Solaris on Merced: What's in it for Sun? The technologist behind UltraSPARC says Sun's Intel deal will help, not hinder his microprocessor By Robert McMillan

January  1998

As chief technical officer with Sun Microelectronics (SME), Michael Splain should have a pretty good idea about whether or not Sun's plans to port Solaris to Intel's upcoming IA-64 architecture signifies the end of UltraSPARC. After all, he's the guy who creates the UltraSPARC roadmap. Splain met with SunWorld recently to discuss the impact of Merced on UltraSPARC. (2,000 words)

Mail this article to a friend

hough Sun Microsystems has ported Solaris to Intel's processors since 1993, last month's announcement that Sun would do a port of Solaris for Intel's next generation IA-64 architecture has been seen by some pundits as a first step towards abandoning the RISC (Reduced Instruction Set Computer) architecture. Some are even saying that Sun's days as a hardware vendor are numbered.

Hewlett-Packard, for one, is keen to propagate the idea that UltraSPARC is dead. HP Worldwide Marketing Manager Nick Earle claims that it has long been apparent that "the economics of the chip business would force Sun to drop SPARC." HP, which makes its own line of PA-RISC processors, maintains that it costs about $3 billion in R&D, design, and fabrication costs to produce one iteration of a high-performance RISC chip. And, says Earle, it was this massive up front cost that, back in 1994, caused HP to enter into partnership with Intel and codesign the IA-64 architecture, which is supposed to replace HP's PA-RISC line sometime after the year 2000.

Right now, Sun executives dispute HP's claim that the move from RISC to IA-64 is inevitable. According to Sun, the cost of one UltraSPARC iteration is in the low hundreds of millions of dollars.

It should be remembered that HP, at one time, also felt the costs of operating system development were too high. Two years ago it embarked on an ill-fated attempt to codevelop a 64-bit Unix, called 3DA, with Santa Cruz Operations (SCO). That initiative has since been shelved.

It's all in the optimization
"What happens with microprocessors is that we [various chip manufacturers] pretty much do similar things at a gross level," says Sun Microelectronics Chief Technical Officer Michael Splain, "and then all the companies go off and make their own optimization points." Splain downplays HP's claims that the first IA-64 chip, code-named Merced (expected in 1999), will have at least twice the performance of the forthcoming UltraSPARC III (expected in late 1998). "The question for Intel is going to be, 'What are they going to optimize?'" He continues, "What are they going to go after versus what we've gone after?"

While UltraSPARC has traditionally competed successfully on things like scalability, data management, and system and I/O bandwidth, IA-64 may render many of Sun's traditional advantages obsolete. It's hard to say for sure, notes Splain, because Intel is not expected to reveal many specifics about IA-64 until later this year.

While Intel's x86 chips continue to gnaw chunks out of Sun's workstation business, D.H. Brown analyst Tony Iams says that Merced systems, which are expected to excel in integer rather than floating point performance, will be more competitive in the server market. In fact, Iams says that the performance estimates he has seen show that UltraSPARC will still have the advantage over Merced in floating point performance, which is crucial in the workstation space. Iams says that UltraSPARC and Merced are expected to be equal in terms of integer performance. Of course, Intel, he says, "is going to have a lot of price-performance benefits."

Ubiquity
Sun says that Solaris on IA-64 will not be a repeat of the Solaris on IA-32 scenario. And there is reason to believe that this may be true. "Solaris for Intel, in the past, was really more of a technical novelty," says Iams. "Sun didn't have a formal relationship with Intel, and they didn't have a deal with any OEMs." With the IA-64 port, this has changed. Sun will work directly with Intel, and it has already signed up one OEM: NCR.

Splain says that Solaris on IA-64 will be good for UltraSPARC because, simply put, more Solaris equals a healthier OS community: more applications, more support, and more people who know how to use Solaris.

Though some have suggested that Sun will be selling two very different operating systems for SPARC and IA-64, both called Solaris, Splaine says he doesn't expect this to happen. Though Solaris for UltraSPARC and Solaris for IA-64 will not share binary compatibility, they will have the same "source base" (as does Solaris for IA-32). Sun says that the different Solaris's will be different only at the device driver level.

Sun executives are spinning December's Intel announcement as a triumph for the Solaris APIs, and confidently predicting, as they have been for months, that a number of unnamed system vendors, spurred on by both the promised availability of Solaris on IA-64 (expected with Merced, in 1999), and a keen desire to avoid the cost of developing their own Unix for Merced, will soon announce plans to ship IA-64 systems running Solaris. NCR Corp. was supposed to be the first in an avalanche of such companies, but just this week Sequent Computer Systems dropped a bomb on Sun's OEM efforts, when the company announced that after a year of negotiations with Sun, it had suddenly decided instead to partner with Digital.

Sequent VP of Marketing Jeff Pancottine says his company felt that Sun was committed to Solaris on UltraSPARC, over and above Solaris on Intel. "Solaris on Intel is not the same as Solaris on SPARC," he says. "If they [Sun's products] scale on Intel as well, what's the point of SPARC?" he asks. "They [Sun] have no real Intel strategy other than to port Solaris."

According to Pancottine, Sequent also wanted more control over the development of the operating system and a better NT integration story -- two things he claims his comapany got from Digital.

Solaris Director of Product Marketing Brian Croll says that Sequent's deal with DEC came at a cost too. He claims that Sequent will suffer by aligning itself with the less-popular Digital Unix. Not as many applications will run on its systems (Digital Unix has about 4,500 applications; Solaris about 13,000), and since Digital has a lower market share, quite simply, less people will know how to use it. "Sequent is really gunning for a high-end handcrafted boutique type of server," says Croll.

