The New Power Equation

For business-technology managers, the marching order for the past few years has been "Do more with less." Now comes a technology--multicore microprocessors--that does just that. The breakthrough chip technology is ready to deliver on its promise to provide more computing horsepower while taking up less space and generating less heat than conventional chips. Multicore processors, which pack multiple processing cores on a single chip, are transforming the design of data centers and providing a way for businesses to add more power and performance to their IT infrastructures without the problems that come with server sprawl.

The two major x86 server chip vendors, Advanced Micro Devices Inc. and Intel, have reason to believe that multicore technology will be adopted quickly and have a major impact on business computing. Sun Microsystems last week debuted servers that use its new UltraSparc T1 microprocessors, which have eight processing cores. And several other microprocessor suppliers are transitioning their products to multicore technology to increase performance while capping the heat generated by the chips. The chipmakers can "dial down" the clock speed of a processor slightly to reduce power requirements and associated heat dissipation, while increasing overall performance by adding two or more processing engines that perform tasks faster than their single-core counterparts.

The Tokyo Institute of Technology doesn't need persuading. When the school developed a plan to create the largest supercomputer in Japan and one of the largest in the world, using multicore processors solved the problem of having a limited amount of space for a data center and the related issue of heat generation inside the center.

"We envision a 100-teraflop-scale system running Windows, Linux, or Solaris, which is obviously applicable to researchers using thousands of teraflops of power," says Satoshi Matsuoka, professor in charge of research infrastructure at the Global Scientific Information and Computing Center at Tokyo Tech. "It's pretty obvious this capability wouldn't have been reached without using this dual-core processor technology because we are heavily constrained in terms of space and power budget."AMD began by delivering dual-core processors for the server market in April and got a market-share boost for its Opteron processor by being the first with multicore technology. Intel recently began shipping its first dual-core Xeons. Both AMD and Intel project that 90% or more of their server-processor shipments by the end of 2007 will be dual core. Startup Azul Systems Inc., meanwhile, offers systems based on a device that integrates 24 processing cores on a single chip.

Risky Business
Still, there's always a risk in being among the first to deploy new and unproven technology, even new chip designs from companies with a long track record like AMD and Intel. Flaws can show up as more customers use the chips for a variety of applications. And it will take time to show whether most applications can scale to take advantage of the performance boost that multiple cores and threads offer.

Multicore technology is showing up first in areas such as clustered computing, and for applications where the ability to handle high volumes of transactional data is crucial. Over the next year, however, multicore processing will spread throughout companies as the technology becomes the means for increasing performance.

Creating the Tokyo Tech supercomputer using single-core processors would have required a data center twice as large as the current one and would have generated nearly twice the heat, Matsuoka says. Just as important, a dual-core implementation has half as many servers, meaning it will be more reliable and less costly to maintain because there are fewer systems to monitor.

Tokyo Tech is among the first to begin installing servers from Sun that are based on its Galaxy platform, which uses eight dual-core AMD Opteron processors, effectively creating a 16-way system in an eight-way machine. The school is installing more than 5,000 dual-core Opteron processors, as well as accelerator-board technology from ClearSpeed Technology plc, which helps computers process large amounts of data in parallel at low power consumption.Multicore also is showing up in high-end chips. While Sun works closely with AMD on the dual-core Opteron chips, it's also pushing its own UltraSparc T1 processor. The T1 has eight processor cores, and each core has four independent threads, allowing systems based on the chip to theoretically handle 32 operations simultaneously. The multicore approach "will create interest from customers who otherwise would have avoided proprietary processors like the plague," says Nathan Brookwood, an analyst with Insight 64.Azul is pushing the multicore approach to even greater extremes with servers that incorporate the company's Vega processor, which integrates 24 RISC-based processing elements on a single core. Azul is offering systems that combine up to 16 Vega chips for a total of 384 processing cores, and it most recently added a platform that combines Vega-based servers with servers using Intel Xeon processors.

Pegasus Solutions Inc., which provides transaction-processing and E-commerce services to the hotel industry, is installing two Azul servers with a total of 192 processor cores, plus a 96-processor-core server that will serve as a testing environment.

Pegasus has seen use of its hotel-booking Web service increase about 25% annually, and in August it handled more than 1 billion transactions, creating 3.3 million bookings. The company usually spends around $4 million each year to increase its data-center capabilities. But the Azul machines will cut that outlay in half, says Steve Lapekas, Pegasus' chief technology officer. The new servers need less space and power than other servers, Lapekas says, and he can get the computing power he needs in one rack of servers instead of five.

"It's a no-brainer that multicore is what will eventually happen in the data center, but what Azul is doing is an extremely disruptive technology that has taken years out of the road map some other chip vendors are slowly rolling out," he says.

Is it too risky using a relatively unknown vendor to run the business' core systems? The Azul platform is optimized to run a Java 2 Enterprise Edition environment used by Pegasus and lets the company redirect to other projects the $2 million it saved by using multicore servers to expand its data center, Lapekas says. "We're not risk takers," he says. "You can't overbook a hotel like you can a plane, because no one is going to take a $200 voucher to sleep in the street."Most companies aren't ready to leap from dual-core to megacore processing, says Jason Waxman, director of server segments and technologies for Intel's digital enterprise group. "Customers, when they think about quad-cores being a couple of years away, start worrying about whether their applications will scale," he says. "Customers don't want to take a huge step back in frequency to get multiple cores on a chip, and they aren't convinced that the multithreaded environments and an ability to get to service-oriented architectures are so imminently achievable. They aren't ready for eight-core processors today. They want a transition that makes sense."

Reasons For Switching
Paradigm Geotechnology B.V., a provider of services and software used in oil and gas exploration, is moving to dual core to increase its computing capabilities and reduce electrical costs without increasing the size of its data center. As important in the final decision was an easy migration, says David Verdun, VP of R&D and technical support.

The company is switching out 280 nodes of a Dell PowerEdge 1850 computing cluster in its Houston facility with dual-core Xeon processors. Para- digm will be able to reuse the existing chassis that holds its single-core Xeon platforms. If all goes well, Verdun plans a similar size switch within the data center in 2006. "We're getting a good deal better throughput and performance, and it's a bit cheaper in cost and fits into our current infrastructure," he says.

Besides Houston, Paradigm operates data centers in Africa, Canada, and India, and also leases computing resources from outsourcing vendors. The Houston center was nearly full, so the company was faced with building an additional facility, increasing its outsourcing, or finding a way to better use its existing floor space. Dual core was the answer.

It may become the answer for more businesses willing to take some early-adopter risk in return for big benefits: eliminating the need to expand a data center, reducing air-conditioning requirements, improving system reliability, and providing more computing horsepower for lower overall costs.

Continue to the sidebar: More Cores For Handheld Devices