The crown jewel of the University of Maryland's supercomputing center is the "Connection Machine," a sleek, state-of-the-art supercomputer capable of executing close to a billion calculations per second.
Researchers from as far away as Tel Aviv, Israel, are using the supercomputer, manufactured by Thinking Machines Corp. of Cambridge, Mass., for molecular modeling, environmental simulation and numerical analysis, to name just a few projects.
If the university had bought the Connection Machine, or one like it, on the open market, the two-unit system would have cost about $2.5 million, estimates Larry Davis, director of the university's supercomputing center.
But the university has contracts with the Defense Advanced Research Projects Agency, the Defense Department's research and development arm, to work on projects. So the university was able to buy its Connection Machine through DARPA for about $250,000. As an added bonus, DARPA threw in two years worth of free maintenance, saving the university another $300,000.
With DARPA offering state-of-the-art equipment at fire-sale prices, Mr. Davis said there is little reason for him to buy commercially.
"No way," says Mr. Davis. Supercomputers are "too expensive. We would never buy from anybody but DARPA."
DARPA's approach to spreading the gospel on high-performance computing may be beneficial to universities and to the scientific community at large. The agency can champion new, high-risk technologies that industry might otherwise shy away from because their profitability is uncertain.
DARPA, for example, has been credited for pioneering efforts in creating "stealth" anti-radar and UNIX computer software technologies. Both have contributed greatly to U.S. defense efforts.
But critics say some DARPA practices are stifling competition in an industry still in its infancy. They fear that DARPA is turning into a kingmaker for a few select companies that specialize in making high-performance supercomputers known as "massively parallel processing" systems (MPP).
Left unchecked, critics say, such practices could seriously damage the viability of the market.
"Where it becomes touchy is when DARPA is no longer stimulating a market but impacting the overall competitive structure of the market," said Jeffrey Kalb, president of MasPar Computer of Sunnyvale, Calif., a manufacturer of MPP systems. "I think we've at least reached that point."
Repeated attempts to interview DARPA officials were unsuccessful. Stephen Squires, head of the unit in charge of DARPA's MPP effort, agreed to an interview but failed to answer or return repeated telephone calls last week.
At issue is DARPA's increasingly active role in promoting the companies it has chosen to advance the MPP concept. That short list currently includes Thinking Machines and Beaverton, Ore.-based Intel Scientific Computers, a division of Intel Corp.
DARPA is supposed to promote technologies -- not vendors. But critics say some DARPA practices suggest the agency is doing otherwise. Those practices include:
* Giving away millions of dollars in DARPA-funded machines, upgrades and maintenance for little or no cost to universities and research labs, the prime market for MPP systems.
* Lobbying would-be commercial customers to choose DARPA-endorsed machines over those offered by commercial vendors.
* Providing sales leads to representatives of DARPA-funded companies.
* Holding user conferences to stimulate interest in DARPA-funded machines to the exclusion of similar machines sold by competitors. Invitees include people from DARPA as well as non-DARPA-funded research universities and labs.
The irony is that these practices may stifle the very market that DARPA is trying so hard to stimulate.
All sides agree that DARPA has done a good job in keeping America ahead in the global technology race -- a feat considering that DARPA has a relatively small staff and limited funding.
DARPA, which has about 140 employees, has an annual budget of about $1 billion, slightly more than the cost of a single B-2 stealth bomber. DARPA has spent about $130 million a year on high-performance computing over the past several years, and its spending could rise to $238 million next year if President Bush's 1992 budget is approved.
DARPA was one of several government agencies to recognize early on that the MPP technology could help meet America's seemingly insatiable appetite for processing power -- at far less cost.
Some of the pioneering efforts into MPP technologies, at the California Institute of Technology, were funded by DARPA. Since then, about a dozen companies specializing in MPP systems have sprung up.
But interviews with more than two dozen computer scientists, government researchers and industry experts suggest DARPA may be unduly influencing the market it helped create. Some worry DARPA may wind up inadvertently squelching, instead of stimulating, the very market it has tried to foster.
Steven Wallach, co-founder of Convex Computers Inc. of Richardson, Texas, said his company had to do a last-minute rTC sales job to hold onto a customer that DARPA was trying to woo away. "DARPA was leaning on them to buy one of these two [Thinking Machines or Intel] machines," recalled Mr. Wallach. "We were successful that time."
