When it comes to Intel-based servers in the data center, both IBM and Dell want your business. But the two companies have very different ideas about what you need in your data center---and what you're willing to pay for it.

Three years ago, IBM decided that soon--right about now, in fact--corporate data centers would start to consider Intel-based systems as an alternative to more expensive big iron. The big iron maker also decided that it wanted to be the undisputed Intel-based server market leader.

That ambition, IBM knew, would require a server that was faster, more reliable, and more scalable than those comparably equipped from the competition--but without the onerous premium customarily attached to such extras. So, the company locked a bunch of its engineers, some big iron technology, and an Intel roadmap into a hermetically sealed jar, and waited three years before prying the lid off.

Now, three years later, here's the burning question for IT managers: Is the jar's unique elixir--Enterprise X-Architecture (formerly code-named the Summit chipset)--just what the doctor ordered? Or is EXA over-engineered to the point that it's not worth the premium it commands? (That premium couldn't be eliminated completely.)

One thing is for sure. The games have begun. Thanks to the increasing maturity of the two primary Intel operating systems (Windows and Linux), IT managers--and in particular, those in charge of data centers--are indeed considering Intel-based systems as a data center option. Meanwhile, as a result of its merger with Compaq, Hewlett-Packard is way out in front in terms of Intel server market share. Dell is a distant second, and IBM is nipping at Dell's heels. That makes Dell the first hurdle for IBM and its EXA to overcome.

Dell considered EXA

That market dynamic is somewhat serendipitous, since Dell itself was considering use of the EXA chipset in the servers it sells before settling on Intel's E8870 chipset instead.

Although Dell Vice President Randy Groves acknowledges some of EXA's unique features, he says they're just not what the market is asking for. For example, EXA offers a remote I/O port that allows a server to easily be expanded by an additional 12 PCI slots. "That's nice," says Groves, "but with new interconnects like gigabit Ethernet and FibreChannel showing up in systems, the demand for PCI slots is going down, not up." Also, he says, licensing EXA from IBM would have been more expensive than licensing the Intel alternative. Paying more for something it didn't think its customers needed didn't make sense at Dell.

IBM differs with Dell over both cost and customer needs. IBM's director of high performance xSeries (Intel) servers, Doug Oathaut, knows how much a license to Intel's E8870 chipset costs because it was offered to IBM as well. According to Oathaut, "We were competitive with what Intel was quoting us. The chipset that [Dell] is using for their design is the same one that Intel is using for [Itanium 2]. We know from what [Intel] told us about [Itanium 2] that our pricing [for EXA] is on par with the cost of the Intel chipset." Perhaps IBM and Dell have different ideas about cost parity. Or perhaps Intel is offering a better deal to Intel-devotee Dell.

The price of the actual servers is where the rubber meets the road. However, it's almost impossible to figure out the prices for comparable (and data center-capable) systems from the two vendors. One reason for this is that Dell won't be shipping the systems that are based on the Intel E8870 chipset that it picked over EXA until later this year. Intel doesn't have the E8870 ready yet. So, if there's one advantage that IBM has because of its investment (and that Dell may have lost by not picking EXA), it's that IBM has some first mover advantage when it comes to four- and eight-way multiprocessor systems based on the latest multiprocessor-capable XEON CPUs from Intel (the ones that run at 1.4GHz). IBM expects to ship its 12- and 16-processor configurations--configurations that exceed the expected eight-processor capacity of the Intel chipset--later this year (perhaps another advantage for IBM).

That said, Dell spokesman Bruce Anderson took a look at the pricing that IBM gave me for two configurations of its EXA-based x440 server that it says are available: a two-processor configuration and an eight-processor configuration. In both cases, Anderson assured me that when Dell shipped its directly competing products, the prices would come in "several thousand dollars less."

While I trust my sources, bear in mind that a lot of this is vapor. IBM is saying it will ship its 12P and 16P configurations later this year. Dell says it will ship its systems based on the Intel chipset for less money, but also later this year. We'll never know for sure until all this happens.

In terms of customer need, Dell is no doubt in touch with its customers, many of whom are in the Fortune 500. But IBM has some history with that crowd as well, and the overarching goal of EXA was to prep the Intel architecture for the data center by introducing some of the features that data center managers like to see in their big iron.

To that extent, EXA does fill a gap that neither Intel nor any other chipset vendor has filled. But you have to dig deep to find that gap and to figure out how it sets IBM's offerings apart. For example, EXA has a fault-tolerant feature called memory mirroring that keeps the system from crashing after a memory bank fails. Dell offers the same feature, but only in its four-processor 6600 and 6650 models that use a chipset from ServerWorks. Dell's current eight-way systems (the rack mountable PowerEdge 8450, which is based on the older and slower XEON processors) are based on Intel's ProFusion chipset, which doesn't offer memory mirroring and may therefore not be as resilient as IBM's eight-way EXA-based offering. Groves says Dell's forthcoming eight-way servers will catch up when Intel ships its new chipsets.

Another differentiator: IBM's x440 server is the only high-end XEON server on the market that, once the 12- and 16-processor configurations are ready, can flexibly scale from four to eight to 12 to 16 processors without requiring the end-user to buy most of what's needed for the maximum configuration. Oathaut compares this to the approach that Unisys takes with its ES7000 where, he says, "regardless of the number of processors you need, you still have to buy a full-blown refrigerator-sized system with 96 PCI slots that accommodates the maximum processor configuration [32]."

Oathaut also says the data center crowd likes features such as the ability to predict component failures before those failures occur and hot-swappable memory. (The latter is supported only in IBM's EXA-based x360 and x440 servers under Windows; Linux support is coming.) Oathaut says features like this, which are built into EXA, are a part of IBM's company-wide eLiza project, an initiative to make all of the company's systems self-healing, self-diagnosing, self-optimizing, and self-managing.

Another unique feature of which Oathaut speaks fondly is the minimum of 32MB of Level-4 (L4) cache that can keep main memory from bottlenecking the performance of the Xeon processors, which only come with 1MB of L3 cache. "None of the other chipsets have that," says Oathaut.

But Dell's Groves thinks IBM's non-uniform memory access (NUMA) methods involve performance compromises that customers are not willing to accept. He says the performance of NUMA-based systems doesn't scale as well as non-NUMA systems. To support his claim, he cites the lack of any published benchmarks of EXA-based systems that go beyond four processors. Earlier this year, Hewlett-Packard raised the same issue.

Philosophically speaking, the two companies could not be further apart in terms of how to address the market. Each will go to great lengths to discredit and nab market share from the other. IBM, for example, likes to point out how it gained a fraction of Intel market share at Dell's expense as though its EXA investment brought that about. The truth is that the most recent numbers are probably based on the sales of pre-EXA systems and cannot be considered a foreshadowing of EXA's ultimate effect. But two things are for sure. First, both IBM and Dell have made strategic choices that carry long-term repercussions for the companies and their customers. Second, if and when the issue boils down to price, Dell is not a good company to get into a price war with (boy, talk about fighting a company on its own turf!). Perhaps two years from now, we will have a better idea of who was right.

IBM and Dell are betting they each know what's best for your data center. Why not tell them what you really need? TalkBack below or e-mail me at david.berlind@cnet.com