Interestingly, Digital, along with Santa Cruz Operation (SCO) may be Sun's strongest competition in this rush for Unix OEMs (in a sense, to be the Microsoft of the Unix world). HP is apparently not courting any, and IBM has yet to announce plans to do Unix on IA-64.

More Solaris system vendors would mean more applications, and the success or failure of Solaris on IA-64 rests on how well Sun develops these channels. "In the near term the goal is to sign up the OEM partners," says Jean Bozman, an analyst with International Data Corp.

So while the Solaris OEM channel strategy, despite its unremarkable history with IA-32, and the stumble with Sequent, could some day become a triumph for Sun, it is possible to also see the decision to port Solaris to IA-64 as a hedge. If Merced-based systems do end up threatening Sun's hardware sales, the company will now have its operating system up to speed for new Intel-based hardware systems. While she doesn't see the deal as an abandonment of UltraSPARC, Bozman says the idea of Sun selling Intel-based hardware is "a very real option."

Ed Zander, the president of the Sun Microsystems Computer Company subsidiary that would purchase these hypothetical chips, disagrees. In a recent interview with Reuters, he denied that Sun was toying with the idea of selling Intel hardware. According to Zander, "Sun doesn't do insurance bets. It doesn't do hedges."

The view from Sun Microelectronics
Splain puts it a different way. UltraSPARC doesn't have a "finite lifespan in the time horizon we can contemplate," he says. How long is that? Five years. "Beyond that," he pauses, and then says, "we just don't contemplate that far in advance." Five years from now is just about the time that Sun is expected to ship its UltraSPARC V systems, the most advanced SPARC processor on the drawing board today, according to Splain.

In that light, the difference between HP and Sun is that HP has not committed to producing RISC chips beyond its PA-8700, which is expected by the end 2000. Sun, meanwhile, says it will still be delivering RISC chips (the UltraSPARC V) three years after that.

Netra-SAP?
Though loathe to reveal specifics about how Sun plans to compete with IA-64-based systems, Splain says Sun has a number of options.

While UltraSPARC III systems will, in theory at least, scale to as many as 1,000 processors, Splaine admits that, at a certain point, simply scaling to more processors won't give Sun a great advantage. "I don't think we're going to find that scaling to 100,000 [processors] is a useful paradigm," he says. A more likely competitive advantage would come making systems that deliver better performance for specific types of applications. For example, "take the 1,000 [processors] and make them do something better" says Splain, "like run every MasterCard transaction in the United States in one computer."

Splain believes that benchmarks monitoring performance in specific applications will come to be as important as previous benchmarks such as dhrystone, specint, and specfp. In the "application performance era," that Splain predicts will soon be upon us, microprocessor performance will, more and more, come to be measured in terms of specific applications, like those from Baan, PeopleSoft, and SAP.

With control over the chip and system design, Sun could, in theory, have more maneuverability to optimize for specific applications. Splaine maintains that the plan is not to design application-specific systems like say a Netra-SAP, "it's more optimizing," he says, in the same way that the UltraSPARC II is optimized for database applications without necessarily being designed exclusively for them.

This has been Sun's competitive strategy relative to Intel for some time. Unlike Sun, Intel's business model is based upon "volume, volume, volume," notes Tony Massimini, the chief of technology with Phoenix's Semico Research Corp. He says that it is highly unlikely that Sun will decide there is no niche left where it can compete with Intel and suddenly become a volume systems vendor. "There are some things that are going to be of a more specialized nature that SPARC is better suited for," he predicts.

That Java niche
Another niche area Splain sees for UltraSPARC is running object-oriented applications. He hints that it will be interesting to see what impact object-oriented languages have on the chip industry. "The underlying way they execute on hardware is a little bit different than C," he says. "They have more methods, they're called more frequently, and they put a little more stress on the memory system than procedural languages."

Of course, it's hard to hear someone from Sun talk about object-oriented languages without mentioning the word "Java," but Splaine doesn't bring up the "J" word on his own. When prompted, he replies, "we think there are some good opportunities there." He says that a lot of the "thinking" from the design of Sun's MicroJava 701 processor is having an effect on the UltraSPARC line, particularly in the post-UltraSPARC III time frame.

Ways of doing faster thread synchronization and reducing application hang time during garbage collection are "just the early examples" of how Sun's processors could be optimized for Java based computing -- something that Sun hopes will become more important to the UltraSPARC world as its Java for the Enterprise platform matures. And if this goes as Sun plans, with important enterprise applications coming to rely on server-side Java performance, then Splaine's group, with its sister company JavaSoft just down the road, stands to gain an important competitive advantage over the Intel camp.

Of course, that's a pretty big if.

It is a fine line that Sun must walk. Its software arm needs to convince potential OEM customers that Solaris on IA-64 will not be the same as Solaris on IA-32. Sun needs to popularize those Solaris APIs as an alternative to NT, and for that, the OS must be competitive on the Merced platform. At the same time, Sun thinks it can best compete as a systems vendor with its own chip architecture -- and as such must compete directly with IA-64 systems.

So this is Sun's dilemma: one arm reaches out to embrace Merced, while the other fights it every step of the way.

Resources

• Sandpile IA-64 speculation http://www.sandpile.org/80x86/ia64/index.shtml
• Sun marketing material abou Solaris on Merced http://www.sun.com/realitycheck/headsup1218.html
• Sun Microprocessor information http://www.sun.com/microelectronics/products/microproc.html
• Hewlett-Packard http://www.hp.com
• Intel http://www.intel.com

If you have technical problems with this magazine, contact webmaster@sunworld.com

URL: http://www.sunworld.com/swol-01-1998/swol-01-ia64.html
Last modified: