Analysis: Sun sings a new tune with chips
- 09 December, 2002 08:07
Sun Microsystems has been criticized for having some of the slowest chips around, but the company now has revamped its processor plans with some novel new designs and retooled its server strategy in hopes of distinguishing itself from IBM and Intel.
In the past, Sun refused to utter a word about future processor plans, but a timely acquisition has led to the company being far more vocal about chips that will not appear until 2004. Sun is designing a number of specialized servers that will run on tight groupings of low-cost chips instead of its traditionally larger and more expensive UltraSPARC processors. The servers based on these new multicore chips each will be built to run specific types of software, such as application servers, Web servers and databases.
Sun has placed Andy Ingram, vice president of marketing in Sun's processor and network products group, at the head of its effort to convince users that this new class of multicore processors can revitalize Sun servers. Ingram has the task of explaining how Sun can shake its reputation for having slower chips than rivals by throwing groups of low-cost, specialized processors at high-end applications.
"In the past, Sun has taken a fairly traditional approach in that they didn't talk about their chip roadmaps," said Gordon Haff, an analyst at Nashua, New Hampshire-based Illuminata Inc. "But they are in a position where they are perceived, with some justification, as being a bit behind in the microprocessor game, certainly when compared to IBM. So now, they are talking a little bit more about future development, and it's no surprise that Andy Ingram, a guy who was aggressive pushing Solaris 9, is in this role."
The main driver behind Sun's new strategy is its acquisition earlier this year of Afara WebSystems Inc. This small company, led by ex-Sun UltraSPARC chip designers, has come up with a way to link numerous processor cores to form one more powerful chip. In many ways, these types of chips will be miniaturized versions of Sun's large SMP (symmetric multiprocessing) servers that have as many as 106 chips, Ingram said.
Sun is still keeping the "magic" behind the Afara technology a secret, but analysts suspect it involves fusing together simplified processor cores that run slower than large chips but are able to divide up workloads more efficiently. Most of the software that already runs on Sun's servers is tuned for an SMP system where different software threads can be spread across multiple processors, so analysts say this approach may succeed.
"For the kind of market Sun is serving today and hopes to go on serving, it is a very interesting approach," said Nathan Brookwood, principal analyst with the consulting company Insight 64, in Saratoga, California. "Commercial data processing typically doesn't require really massive single-processor performance. A lot of what goes on inside a system today is lots and lots of threads, each of which can do well on simple-minded processors."
In the meantime, Sun will continue to use its standard UltraSPARC line of chips to compete with IBM's Power4 and Intel's Itanium processor in the high-performance computing arena where a hulking chip is needed. Sun plans to roll out a dual-core UltraSPARC IV chip next year that supports multithreaded applications and is expected to fare well against its competitors even though it will lag in pure speed. This chip will be the main focus for Sun until the Afara technology is ready, most likely in 2004.
Analysts said Sun still has to prove it can pull off the Afara-based chip design and show that it will boost software performance, but even in concept alone, they said it could create an interesting choice for customers.
IBM's Power4 chip already packs two processor cores closely together, which has helped IBM create some of the most powerful Unix servers to date. However, the company has yet to describe any chip technology similar to the Afara designs, which can pack tens of processor cores together.
Intel, on the other hand, has said it won't be able to produce a dual-core Itanium chip until 2005 or 2006 due to the sheer size of its chip. It prefers to present Itanium as the answer for high-end systems and its 32-bit Xeon and Pentium 4 chips for the types of software that might be handled by multicore processors.
"Itanium is a very different approach," Illuminata's Haff said. "There are certainly differences as well with where IBM is going with multicore chips. It looks like Sun is taking a fundamentally more modular approach than IBM is with future designs."
Sun expects to make various kinds of specialized servers built around these new chips that will be designed to handle different classes of applications. If a customer wants to run an application server, for example, Sun might tweak the processor cores to be more suited for handling transactions. The servers for less demanding applications, such as Web serving, may use multicore chips with less shared memory, thus lowering the cost of these lower-end systems. For the most demanding software, Sun will make servers that include pricier technology for making sure chips can recover from errors and fail over to other processors if needed, Ingram said.
"I think what you will see from Sun is us beginning to deliver systems optimized for specific software implementations," Ingram said. "You want to get the highest performance at the lowest cost for each application tier."
Sun has some technological advantages in making this vision a reality, but it also has huge obstacles -- namely, IBM and Intel -- that it must overcome to keep its customers, according to analysts.
One advantage is that Sun develops most of the technology used in its servers in-house. Sun itself designs the chips, operating system and interconnect technology that powers its servers. It has worked on refining a tight relationship between all these products for years, giving it some of the most highly regarded systems around.
These are the kinds of skills that will be needed to create the types of specialized servers Sun is shooting for. This model of system design also contrasts to a company such as Intel, which has to rely on server vendors and on unproven 64-bit versions of Microsoft Corp.'s Windows and Linux to make its chip plans succeed.
On the negative side, Sun has to prove that the Afara technology can work and that it can compete on price with Intel and IBM. Sun pours hundreds of millions of dollars into designing its UltraSPARC processors, but as a company, it is only a fraction of the size of its competitors. Rivals question whether Sun can maintain this level of investment, especially in the current economic climate.
Sun also will need to attract new types of customers to keep its revenue high, as spending by the dot-coms and telecommunications companies that were the bulk of its business dries up.
"In terms of Sun's long-term direction, I think (the Afara technology) makes a lot of sense," Brookwood said. "I still have reservations about the overall economics of proprietary technology versus industry standards, but to the extent that Sun delivers something that has real value to its customers, this could be an interesting approach."
Sun's major tasks will be convincing customers to stick with it in 2004 and 2005 as Itanium begins to take off and the Afara technology is only beginning to appear, said Vernon Turner, group vice president of global enterprise server solutions at market research company IDC, in Framingham, Massachusetts. (IDC is a division of International Data Group Inc., parent of IDG News Service.)"IBM and Sun will have had experience (building) multicore chips, and it will offer up an interesting challenge to their customer base," he said. "Do you stay with their strong roadmap or are you compelled by Itanium, which could potentially be cheaper?"