January 28, 2010

No Sign of HPC on Sun-Oracle Roadmap

Michael Feldman

After nine months of thrashing, Oracle finally completed the acquisition of Sun Microsystems. As expected, the EU approval of the deal last week cleared the way for the formal merger of the two computer industry behemoths in what is being described as the biggest IT deal of the 21st century.

Unfortunately for high performance computing lovers everywhere, there was nary a mention of HPC in Oracle’s merger party on Wednesday. The five-hour event, which was intended to provide an outline of the new company’s roadmap, mostly focused on Oracle’s infatuation with using Sun hardware to provide “complete business systems” and thus being able to compete with other vertically-integrated systems from the likes of IBM and HP. And while nothing was explicitly said to indicate Sun’s HPC customers would be ditched, Oracle’s focus right now is obviously on business computing, not scientific computing.

Specifically, Ellison and company plan to leverage well-known Sun technology like Java, Solaris, Sun servers, SPARC chips, ZFS storage, StorageTek tape systems, and Sun flash memory to build high-value enterprise systems and appliances. The Sun Oracle Database Machine, which combines Sun hardware (flash memory, servers, disk storage, and InfiniBand) and Oracle software, is the model they want to use to build a whole product portfolio of Oracle-branded business machines. In fact, the plan right now is to increase investments in these core Sun technologies to accelerate this strategy.

So what’s to be left behind? Not much, it seems. At the business level, Oracle wants to change the sales model so that it is selling directly to its top customers, rather than using the Sun partner model. Hardware-wise, it intends to back out of the low-end x86 server market. (According to Oracle president Charles Phillips, they’re going to leave that business to companies like Dell). And while Oracle will probably cut 1,000 employees or so, the initial plan is to hire another 2,000 new engineers and sales people to support the new roadmap.

A hint of what may be in store for the HPC business came from John Fowler, who was formerly the executive vice president of Sun’s systems group and is now the executive vice president of hardware engineering for Oracle. Like the other speakers at the event, Fowler reiterated the push to integrate Sun’s server, storage and network technology with Oracle’s enterprise software. At one point though, without mentioning names, he remarked they would “focus the product set,” rather than compete in every possible application area. You can read into that what you will.

Fowler also presented a roadmap, of sorts, for the SPARC chips. Both the UltraSPARC T series and SPARC64 series chips are in line for upgrades. According to Fowler, an UltraSPARC T3 is scheduled to be released later this year, with double the cores, bigger cache, and faster memory compared to its T2 predecessor. In addition, a new SPARC64 is slated to appear in the next 12-15 months. However, both chip’s lines are being groomed for enterprise duty: CRM, database, OLTP, and so on. Although HPC’ers have dabbled with SPARC servers from time to time, it’s difficult to imagine Oracle putting much effort into SPARC-based machines aimed at the high performance computing market.

It looks like Oracle will keep the Lustre file system software, at least for the time being. But there wasn’t much talk about it at the event. I wouldn’t be surprised to see the Lustre effort spun off to someone else if the company can’t find a way to leverage it in its business products.

In fact, it’s difficult to imagine an Oracle regime making a strong commitment to HPC in the near term. Certainly, if big customers want to upgrade their supers with Sun gear, Oracle will be glad to close those deals, as long as it can turn a profit. But in general, selling HPC servers is not the kind of high-margin business Oracle is used to and wants to continue. And right now, there’s no example of a public company that appears to be generating consistent profits from just selling HPC systems. (IBM, HP and Dell may be profitable as a whole, but don’t split out HPC revenue and profit numbers.)

If Oracle decides to make a push into HPC, I think that’s going to be down the road, after it’s matured its business portfolio a bit. If they prove me wrong tomorrow, so be it. But for now, color me skeptical.