Technology has always been essential to military strength, but breakthroughs developed within the military often are not limited to weapons. This special report introduces some of the Pentagon’s most advanced information technology projects, in the context of their relation to commercial products and battlefield necessities.
[IMGCAP(1)]The Defense Advanced Research Projects Agency has fostered technologies ranging from the Internet to artificial intelligence research. Nowadays, the scientists it supports are pushing IT ever closer to achieving the processing power and cognitive awareness of living beings. At the same time DARPA is applying technology to the pressing threats imposed by current conflicts, the agency is sponsoring more than a dozen innovative projects, including a bid to perfect cheap, extremely accurate and nonradioactive atomic clocks for use in battlefield systems.
Advances in the mathematical algorithms for cryptography and the processing muscle behind them soon will transform the platforms that handle cascades of classified data, for example. National Security Agency officials characterize their work as a process of continuous ploy and counterploy in the rarefied realms of logic and computing.
The Grand Challenge of bringing practical, remotely piloted or autonomous land vehicles into use also is advancing via the competitive work of several teams. And in its approach to supercomputing, the Defense Department could be changing the way high-performance systems are measured, developed and purchased.
Mutating threats shape DARPA’s research in a wide range of new technologies
In a conflict where the biggest threats to soldiers often are low-tech, homemade explosives, it might not be obvious why troops need a more precise atomic clock to support their efforts. But the Defense Advance Research Projects Agency is working to deliver such precision, along with 13 other future icons that span a range of science and technology, from networking to air vehicles, biology and lasers, DARPA Director Tony Tether said.
The Chip Scale Atomic Clocks (CSACs), for instance, would perform key control functions throughout Pentagon networks and also could help warfighters detect an enemy’s presence.
All the Future Icon projects involve the application of computing resources to solve present and future defense missions, and some directly attack the problems of improving information technology performance for existing systems and futuristic computer architectures.
And they are the types of projects whose impact often extends beyond their original scope, affecting the development of technologies used elsewhere in government and commercially.
“They are tremendously difficult technical challenges that will be hard to solve without fundamentally new approaches — ones which may require bringing multiple disciplines to bear and perhaps even result in entirely new disciplines,” Tether said in testimony submitted recently to the House Armed Services Subcommittee on Terrorism, Unconventional Threats and Capabilities.
One of the most ambitious of the futuristic computer design projects is a five-year project to build a system modeled on the human brain, which would reflect and incorporate human assessments of the roles and intentions of people (see sidebar).
The research agency is also probing highly advanced IT challenges such as the Programmable Matter project, which aims to develop software that would allow physical objects to change their size, shape, color and other attributes to fulfill changing functions within, say, a military communications system.
CSACs would tackle more immediate concerns in defense networks and in helping soldiers detect enemy vehicles and facilities, according to a leading scientist at the National Institute of Standards and Technology who is researching the technology with DARPA support.
DARPA’s research is honing computer-based methods of detecting purposely hidden or naturally elusive enemy targets underground or on the high seas.
The CSAC project has been driven by the increasing need to reliably assure continual synchronization of systems linked via the Global Information Grid, said Thomas O’Brian, chief of the Time and Frequency Division at NIST’s laboratory in Boulder, Colo. The lab receives DARPA funding to support the development of chip-scale atomic clocks.
The tiny clocks could be deployed in hundreds of systems that military organizations at all levels rely on, including not only radios but also radars, sensors and location units that use the Global Positioning System, O’Brian said in an interview. The atomic clocks promise to make GPS systems more reliable while using little power, along with providing other helpful features, such as low weight and small size, he continued.
The CSACs “are significantly more accurate than the quartz crystal units ,which have been the standard” for such timekeeping, O’Brian said. The new generation of small clocks relies on the vibration frequency of elements such as cesium and rubidium to maintain their steady timekeeping and does not involve radioactive materials.
The tiny clocks can operate for as long as two days or more using the power available in a AA battery, O’Brian said.
“Another aspect of these devices is that they can serve as magnetometers,” he added. As such, the CSACs could sense the presence of metallic objects, such as mines or tanks. “You could scatter them across a wide area so when a Jeep or tank drives over, they might detect it,” O’Brian said. “Or they could detect the presence of ventilating fans in [al Qaeda caves] in Tora Bora [Afghanistan].”
CSACs already have proved themselves in demonstrations using GPS devices, and the technology showed that it could help navigation units function when satellite signals aren’t available, O’Brian said.
Some of the main tasks remaining before the CSACs reach routine use include:
- Developing efficient, low-cost mass-production methods.
- Improving the small clocks’ resistance to field conditions such as vibration, temperature and pressure variations and shock.
- Reducing power consumption.
O’Brian expressed confidence that researchers could soon achieve those improvements.
