|
Super contract for Cray Inc.
 |
Dave Kiefer, left, vice president and customer development representative with Cray, and Steve Scott, vice president and chief technical officer, were all smiles in November when it was announced that the company won the largest supercomputer research and development contract in history.
Photo by Candice Novitzke
|
|
|
By Mark Gunderman mark.gunderman@lee.net
Everyone in business who has information technology people on the payroll knows those invaluable experts don’t deserve the “nerd” and “geek” caricatures of popular culture. Still, one can’t help but smile and imagine guys throwing their shirt-pocket protectors in the air when hearing about workers nearly dancing in the hallways at Cray Inc., a few months ago.
There was cause to celebrate in November, because the company had just landed a big government contract.
OK, Cray has been landing big government contracts for decades. If it weren’t for government work, it’s doubtful the company could survive, according to vice president and customer development representative Dave Kiefer, the top man at the Chippewa Falls facility.
But this wasn’t just any big contract. This was the High Productivity Compiling Systems research and development contract for DARPA — the Defense Advanced Research Project Agency.
Still not impressed? It was for $250 million.
“This was clearly the most anticipated and closely-watched contract in supercomputing industry history,” said Kiefer.
It is the largest-ever computer research and development contract, and it went to just two companies. IBM got the other one.
“It means I have a job for the next four years,” said Steve Scott, vice president and chief technical officer.
Close partnership
No one needs supercomputer power more than the federal government with its broad reach into areas as diverse as energy research, space exploration, nuclear weapons testing and defense intelligence gathering. Money that the government spent fueled the supercomputer industry and private enterprise began finding applications for the most advanced systems.
“Originally Seymour (Cray) was designing machines for the government. There were no commercial accounts early on,” said Kiefer. “They were developed after 1980.”
Government purposes were similar to what they are today: weapons design at the Los Alamos National Laboratory, cryptography (code-breaking) and weather forecasting.
Those applications for supercomputer power continue to this day, but the government has added a long list of tasks it needs the world’s most powerful computers to do, Kiefer said. Those tasks involve climate modeling (read: global warming), energy research, and materials sciences.
“For the most part, it’s simulation. The biggest use of supercomputers is simulating things in the natural world,” Scott said.
“They’ve been used to replace experimentation,” Kiefer added.
Here’s a simple example but from the private sector: An auto manufacturer used to build prototype vehicles and crash them to see what happened. Now computer modeling saves a lot of time and wreckage.
For the government, the simulations are far more complex, just as Cray Inc.’s machines are more complex and powerful than ordinary high-performance computers.
Just because there’s a nuclear test ban treaty doesn’t mean the military doesn’t know what its nuclear weapons will do.
“There was a huge thing in the 1990s, in nuclear stockpile stewardship,” said Kiefer. “They were testing new designs and also testing the safety of the existing stockpile.“
“In the area of drugs, (supercomputers) replaced a lot of the wet lab stuff,” Scott said.
“The schedule is a large part of the impetus. It’s faster than experimentation,” Kiefer said.
Competition driven
In business, meeting customer needs drives industries forward. The other part of the formula, though, is competition, which also plays a role in progress.
The early Cray computers employed what are known as vector processors, and Cray Research dominated the field. Then competitors found ways to accomplish many of the same tasks by linking traditional-processor computers together in a system called massive parallel processing. People in the industry talk of such concepts, along with horizontal and vertical scalability, but what it all means is the same as in any industry: Trying to get a job done as fast as possible and in the most cost-efficient manner.
Companies like NEC, Fujitsu, Hitachi and IBM were slugging it out for a share of the market of which Cray controlled 70 percent.
Frankly, Cray didn’t always meet the challenges in the 1990s. Its parallel vector processors were less competitive as massive parallel processing came on strong. Cray got into the MPP game, too, and the T3E was the most successful machine ever to employ that technology.
But follow-ups were cancelled in a move some saw as a mistake.
The end of the Cold War didn’t help on the demand side, either.
“That had a big impact on Cray. That meant less revenue in the supercomputer industry,” Kiefer said.
Also working against Cray was the trend away from custom-designed processors to use of commodity processors in supercomputing. That is, using basically the same kind of processors available on the commercial market and going into ordinary servers and desktop computers.