Several other executives were willing to talk about their experiences with DARPA, but none wanted to be quoted by name for fear of reprisals by the agency.
One computer company claims it lost at least 10 sales due to DARPA meddling. According to an official of the company, developers proposed using his company's systems on 10 different occasions -- and were turned down by DARPA each time.
"They told us they were directed by DARPA to use a product from Intel," the official said.
One consultant said he knew of an incident where a research lab was taken to task by DARPA for submitting a proposal for a non-DARPA-funded machine. According to the consultant, a DARPA official called up the researcher who had placed the order and demanded to know why the proposal did not include one of the systems funded by DARPA. The matter has yet to be resolved.
"I have no problem with DARPA giving companies references,said one computer executive. "But when that reference also has a big pot of cash, well, that's when I have a problem. Not many of my references are going to pay to have someone use me."
Even DARPA supporters -- and there are many -- acknowledge the troublesome issue.
"There have been some definite perceptions that some companies know how to get DARPA's attention," said Charles Seitz, a professor of computer sciences at Cal Tech who has worked with DARPA on MPP projects for more than a decade. "I think that is only true a little bit, but it is true to some extent. Because what it comes down to is people, and some get along with DARPA, and some don't."
Though DARPA officials couldn't be reached for comment, one former DARPA official did step forward to defend the agency. The former official, who spoke on the condition of anonymity, said DARPA's seeming favoritism toward Thinking Machines and Intel isn't favoritism at all, just an attempt to make sure the MPP technology succeeds.
The official said being a good manager at DARPA means making sure projects are successful. That's because successful projects translate into future funding, and funding, after all, is what every government agency wants, he said.
Toward that end, he said, managers are expected to "nurture" technologies to make sure they can survive in the field.
In the MPP arena, that nurturing process varies from manager to manager. But he said it typically includes encouraging research labs to try out the equipment, holding user conferences and underwriting corporate brochures to make sure that users get the word on the latest from DARPA.
But DARPA doesn't discourage researchers from using the machines of non-DARPA-funded companies, he said.
"There's was never any intent, from our perspective, to say, 'Thou shalt not buy Machine X,' " the former official said.
Though industry may privately grumble about DARPA, nothing in the public record indicates that the agency treats companies unfairly.
The House Armed Services Committee, which has oversight responsibility for DARPA, has not received even a "single complaint" from industry about the agency's practices, a staff member said.
And companies that work with DARPA on MPP projects say the agency is even-handed in the marketplace.
Tim Browne, director of communications for Thinking Machines, said DARPA's seal of approval has helped in the marketing game, but only to the extent that DARPA's backing indicates Thinking Machines has a solid product.
"Having DARPA's help means the government has taken a long look and decided a particular technology is one that is worth investment for the future of America," he said.
Mr. Browne said he is sympathetic to the complaints of companies that haven't received funding and assistance from DARPA. But he said DARPA can't fund everybody.
"They simply cannot fund the amount of projects that they arconfronted with every year," he said. "I know of so many incredibly worthwhile projects in my own sphere that want DARPA help and didn't get it. But that's just the way it is."
Those views are echoed at Intel, which is working with DARPA on an MPP project dubbed "Touchstone."
Sig Lillevik, Touchstone project manager, said DARPA backing has been extremely helpful in developing Touchstone, but he said the project wouldn't go away if DARPA did.
"There are possibilities for funding from DARPA, generally speaking," Mr. Lillevik said. "But we would continue to develop these technologies one way or another."
Mr. Lillevik said Intel, like Thinking Machines, enjoys a certain amount of prestige because of its association with DARPA. Likewise, he said, that stamp of approval sometimes translates into marketing leverage for Intel.
"A lot of these folks are in markets that, when somebody gets a newfangled whatever, they all want one," said Mr. Browne. "It's like Air Jordan [basketball] shoes. When one kid gets them, they all want them."
The domino effect has not been lost on the competitors of Intel and Thinking Machines.
Even companies like Cray Research Inc., which doesn't yet compete head-to-head with the MPP makers, say DARPA's influence -- whether intentional or not -- could make selling a lot harder in the future.
Bill Bartolone, Cray's legislative-federal program director in Washington, asked, "Where the government is acting as an agent, how do we compete against that?"