Summary
Larry Ellison will not kill and/or disinvest in Sun's Sparc architecture. 1) Oracle's data base software runs very efficiently on Sparc and especially on the newer CMT (Chip Multi-Threading) Niagara architecture. 2) The next generation 16 core CMT processor, code named ROCK, is due out this fall and incorporates innovative speedup technology that Oracle can leverage 3) CMT chips are much more power-efficient per unit of work accomplished than Intel or AMD based X86 architecture chips 4) Larry Ellison's core mantra is that hardware and software designed together is always more efficient and elegant than systems designed in isolation. ...And, Larry can do this without sacrificing Oracle's historically high margins!
Analysis
Contrary to what many analysts have been saying, Larry Ellison did not buy Sun just for its core software - The Solaris OS and JAVA. Rather, Larry is taking a much larger - and longer - view of what he wants Oracle to morph into.
Larry realizes that Oracle has not grown organically over the past five or so years, and it will not be able to change this trend. The Data Base marketplace for Oracle products is near saturation, and is growing very slowly. In this current economy, it is arguably shrinking. In order to continue the historical growth rate of Oracle, Larry has to change from being a pure software (Data Base and Middleware) play into a complete systems company. Attempting to do this on the X86 compute platform is contrary to Larry's stated intention to preserve his historical high margins. The Sparc architecture, on the other hand, need not be commodity priced provided it can bring unique and compelling features that are perceived to have large value to Oracle's customers.
Sun's current line of Sparc chips can be broadly classified into two types: Traditional Sparc architecture and the newer CMT (Chip Multi Threading) architecture. The traditional chips are now primarily designed and manufactured by Sun's partner, Fujitsu. Sun has kept the CMT designs completely in-house. It turns out that Oracle's software runs extremely efficiently on the CMT chips in particular, and with some additional work on both Oracle's part and Sun's could be significantly improved.
Coming down the road later this year is the next design center in CMT architecture, the ROCK processor. ROCK incorporates 16 cores and 64 hardware threads per chip; an architecture that Oracle can leverage to dramatically speed up data base performance in general and specifically in large data mining applications. Additionally, ROCK incorporates an innovative feature called the Scout. The Scout examines not yet executed future software instructions for code branches and predicts which way the code will branch. This allows the chip execution unit to load the proper next instruction(s) with a very high degree of accuracy, thus eliminating for the most part, the need to flush the instruction queue and try again; a very time consuming operation.
Now think Data Base query - It is almost completely comprised of a set of potential branches (tests) to either select an element in a table or not; in reality, the poster child application for the Scout.
If, for instance, the branch prediction rate could be improved from 50% to just 75%, compute efficiency would be improved by up to 50%. I suspect that most customer would be very interested in having a data intensive search run 1.5 times faster. I also suspect that they would be willing to pay a premium for this.
Now comes the hard part: How does Larry pay for a continuing (Very Expensive) stream of chip development and deliver good margins and make good on his promise to turn Sun cash flow positive to the tune of $1.5 Billion next year? Back at the beginning of this article, I mentioned that Sun had partnered with Fujitsu to continue the development of the older, traditional Sparc chip architecture. I suspect that Larry will turn to Fujitsu and expand the current relationship to include ROCK and its follow-ons. Additionally, I would expect the engineering relationship between the companies to deepen, with significantly more interaction between Oracle's core software team and the Fujitsu chip design team.
This potentially could give Larry the best of both worlds: A continuing exclusive platform to deliver his software on that he can sell at a a premium price, and an affordable method to continue the synergistic development of CMT style Sparc chips.
While this will not be easy to pull off, what with the no doubt tough negotiations with Fujitsu over fees, licensing etc., and the difficulty of establishing a close engineering relationship with Fujitsu, the rewards of success could be bountiful. The icing on the cake is that Larry also can leverage the Storage unit of Sun to optimize the back-end I/O capabilities of his systems to accelerate Oracle software on Oracle provided platforms.
In summary, Larry sees a path to build the "turbo" I-Phone of data base appliances; elegant, easy to use, power efficient and lightning fast. ...and Larry has never ever shied away from a challenge.
