Sun, Fujitsu to Partner for High-End Servers?

A Japanese newspaper reports the two companies may combine their server business as early as next year.

Last week, Japanese newspaper Nihon Keizei Shimbun reported that Sun Microsystems Inc. and Fujitsu Ltd. were discussing the possibility of combining their server businesses by as early as 2004.

Nihon Keizei Shimbun did not cite any sources for its story, and neither Sun nor Fujitsu has confirmed or denied the report, but analysts say that the proposed combination actually makes a great deal of sense.

This much is known: Sun CEO Scott McNealy met last week with Fujitsu Chairman Naoyuki Akikusa and other company executives. Sun did not respond to requests for comment, but a Fujitsu spokesperson explained that “Fujitsu has enjoyed a close relationship with Sun Microsystems and we have had a number of discussions about the benefits of working together to deliver the best solutions to our customers," but declined to elaborate further.

According to Nihon Keizei Shimbun, Fujitsu will assume the responsibility of manufacturing Sun’s high-end servers, while Sun will continue to manufacture its low-end servers. The two companies will also jointly develop SPARC-based microprocessors that will be manufactured by Fujitsu. Sun currently develops its UltraSPARC architecture; Fujitsu, an architecture called SPARC64. Sun’s UltraSPARC chips are manufactured by Texas Instruments.

There are obvious hardware and platform synergies—both Sun and Fujitsu develop microprocessors based on the SPARC architecture, and Fujitsu distributes Sun’s Solaris operating environment with its PrimePower Unix systems. Analysts say the arrangement could also work because Sun and Fujitsu aren’t competitors on the same scale as, for example, Sun and Hewlett-Packard Co.

“They cooperate in some areas, they resell parts of each other’s product lines in some areas, there’s some cooperation because Fujitsu has a certain base in Asia-Pacific, and Sun of course has a large base here in North America, so I can see them cooperating,” comments Richard Partridge, vice president of enterprise servers with consultancy DH Brown Associates.

Moreover, most analysts agree, Fujitsu’s SPARC64 chips are faster and more scalable than Sun’s UltraSPARC III microprocessors. Fujitsu should continue to outpace Sun even after the Unix giant introduces its UltraSPARC IV chip early next year. UltraSPARC IV is expected to debut as a multi-threaded chip – with integrated two UltraSPARC III cores—running at a speed of 1.2 GHz.

Fujitsu, for its part, has announced an ambitious roadmap that will see its SPARC64 chip hit speeds of 2.4 GHz by early 2004. SPARC64 is expected to be available as a dual-core chip by the end of 2004, at which time it will also run at 2.4 GHz – although Fujitsu says that it this iteration of the chip will reach 3.0 GHz before it runs out of gas. Fujitsu says that its seventh generation SPARC64 architecture will feature integrated four-chip cores and should reach speeds of 5.0 GHz by mid- to late-2005.

“[The edge in processor speed and performance is] definitely an advantage for Fujitsu, and it suggests at the very least the potential for some cross-licensing or some other kinds of agreements where Sun would use [Fujitsu’s SPARC64] on the high-end,” speculates DH Brown’s Partridge.

On the high-end, Fujitsu’s high-end PrimePower 2500 system currently scales to 128 processors; Sun’s SunFire 15K, on the other hand, can scale to 106 processors in a special configuration, although when Sun ships UltraSPARC IV, that number will effectively double.

Because of these factors and others, Gordon Haff, a senior analyst with consultancy Illuminata, says that an arrangement of some kind between Sun and Fujitsu makes sense. “It would be surprising if the higher ups in Fujitsu and Sun would not at least be interested in exploring the possibility, although the devil is always in the details,” he comments.

Haff suggests that as a result of the ambitious Throughput Computing initiative Sun outlined earlier this year, the company may have stretched itself too thin on the conventional UltraSPARC design front. In this respect, he agrees, there could be a role for Fujitsu to play: “People are already raising questions about whether Sun is capable of developing SPARC as we know and love it today, because Throughput Computing implies an expansion of Sun’s microprocessor development. So one could imagine a partnership with Fujitsu where effectively Fujitsu takes over that high-end, or part of that high-end, chip development for the Big Iron servers, really allowing Sun to focus most of its energies on the more Throughput Computing-type of approach.”

Not everyone thinks that a Sun-Fujitsu pairing is an automatic slam dunk, however. Nathan Brookwood, a principal analyst with microprocessor consultancy Insight64, points out that Sun can’t simply rip and replace its UltraSPARC chips with Fujitsu’s own SPARC64 microprocessors. “Clearly, if Sun was to go with the Fujitsu chip, it would require a dramatic reengineering of their entire product line,” he asserts. “Although both [chips] run the same software, the physical interfaces to the chips are very different, and therefore it’s hard to imagine that a Fujitsu chip would fit easily into the high-end, multi-processor system context that Sun has laid out.”

Because of this, Brookwood says, if Sun did enter into an agreement with Fujitsu, it would effectively be reselling Fujitsu hardware, rather than its own SunFire systems, which – in spite of their laggard UltraSPARC III chips – boast impressive performance and scalability in large SMP configurations. “Sun right now has a high degree of scalability with their UltraSPARC III chip, even though it may not do as well in the uni-processor benchmarks,” he concludes.

Another upside, Brookwood says, is that Sun could eliminate the approximately $200 million that it spends annually to support UltraSPARC design. It could better devote this money to ThroughPut computing and other endeavors, Brookwood says: “Sun is visibly pursuing … Throughput Computing, and that to me is a very interesting endeavor because if it works, it could be a very important breakthrough and nobody else is really pursuing a similar kind of approach.”

About the Author

Stephen Swoyer is a Nashville, TN-based freelance journalist who writes about technology.