The research agency’s push in the fields of “detection, precision identification, tracking and destruction of elusive targets” has spawned several research projects. One group of them aims to improve methods for finding and investigating caves, and another centers on tracking seaborne vessels.
The cave research has gained momentum partly from the response of adversary countries’ forces to the success of the Pentagon’s spy satellite technology. Countries such as Iran and North Korea reportedly have built extensive underground facilities to conceal some of their nuclear-weapon production facilities from orbiting sensors.
The underground research spurred by such strategic threats also has led DARPA to study how better cave technology can aid tactical operations, such as by helping soldiers discover enemy troops and weapons lurking in small caves and by helping detect cross-border smuggling tunnels.
The Counter-Underground Facilities program aims at developing sensors, software and related technology to:
- Pinpoint the power, water airflow and exhaust vents from cave installations.
- Evaluate the condition of underground facilities before and after attacks.
- Monitor activities within cave structures during attacks.
According to DARPA procurement documents, the Pentagon’s cave program began by developing methods to learn about those conditions and other features of caves via Measurement and Signature Intelligence (Masint) technology.
Masint methods involve the use of extremely sophisticated and highly classified technology that can integrate information gathered by various types of sensors, including acoustic, seismic, electromagnetic, chemical, multispectral and gravity-sensing devices.
DARPA’s underground facility research project also involves investigation of the effluents coming from vents connected to cave complexes. Effluents for Vent Hunting research can involve the computerized evaluation of smoke to distinguish, for example, between decoy cooking fires and real cooking fires in an area where hostile forces may be roaming.
On the high seas, the Predictive Analysis for Naval Deployment Activities (PANDA) project is refining its existing technology to track the location and patterns of more than 100,000 vessels and to detect when ships and boats deviate from normally expected behavior.
As such, the PANDA research is similar to other systems that use exception detection to pinpoint unusual behavior by people in airports or train stations. Developers of those counterterrorism systems have carved out the task of teaching systems what types of events to watch for among the countless mundane activities observed via video cameras in the transportation hubs.
Like the PANDA system, the exception-detection software for airports flags unusual events — such as an errant freighter in one case or an unattended satchel in the other — and brings them to the attention of human analysts.
At the edges of computer science, DARPA is approaching the problem of attracting and cultivating talent to the field of computer science partly by asking promising students to choose projects that strike them as interesting and attractive.
“One of the ideas the students liked is Programmable Matter,” Tether told the congressional subcommittee members. “It is an important idea that is of significant relevance to DOD. The challenge is to build a solid object out of intelligent parts that could be programmed so that it can transform itself into other physical objects in three dimensions. It would do this by changing its color, shape or other characteristics.”
The programmable matter project could, for instance, lead to the invention of a malleable antenna that could change its shape depending on the radio or radar to which it is connected, Tether said.
“The computer science challenges are to identify the algorithms that would allow each element of the object to do its job as the object changes, while staying well coordinated with the other elements and functioning as an ensemble,” he added.
DARPA throws down the challenge on cognitive computing
The Defense Advanced Research Projects Agency’s research in the field of cognitive computing could progress to the point of a Grand Challenge that would pit alternate methods of building brainlike systems against one another.
The agency’s Biologically-Inspired Cognitive Architecture program is pushing artificial intelligence in the direction of building software that mimics human brain functions.
BICA relies on recent advances in cognitive psychology and the science of the human brain’s biological structure to build software that comes much closer to human abilities than previous AI. The research agency’s Information Processing Technology Office is leading the BICA research process by funding research teams based mainly at universities.
AI traces its roots back to designs such as expert systems and neural networks, familiar since the 1980s, which held out the promise of transforming information technology by adopting human learning and thinking methods. Those classic AI approaches proved to be useful in some commercial and government systems but were less effective than conventional IT architectures for most uses.
BICA’s leaders note that AI progress has been slow and steady in recent decades. “However, we have fallen short of creating systems with genuine artificial intelligence — ones that can learn from experience and adapt to changing conditions in the way that humans can,” according to DARPA. “We are able to engineer specialized software solutions for almost any well-defined problem, but our systems still lack the general, flexible learning abilities of human cognition.”
The BICA program has completed its first phase, which commissioned eight research teams to combine recent findings in brain biology and psychology to help build blueprints for functioning computers that could learn and understand like people. In the second phase of the five-year BICA program, which is now under way, the military research agency is seeking proposals for vendor teams to develop and test models of human cognition, or thinking, based on the architectures built in the program’s first year.
DARPA has not yet announced plans for a grand challenge competition to pit the resulting AI-like systems against one another. But vendor documents submitted in response to BICA’s first phase refer to an anticipated challenge stage of the program.
