3 billion market. HP's Unix sales were down 4 percent, to $5.3 billion, but it still had 31.7 percent of that $16.7 billion Unix server market. And across all platforms, HP also saw 60 percent revenue growth on Linux servers, with $927 million in sales. This gave HP a 33.5 percent share of that Linux server market.

Ranked by aggregate revenue, Sun Microsystems was still the number-three vendor by a comfortable margin over number-four Dell, but Sun's sales did decline by 15 percent in the year, to $5.5 billion. Sun lost 2.8 points of market share, which can't make Sun happy. But a lot of this water is under the bridge, and Sun has recognized that it needs to be aggressive in pushing Solaris on X86 platforms and still support its vast Sparc server installed base with enough price/performance improvements to keep them from jumping to alternatives. Sun doesn't really register as an X86 server vendor yet, but it still edged out HP for dominance in 2003, with $5.4 billion in Unix server sales and 32.6 percent of the market. That said, if current trends persist, the Unix market will be a three-horse race this time next year, with HP as the top Unix server maker and Sun and IBM neck and neck for the number-two spot.

Dell was once again the leader in terms of server revenue growth in 2003, pushing sales up 22 percent, to just under $4 billion for the full year. Dell gained six tenths of a point of market share in the X86 server market, with 19.6 percent of that pie. The company only managed to grow Linux sales by 40 percent, according to Gartner's estimates, to $521 million for all of 2003. However, that gave Dell an 18.8 percent share of the Linux server pie, which is roughly its share in the overall X86 server space. Interestingly, the Linux server market outside of sales by IBM, HP, and Dell is growing sales by 90 percent year-on-year, with $766 million in sales in 2003. The relative newness of the Linux server market and the enthusiasm of Linux on various special-purpose servers show that there is room for players besides the majors in the Linux space--at least for now, anyway.

VMware Tweaks GSX Server with Version 3

The VMware unit of disk array maker EMC this week announced enhancements to its operating system virtualization software for 32-bit X86 platforms, GSX Server 3. Soon-to-be rival Microsoft is apparently slipping as it tries to bring Virtual Server 2004 to market, so VMware has a bit of breathing space in a virtualization market it pretty much owns. That said, VMware is going to have to innovate like crazy to keep from being crushed by the onslaught of the Microsoft marketing machine.

A lot of the big innovations for VMware's virtualization products were brought to market last year in anticipation of an initial public offering that never materialized as VMware instead sold itself to EMC for $635 million as 2003 was winding down. As such, GSX Server 3 is not a big leap for the company, but VMware has to keep adding features that make its GSX Server and ESX Server virtualization tools easier to use and better at integrating with other tools for Windows, Linux, NetWare, and other environments so it can have a better story to tell than Microsoft.

To that end, VMware says that GSX Server 3 can now support a maximum of 3.6 GB of main memory per virtual machine, up from 1.2 GB with the prior version of the program. (A reminder: ESX Server is distinct from GSX Server in that ESX Server goes down to bare iron and creates virtual machine partitions that can support Windows or Linux. GSX Server runs inside an existing Windows or Linux instance and it creates virtual machines akin to a DOS Window running on Windows platforms. These GSX Server guest environments can run a full Windows, Linux, or NetWare operating system.) GSX Server 3 also offers 10 to 20 percent better disk and I/O performance, according to the company, and now has the ability to gang up network adapters to create large virtual network links spanning multiple NICs. The latest version also knows how to talk to SCSI tape backup devices, which is important for most commercial servers.

GSX Server 3 has also been integrated with the company's VirtualCenter systems management and provisioning tool, which allows system administrators to quickly provision and re-provision servers with any number of virtual partitions running Windows NT, Windows 2000, Windows 2003, Linux 2.4, and NetWare 4, 5 or 6. GSX Server 3 also now supports automatic startup and shutdown of virtual partitions, which means a partition doesn't have to be shrunk down (yet still be using resources) to allow other partitions to extend themselves dynamically as their policies and workloads demand. Now, if a partition is not doing anything useful, it is removed from the system and uses no resources. The updated GSX Server can also migrate virtual machines to ESX Server, which means companies who start out with GSX Server but who want the better fault isolation and control that comes with ESX Server do not have to start from scratch and rebuild their virtual machines.

One of the tricks with supporting so many guest and host environments is keeping up to date. GSX Server 3 provides support for Red Hat Enterprise Linux 3 hosts. VMware has added support for the following guest environments: Red Hat Enterprise Linux 3; SuSE Linux Enterprise Server 7 (patch 2); NetWare 6.5 Server; FreeBSD 4.6.2, 4.8, 5.0 and 5.1 (prerelease version); and Turbolinux Enterprise Server 7.0, 8.0 and Workstation 8.0 with GSX Server 3. The company is also supporting in "experimental mode" the future Microsoft "Longhorn" operating system as a guest and will also support the open source Linux 2.6 kernel as a guest.

GSX Server 3 will be available at the end of February for Windows or Linux host machines. Pricing has not changed from the prior release GSX Server 2.5, and still starts at $2,500 for a two-way X86 server.

IBM Explains Merged Systems and Technology Groups

IBM last week came clean on the fact that it has indeed merged the Enterprise Systems Group and the Technology Group, and hopes that the merger of the units gives it synergies and efficiencies that were not always possible when they operated separately. The Enterprise Systems Group develops and markets its four server lines and storage devices, and Technology Group has a big custom chip business and had a big disk drive business before IBM sold it to rival Hitachi last year.

At the moment, the Systems & Technology Group has two general managers, and each is maintaining control of his own turf. Bill Zeitler is general manager of the systems side, as he has been for years, and John Kelly is general manager of the technology side, a post he has held for a while, too.

In a conference call with Wall Street analysts last week, neither executive said anything about whether the merger of these units would mean that IBM could eliminate headcount, but they did admit that they no longer would have to do the intricate accounting work that has been necessary since the systems side of IBM has been one of the technology side's biggest customers--buying processors, custom logic, and manufacturing capacity to create its servers and storage arrays.

The Systems Group, which had external sales of $14 billion and pretax income of $2.1 billion in 2003, has about 13,000 employees worldwide. The Technology Group had about $2.9 billion in external sales and a pretax loss of $252 million; it has about 17,000 employees.

Kelly said that because IBM was ramping up production in its 300mm wafer copper chip fab in East Fishkill, New York, it had curtailed its chip business to only chase high-cost, high profit custom designs, and had sold off the unprofitable disk business (which went south largely because of failures IBM had in bringing a new generation of 3.5-inch disks to market). By doing this, the technology side of the IBM company is expected to be profitable in 2004. Just how profitable, we may never know, now that this unit has been merged with the systems unit. With the dollar weak, and server sales improving overseas, IBM could see significant improvement in server sales in 2004, and this will go a long way toward helping IBM's hardware business--of which the systems and technology units are the dominant component--look strong together. By merging them, IBM can disclose as much or as little as it wants about the specifics of how well different pieces of the Systems & Technology Group are doing.

That said, the merger of the two units, which actually went into effect at the beginning of the year, is about more than accounting. It is about having the two halves of IBM's technical brain know what the other is doing, so that it can remain focused on creating only those technologies that will deliver improved systems, and so it can use the vast engineering skills in the technology unit to more quickly bring promising technologies to market. This is why Sam Palmisano merged the four distinct server lines into the eServer brand in 2000 and eventually brought the storage business inside, too. By creating a single design organization for servers and storage across the zSeries, pSeries, iSeries, and xSeries brands, IBM could get more mileage out of its research and development investments, while beefing up all of these distinct products with mainframe-like virtualization and reliability features.

While IBM took a lot of flak for creating the eServer brand and the new names for what were the S/390, RS/6000, AS/400, and Netfinity servers (and rightly so, since changing the name of a product doesn't change what it is), the idea of making better servers by sharing technologies has made IBM's servers better products, and it has made IBM a stronger contender in the server market. It's funny, but what seems clear in hindsight--and what was not clear nearly four years ago, when the rebranding rumors began--is that the eServer campaign was more about changing the way IBM works inside than about presenting a clear marketing signal outside of IBM. IBM's server fiefdoms fought one another as much as they did competitors, and they did not share technology; they hoarded it greedily.

