Trying to boost the IT capabilities at his digital forensics company, Brian Dykstra invested in servers equipped with quad-core processors. After all, he figured, having more cores means a more powerful machine that can do far more work than single-core systems.
However, after shelling out money for the new technology, Dykstra found that three of the four cores were sitting idle because the software he was running wasn't built to make use of multiple cores.
Dykstra isn't alone in his disappointment with the lack of software for multicore chips. As hardware makers increase the number of cores in single chips, most software simply isn't keeping pace, creating a huge drag on efforts to take advantage of potentially significant hardware-based performance improvements.
In order to see that performance boost, software running on multicore chips must be built to let different cores handle different tasks in an application at the same time.
Dykstra noted that while some server software from major vendors such as Microsoft Corp. and Oracle Corp. has been partially multithreaded, there is a dearth of such applications.
Once Dykstra, co-founder and a senior partner at Jones Dykstra & Associates in Columbia, Md., had compiled a list of his firm's most critical software, he picked up the phone and started haranguing the vendors to add support for the chips. He didn't identify the vendors he contacted. Some IT managers have been able to cut costs and hardware needs by using the multicore technology in virtualization projects.
For instance, when a company virtualizes with multicore systems, each core is assigned its own virtual machine, allowing each to run a separate application.
Virtualization on multicore chips is working out very well for Bruce McMillan, manager of emerging technologies at the U.S. division of Solvay Pharmaceuticals Inc. in Marietta, Ga. He has scaled up his virtual machine total by 50% while cutting the number of physical servers in his data center almost in half.
McMillan said he had been running 100 virtual machines on eight servers with single-core processors. He added two dual-core servers about a year ago and was able to scale from 100 to 150 virtual machines.
About a month ago, Solvay installed a quad-core server and retired three single-core servers. The company is now in the process of adding two more quad-core servers, which will replace all of its remaining single-core systems, according to McMillan.
"It's saved me $500,000 just in hardware costs" so far, he said. "I can have much higher consolidation ratios than I had before."
McMillan said he's looking forward to getting more multithreaded software, but for now, he's happy that the multicore machines are allowing him to do more work with less hardware.
"It's a new level of scalability," McMillan said. "It's enabled us to really reduce our footprint in the data center. It's reduced our cooling costs. It's giving us less physical servers to manage. The maintenance contracts are cheaper. We're using fewer network portals because we have fewer machines."
The lack of multithreaded software certainly hasn't slowed the development of multicore processors by the world's top chip makers.
Just this month, Intel Corp. released its new Xeon 7400 server processor series, which includes six-core technology -- a new high-water mark in the semiconductor industry.
And while the step from quad-core to six-core processors was a big one, Intel is expected to soon go a step further.
Eight-core versions of the company's next-generation chip, dubbed Nehalem, are expected to go into production next year. The first releases of the Nehalem chip family are expected to be quad-core server chips that will ship in the fourth quarter of this year.
At the same time, Advanced Micro Devices Inc., still far behind rival Intel in producing chips with more than four cores, has released its own road map for pushing the processor envelope.
The Sunnyvale, Calif.-based company expects to ship its six-core Istanbul server processor in the second half of 2009 and a 12-core server processor during the first half of 2010.
And IBM is building supercomputers that run the eight-core Cell chip, which the company jointly developed with Sony Corp. and Toshiba Corp. to run large computations on Sony's PlayStation 3 video game system.
Possibly the farthest-reaching project is under way in Intel's research labs, where engineers are working on an 80-core processor. The company showed off the technology, still in development, at a conference in early 2007.
Though there have been no publicly announced plans to actually build an 80-core chip, analysts say the research into it hints at the future -- possibly the not-so-distant future.
"You know, at this point, everyone knows we're going to go up with the multicores: quads to six, to eight, to 12 cores," said Jim McGregor, an analyst at In-Stat in Scottsdale, Ariz.
"The road maps are out there for multiple cores," allowing IT managers to start planning to take advantage of the technology, he noted. "We know the track the technology is taking. This is an evolutionary cycle."
McGregor said he expects to see 16 cores on a chip in 18 months to two years. He noted that the chip-making industry is almost to the point where it's doubling the number of cores on processors every two years.
Value Proposition
The task for IT executives today is to try to determine how helpful that doubling -- or any increase in processor cores, for that matter -- might really be until the software problem is solved. Just ask Dykstra.
"It's really disappointing when you fire up a quad-core and then you see it's really only running on one core," Dykstra said. "All that extra money and expense, and you're not really getting a boost in speed."
In most cases today, only one core of a quad-core chip is used to run software "to its max usage potential, while the other three cores are just sitting there doing nothing," Dykstra said. Taking advantage of all the cores, he added, would boost performance by 300%.
It would also mean that data would be processed and sent out to customers more quickly, Dykstra said. "That's why we go to vendors and harangue them to do better."
But analysts say the effort to get multithreaded software could take longer than many IT managers hope.
The primary reason for this is that building multithreaded software is expensive. It's also a difficult task, especially for the many developers who learned how to code single-threaded software and have done nothing but that for years.
"We have a serious developer problem. People just don't know how to develop" for multicores," said Rob Enderle, an analyst at Enderle Group in San Jose.
"The environment has been single-threaded for so long that developers really haven't developed the skills. It's difficult to take things apart, make them run separately and then have them come together perfectly at the end," he said.
Margaret Lewis, director of commercial solutions at AMD, predicted that software companies will make major advances in writing multithreaded code within five years.
But even when multithreaded software starts flowing from vendors, it won't help the many large companies that run internally developed and legacy applications that can't easily be replaced. Eventually, organizations that want those applications to use multicore processors will have to replace the software or take on the massive job of rewriting it.
Today's volatile economy also isn't helping the cause of multithreaded software development, since companies must prove a strong business benefit when seeking to build or buy new technologies, noted Joanne Kossuth, vice president of operations and CIO at the Franklin W. Olin College of Engineering in Needham, Mass.
"In the economic arena we're in right now, cost is critical," she said. "How much will those 12 cores cost, and then what am I going to not be able to do? Will I be able to get rid of servers? Will I be able to consolidate?
"We can't just ask for the new toys anymore," said Kossuth. "There has to be a business application for them."