The University of Maryland at College Park provided one of the computer architectures for the first phase of the BICA program, basing some of its research on methods of designing a mobile system that could learn the various skills DARPA seeks in a cognitive system. “We are ultimately interested in [designing] an agent that captures many of the abilities of a child, and thus do not focus on a large initial knowledge base,” the University of Maryland computer scientists wrote.
“We keep the environment and input/ output to the system relatively simple so that we can focus on the primary issue of integrating those components and not the important but low-level details that will eventually need to be addressed,” according to their blueprint.
The 14 Future Icon technology areas, as described in testimony by Defense Advanced Research Projects Agency Director Tony Tether before a House committee:
Networks: Self-forming, robust, self-defending networks at the strategic and tactical level are the key to network-centric warfare.
Chip-Scale Atomic Clock: Miniaturizing an atomic clock to fit on a chip to provide very accurate time as required, for instance, in assured network communications.
Global War on Terrorism: Technologies to identify and defeat terrorist activities such as the manufacture and deployment of improvised explosive devices and other asymmetric activities.
Air Vehicles: Manned and unmanned air vehicles that quickly arrive at their mission station and can remain there for very long periods.
Space: The U.S. military’s ability to use space is one of its major strategic advantages, and DARPA is working to ensure the United States maintains that advantage.
High-Productivity Computing Systems: DARPA is working to maintain the U.S. global lead in supercomputing, which is fundamental to a variety of military operations, from weather forecasting to cryptography to the design of new weapons.
Real-Time Accurate Language Translation: Real-time machine language translation of text and speech with near-expert human translation accuracy.
Biological Warfare Defense: Technologies to dramatically accelerate the development and production of vaccines and other medical therapeutics from 12 years to only 12 weeks.
Prosthetics: Developing prosthetics that can be controlled and perceived by the brain, just as with a natural limb.
Quantum Information Science: Exploiting quantum phenomena in the fields of computing, cryptography and communications, with the promise of opening new frontiers in each area.
Newton’s Laws for Biology: DARPA’s Fundamental Laws of Biology program is working to bring deeper mathematical understanding and accompanying predictive ability to the field of biology, with the goal of discovering fundamental laws of biology that extend across all size scales.
Low-Cost Titanium: A completely revolutionary technology for extracting titanium from ore and fabricating it promises to dramatically reduce the cost for military-grade titanium alloy, making it practical for many more applications.
Alternative Energy: Technologies to help reduce the military’s reliance on petroleum.
High-Energy Liquid Laser Area Defense System: Novel, compact, high-power lasers making practical small-size and low-weight speed-of-light weapons for tactical mobile air and ground vehicles.
NSA pushes for adoption of elliptic-curve encryption, whose greater security and shorter key lengths will help secure small, mobile devices
The cryptographic security standards used in public-key infrastructures, RSA and Diffie-Hellman, were introduced in the 1970s. And although they haven’t been cracked, their time could be running out.
That’s one reason the National Security Agency wants to move to elliptic-curve cryptography (ECC) for cybersecurity by 2010, the year the National Institute of Standards and Technology plans to recommend all government agencies move to ECC, said Dickie George, technology director at NSA’s information assurance directorate.
Another reason is that current standards would have to continually extend their key lengths to ensure security, which increases processing time and could make it difficult to secure small devices. ECC can provide greater security with shorter keys, experts say.
The switch to ECC will be neither quick nor painless. It will require mass replacement of hardware and software to be compatible with ECC and new NSA cybersecurity standards.
In fact, the 2010 goal might not be realistic for NSA, where more than a million different pieces of equipment will need to be moved to ECC, George said. NSA’s move could potentially take as long as 10 years to complete, given the project’s complexity and scope. The agency has not set a specific deadline for completing its Cryptographic Modernization initiative, started in 2001 and recognizes that cybersecurity will always be a moving target, he said. The move to ECC is part of the initiative.
ECC, a complex mathematical algorithm used to secure data in transit, will replace RSA and Diffie-Hellman because it can provide much greater security at a smaller key size. ECC takes less computational time and can be used to secure information on smaller machines, including cell phones, smart cards and wireless devices.
The specifics of the changeover were announced in 2005 with NSA’s release of its Suite B Cryptography standards. Suite B falls under NSA’s Cryptographic Modernization initiative and details ECC usage for public keys and digital signatures. The announcement, the first related to cryptographic standards in 30 years, was a watershed event, said Bill Lattin, chief technology officer at Certicom, a pioneer in ECC.
NSA has licensed approximately 25 of Certicom’s ECC patents for use by the government and vendors that develop defense products.
The move to ECC represents a new way of doing business for the NSA. The Cryptographic Modernization initiative “is not just replacing the old with the new. We are upgrading the entire way we do communications,” George said.