“Commodity processors were becoming more powerful,” said Scott.
And IBM came on strong. “All of their systems used commodity processors,” Scott said.
By 2000, the market share picture was reversed, and IBM had 70 percent and Cray was among the others fighting for the remaining part.
Around that same time, the Japanese firm NEC made headlines with its famous Earth Simulator machine, at the time the most powerful and fastest computer ever built. It left an image that the American supercomputing industry was falling behind.
But the U.S. government had a vested interest in not seeing the nation fall behind in supercomputer advancements, and it was ready to put money into the industry as its post-Cold War needs grew.
And it certainly didn’t hurt Cray’s chances that the congressman representing part of the Chippewa Valley was Dave Obey, chairman of the House Appropriations Committee, then ranking minority member, and now chairman again.
The X-1
Everyone who follows the industry knows of SGI’s purchase of Cray Research, and how that brought SGI to the Chippewa Valley. Many insiders see that as a sort of down time for Cray, until SGI sold off the division to a company based in Seattle.
However, it was during the SGI days that the X-1 project was started.
In the past, Cray Research did business in the traditional way: Generally, it designed a product, built it, and sold it to a customer — in most cases, the federal government.
With the X-1, for the first time the government became a partner before the fact.
“Cray never received government funding before the X-1,” Kiefer said. “It was the first time there was co-development.”
It may have been in the backs of some people’s minds that the government needed to support companies like Cray, but market forces really drove the government’s participation.
What Cray was doing at the high end of the computer industry was so specialized that there weren’t a lot of applications for it in the commercial world, and therefore less incentive to develop products that would not have that wider market appeal.
It would have been a huge risk for Cray to use investor money to develop the products the government needed.
“They came to the realization that these machines really could not be made by the commercial industry,” said Kiefer. “Cray was the only company to do that kind of work.”
That work involved marrying the MPP system architecture and Cray’s signature high bandwidth vector processors. The result was great improvement on “the rate at which you could move memory around in the machine,” Scott said.
“The market for that kind of product was dwindling with the end of the Cold War,” Kiefer said.
“It had the capabilities they wanted,” Scott said.
“Even if it wasn’t the most profitable product on the market,” Kiefer added.
Setting the stage
The arrangement with the X-1 project set the stage for the dancing in the hallways in November. Government support of the supercomputer industry is a fact of life now.
According to Kiefer, if all of the government money went away, Cray Inc. would go away, too. Other companies like IBM would survive on their private sector work, but no one would be doing the product advancement on the high end that is driving the industry now.
Cray’s follow-up to the X-1 is called by those in-house “the Black Widow.” It was also developed with some government research and development money. It is shipping later this year to, as the saying goes, “an undisclosed location.”
The big project now is called XMT, for extreme multi-technology. The government is funding software development for it, and that’s where that big contract comes in.
Without getting into the technical aspects of it, basically the government presented to the industry a high productivity problem it wanted solved and waved a whole bundle of money for the companies that could solve it.
Big names like IBM, HP, Sun and others came up with ideas, as did Cray’s technical people. Cray’s and IBM’s proposals were selected in a highly competitive environment.
“It’s funding a whole product strategy called adaptive supercomputing,” said Scott. “It combines multiple processor technologies into a single system.”
And it uses commodity processors.
The devil, of course, is in the details and now Cray has to deliver a system that the government accepts, and (this would be really cool) makes the IBM guys say, “I wish we’d thought of that.”
Figuring it out will keep Scott employed for the next four years.
In the meantime, he and Kiefer and the rest of the computer geeks at Cray are working on the next level.
Scott was recently at a Department of Defense workshop where talk turned to government concerns, such as climate change.
And the computer power to do it will not be measured in the old terms like megaflops, teraflops and petaflops, but exaflops.
“It’s very exciting we are talking about computers at this magnitude,” said Scott.
Reaching the exaflop stage may cause more dancing in the hallways at Cray, but what does it mean?
Well, you really don’t want to know, but at Cray they’re hoping it means another big government contract.
Mark Gunderman is editor of Chippewa Valley Business Report. He can be reached at mark.gunderman@lee.net or at (715) 738-1607.
Email this story
Print this story
|