And if there is one thing IBM cannot afford, especially as it wants to push its Power processors into all aspects of computing--servers, embedded devices, game machines, whatever--it is a Systems Group that has no idea what the Technology Group has cooked up.

The most recent example of this, which I described in a story a month ago, when the rumors of the merger of these two units first came out, bears repeating because it describes how difficult it is for a large organization to keep track of what it is doing. Apple got the PowerPC 970 "G5" processors for its servers and workstations out the door well ahead of IBM in its own pSeries servers, BladeCenter blade servers, and IntelliStation workstations. AIX support for the PowerPC 970-based blade servers is not expected until late summer, and when these blades do start shipping on March 5, they will be able to run only Linux. IBM should have a PowerPC 970 line of entry tower and rack-mounted Unix servers to go after Sun Microsystems, but it doesn't, because IBM's systems people thought that Power4 and Power5 would be sufficient to compete. It is not. Further, Technology Group created the PowerPC 970 chip to give Apple a hot chip, and it definitely succeeded, but it did so without adding the special PowerPC AS instructions that would allow the chip to run IBM's own OS/400 operating system. That is 1980s IBM thinking, and it is as unacceptable as it is short-sighted.

Back in 2000, Zeitler said on the conference call, IBM had identified a bunch of discontinuities in the IT sector, and it created the Systems Group (which had had various names in the past) to chase those discontinuities. IBM predicted that there would be a massive shift from open system Unix platforms, which dominated the 1990s, to open source Linux platforms, which are taking a bite out of the 2000s. The company also predicted a big shift away from RISC and proprietary architectures and toward X86 architectures, which now include both Intel and Advanced Micro Devices variants of the platform Intel created. So IBM created a company-wide Linux organization and ponied up $1 billion to help foster the Linux market. The company moved its Intel-based server organization from the Personal Systems unit, where it had been since the 1980s, and put it side by side with its other servers. And it created a single server development team to work on technologies that would span its mainframe, Power, and X86 architectures.

By IBM's math, which is based on a four-quarter rolling average of market share, based on revenue statistics for the server market, this eServer and Linux strategy has paid off. When the eServer campaign was launched in the fall of 2000, IBM's server market share had bottomed out at around 24 percent. A year later, when the Power4-based machines were launched and the "Summit" chipsets for high-end Xeon servers debuted, IBM's revenue share had grown by a few points, and now, after two generations of high-end mainframes and Power4 boxes and a rapidly growing high-end Xeon server business, IBM's market share has crossed the combined Hewlett-Packard and Compaq, and has hit about 32 percent share in 2003.

Server revenue share, four-quarter rolling average (Source: IBM/IDC)

In the storage business, IBM's share dropped throughout the 1990s as EMC and Hitachi pounded on the company, particularly at the high end, but with the launch of the "Shark" arrays in 2000 and the subsequent "Silvertip" launch and the advent of very cheap, very powerful FAStT arrays for midrange users, IBM's share of the storage market has climbed from about 8 percent in 2000 to about 14 percent in 2003. To be sure, the new Hewlett-Packard (the merged HP and Compaq) and EMC are hovering at around 20 percent of the market and are climbing slightly because of revamped product lines.

Storage array revenue share, four-quarter rolling average (Source: IBM/IDC)

To IBM's way of thinking, it is making some ground in terms of market share, and perhaps more important, it is bringing more profits to the bottom line for systems and storage than either HP, Sun Microsystems, or EMC.

As IBM looks ahead into the remainder of the decade, it believes that the skills it has in creating and building very powerful processors that use less electricity and enable denser packaging are going to be key differentiators in the server and storage market. "Integration on the die will be at least as important as clock frequency increases were," Zeitler said. And he wanted to contrast IBM's approach to future processors from those of rival Sun, which has a line of future processors dubbed "Niagara" and "Rock" that will have relatively low clock speeds and a high number of multithreaded cores. "We believe that integrating more elements of the system onto a chip is more important than integrating a greater number of cores."

