Related topics

Fujitsu, Oracle ironing out Sparc server deal

A 'mutually complementary' relationship

graph up

Fujitsu and Oracle, newly in the server racket thanks to its Sun acquisition, are working on a new contract covering the development and sale of Sparc-based servers.

While Oracle may not be the mood to talk about its future Sparc and x64 server plans, Fujitsu's new president, Masami Yamamoto, wants to look ahead after taking the reins in April following the ouster of his predecessor, Kuniaki Nozoe, for alleged connections to the Yakuza, Japan's homegrown Mafia. Which may be one reason why Yamamoto let the world know about the cooperation between the Japanese server giant and semisecretive Oracle.

Yamamoto conversation with the Wall Street Journal revealed the answer to the question many of us had been expecting, given the big fat question mark on the Sun Microsystems and then Oracle Sparc roadmap: what happens once the 3GHz quad-core Sparc64-VII chips are delivered for the Sparc Enterprise M series of servers?

Well, it was not really a question mark so much as a big fat gray sphere marked as "APL2" in Sun's roadmaps from about a year ago, as El Reg revealed back in September 2009. APL is short for Advanced Product Line, which was the code name (however lame) that Sun and Fujitsu chose when Sun ditched its UltraSparc-V millennium chips in June 2004 and tapped Fujitsu to be its processor and systems supplier for midrange and high-end SMP boxes.

Sun went on with development of its entry and midrange Sparc T series multicore, many-threaded chips, and pushed more aggressively with x64 servers based on Opteron and then Xeon processors from Advanced Micro Devices and Intel, respectively. The plan was to get the initial APL machines into the field in the middle of 2006 or so, but the machines did not come out until April 2007. Neither Sun nor Fujitsu ever explained why, but the delay gave Sun's and Fujitsu's Unix competitors (mainly IBM and HP) some leverage at Sparc shops. As did the delay and killing off of the "Rock" UltraSparc-RK processor, which Sun quietly shot in the head last summer.

The original deal inked between Sun and Fujitsu back in June 2004 called for the two companies to work together through two generations of systems based on the dual-core Sparc64-VI and then quad-core Sparc64-VII processors. The idea was to tide Sun over until the Rock chips and their "Supernova" servers came to market. These were delayed again and again and again. Rock never made it, for reasons that Sun did not explain, and the Fujitsu line has sustained Sun's and now Oracle's big Solaris shops well beyond the expected term of the APL partnership.

Sun and Fujitsu have never been particularly clear about when and how they extended the original APL deal. However, John Fowler — who was general manager of Sun's Systems Group before the Oracle deal and who is now the top ex-Sunner to move to Oracle, taking the job as executive vice president of hardware engineering — told El Reg back in November 2008 that the APL partnership was set to run out and had been extended to 2012. He did not elaborate, but what seems to have happened is that Fujitsu stretched out the life of the quad-core Sparc64 platform, and hence Sun got more Sparc Enterprise M coverage.

Enter Yamamoto, who doesn't need permission from Oracle's top brass before he speaks and who told the WSJ that Fujitsu was negotiating a deal to extend the APL partnership. "We are combining Oracle's strength in application software with Fujitsu's hardware using Sparc processors," he explained. "Our relationship will be mutually complementary and will help us expand globally."

Yamamoto added that he expected to have a final contract done in a month, perhaps two.

He did not elaborate on what the shape of this deal would be, but Fujitsu was left holding the processor bag for Japan's $1.2bn Project Keisoku hybrid vector-scalar supercomputer, which was supposed to marry the vector savvy of NEC and Hitachi and the scalar oomph of Fujitsu to create a 10-petaflops super by 2012 to support research in Japan.

NEC and Hitachi bowed out of the Keisoku effort in May 2009, and it was a funny coincidence that the day before this happened, Fujitsu leaked details about its future eight-core Sparc64-VIIIfx processor, which will now be the sole engines in the Keisoku machine. Last November, the Japanese government was poised to chop the project's $300m budget for 2010 because of the country's fiscal woes, but in December, after what we presume was some pretty deft political jujitsu on the part of Fujitsu, the government reinstated the Keisoku budget, thus saving the Sparc64-VIIIfx processor from oblivion.

Now, Oracle and its commercial Sparc-Solaris customers (as well as those from Fujitsu) can get into position to benefit from the forward-thinking of the Japanese government, just like IBM's Power-AIX shops in industry benefit from all the money the governments of the world (mostly in the US and Europe) shell out for parallel Power boxes.

The fact is, Sun and Fujitsu have needed each other since the late 1990s, and both companies have been too stubborn and proud to admit it. Had they done proper joint development on Sparc for the past decade, Sun might still be a free-standing company. Then again, Sun might have followed the way of ICL, Ross Technology, and Amdahl, too, and been borged into Fujitsu. There were many paths for Sun. ®

Sponsored: Designing and building an open ITOA architecture