In any event, IBM's move from dual-core processors to system-on-a-chip designs is the main reason why all of IBM's future processors and system-on-chip designs will be created using a single engineering team that spans the formerly separate Systems Group and Technology Group. Each side of the company has skills the other needs in order to make system-on-chip designs. This team will be working on future Power6 and PowerPC 980 processors, as well as on variants such as the PowerPC 4XX core, which is at the heart of the Blue Gene/L supercomputer, which integrates two cores and five different types of system interconnections onto a single chip to create the brains in one of the fastest and most efficient supercomputers ever built. Kelly would not say much about the impending Power5 processors, except that it would have two cores, chip multithreading, and an integrated main memory controller. He said that the Power6 design is well underway and that prototypes of the chip are being created now.

"We're very confident of our ability to lead in performance and price/performance with Power5 and Power6," Zeitler said, showing off the latest pSeries 690 benchmarks, which demonstrate that a 32-core Power4+ machine running at 1.9GHz can meet the 1 million transaction per minute performance of a 64-way HP Integrity server using 1.5GHz Itanium processors, and deliver a price/performance that is 35 percent lower than the cost of the Itanium machine.

"We think we have the right model here," said Kelly. "We leverage what we do for ourselves and then drive it into our OEM partnerships." He said that IBM's OEM chip customers were encouraged by the merger of the systems and technology units, and that this gives companies like Sony, Nintendo, Toshiba, Cisco Systems, and nVidia confidence in IBM's commitment to its chip foundries and its Power architecture. While this may be true, only a few years ago IBM Microelectronics became the foundry for the Alpha processors from Compaq and the PA-8800 processors from HP; became the foundry for mainframe processors from Unisys; and did a lot of chip and board packaging work for Sun. IBM wanted a lot more business like this, but didn't get it. It seems likely that HP, Unisys, and Sun are not too happy about closer ties between the technology and systems side, but IBM has not become the raw component supplier to the server and PC industries, as it once dreamed it could be, and if it makes some customers antsy, that is just the price IBM will have to pay to do what is right for itself.

One of the discontinuities that IBM sees now--and which gives a company that moves fast a chance to get profits--has to do with the introduction of 64-bit Opteron X86 processors from AMD last year and from Intel this year, which will roll out its 64-bit Xeon processors next quarter, which are going to have a very important role in IT budgets in the coming years. For years, it looked like the Itanium was the only way to go for Xeon shops in the long run. But now that is not necessarily the case, says Zeitler, echoing an increasing sentiment in the server market. He says that Intel and HP broke one of the cardinal rules of computing--the 64-bit Itanium is not strictly compatible with the 32-bit Xeon--and that means it will be relegated to the high-end of the server market, where the superior design and higher costs of systems based on the Itanium chip, relative to that of the Xeons, will be necessary on technical specs and justifiable in economic terms. (We have covered the latest Intel 64-bit announcements, as well as its future roadmaps, in our Linux platform newsletter, The Linux Beacon, as well as in our new Windows platform newsletter, The Windows Observer.)

The other discontinuities that IBM is going to try to turn to its advantage have to do with Linux moving from being an option to being pervasive in the enterprise. Server architectures are moving from rack-based systems that are clustered to blade servers that are set up using grid technologies. And generally speaking, the value of server and storage systems is shifting away from processors, disks, and operating systems and toward the virtualization and management features built into these systems, which allow them to be used more efficiently, flexibly, and responsively. This is where IBM sees future profits.

The question, of course, is what can IBM do now on the profit front? When pressed to explain the decline in profits on the xSeries, pSeries, and iSeries in the fourth quarter of 2003, Zeitler said that there was pent up demand for a refreshed four-way Xeon box, and that IBM is delivering this now. For the pSeries, he said that continuing competitive pressures have pushed down prices on entry, midrange, and enterprise Unix boxes, but that IBM had been weak in the two-way and four-way markets. He added that the company had pushed out the pSeries 615, a low-cost two-way Power4 server, and was seeing traction with the pSeries 630, too. As for the iSeries, Zeitler said that the OS/400 server base has known that new products are due "in the not too distant future," and that this had caused some buyers to wait and see what IBM would launch--presumably the Power5-based "Squadron" midrange machines in April or May. The pSeries customers might be taking a wait-and-see approach on these boxes, too. When Power5 processors will make it into high-end iSeries and pSeries boxes is unclear, but the scuttlebutt is that it might not happen until later in 2004, since the larger processor complexes are much more complex and take more time to test and validate than the simpler processors used in the midrange Squadron boxes.