Interoperability is the core of the new communications program and the reason for the modernization initiative. NSA plans to work closely with other governments, U.S. departments and agencies, first responders, and the commercial sector, George said. To do so, the agency needs public-key algorithms to securely transmit information among all parties, he said.
“If you go back 30 years, things weren’t nearly as interoperable as they are now. In today’s world, everything is being networked. We have to allow interoperability. And the cryptography has to match [among devices] because if it doesn’t, it is not going to be interoperable,” George said.
These interoperability goals will most likely extend across federal, state and local governments in addition to law enforcement agencies nationwide.
Although RSA and Diffie-Hellman are both public-key algorithms, experts say they don’t scale well for the future. To make RSA and Diffie-Hellman keys, which now can go to 1,024 bits, secure for the next 10 to 20 years, organizations would have to expand to key lengths of at least 2,048 bits, said Stephen Kent, chief scientist at BBN Technologies. Eventually, key sizes would need to expand to 4,096 bits. “That’s enormous keys. To do the math operations underlying the keys takes longer and is more computationally intensive,” Kent said.
Thus, NSA’s decision to move to ECC, which appears to be the only option. Experts agree that there is no new technology comparable to ECC. Although there are a number of protocols, there are only two basic technology approaches, George said: integers, used by RSA and Diffie-Hellman, and ECC, he said.
“ECC is the only impressive thing out there,” Kent said. “People don’t get excited every time a new thing comes along. We wait several years and let people try to crack it first. ECC definitely passed the test in this regard.”
NIST, which develops government- wide cybersecurity standards, also sees a need to move to ECC, although its recommendations are less stringent than NSA’s, whose ECC guidelines are a subset of NIST’s.
“I’m pretty sure [RSA and Diffie-Hellman] will be broken within a decade or so,” said Bill Burr, manager of NIST’s security technology group. “We are trying to end the use for most purposes of RSA and Diffie-Hellman with 1,000-bit keys by the end of 2010. And if you are real conservative, we are late.”.
“NSA has been fairly aggressive to standardize on ECC,” Burr said. We are slower, partly because we think it will naturally happen anyhow.”
John Pescatore, vice president and analyst at Gartner, does not see a need for the average user to switch to ECC unless it is to take advantage of its smaller size, such as securing cell phones and smart cards. With NSA, those technologies might include “things that a soldier carries around…and [has] strict limits on power consumption,” Pescatore said.
Burr expects ECC to become a universal standard by 2020, when most ECC patents owned by Certicom expire. “If it’s not a big problem today, it may be hard for the CIO to motivate people to transition to ECC,” said Kent.
DARPA’s Grand Challenge moves downtown, where teams will test their vehicles against city traffic
The Defense Advanced Research Projects Agency’s competition for autonomous vehicles has seen great leaps forward in its first two incarnations. This year, the ride could get rather bumpy, as the Grand Challenge moves from the expanses of the desert to the mean streets of the city.
The competition, called the Urban Challenge for 2007, is no mere sporting event. DARPA’s goal is to use the challenge to help develop technologies for self-guiding military vehicles that could reduce the deadly toll of vehicular-related battlefield casualties among U.S. military personnel.
Approximately half the U.S. soldiers killed to date in Iraq have died in enemy attacks on vehicles, whether by live enemy fire or by improvised explosive devices or, to a lesser extent, in vehicular accidents.
Based on results from the two previous Grand Challenges and a preliminary look at the entrants in DARPA’s Urban Challenge contest now under way, “we think that over time we will be able to build vehicles that will be able to drive as well as humans in certain situations,” said Norman Whitaker, program manager for DARPA’s Urban Challenge.
In May, DARPA trimmed the roster of teams competing in the Urban Challenge from 89 to 53 and will further narrow the field to 30 semifinalists this week based on scores issued during site visits DARPA officials have been conducting since May. The agency also will name this week the location of the competition’s Qualification Event scheduled for Oct. 26 to 31 and the location for the final contest Nov. 3.
To date, DARPA has said only that both events would take place in the western United States, although its placement in a simulated urban combat zone has become the theme of this year’s contest and considerably upped the ante for the level of vehicle proficiency that will be required to successfully complete the contest’s 60-mile course in six hours.
The complexities of a city environment and the introduction this year of other moving vehicles along the course increases exponentially the sophistication of the sensing, data processing and guidance technologies required, Whitaker said.
DARPA’s goal in its successive challenges is to raise the bar each time, he said, although the addition of moving traffic represents the biggest obstacle ever added to the contest.
The first Grand Challenge in 2004 ran over a 142-mile course in the desert, but the competition looked more like the Keystone Cops than Knight Rider — no vehicle made it past the eight-mile mark. Still, DARPA officials said they saw promise, which came to fruition in 2005, when four vehicles covered a 132-mile desert course. With those results, the decision was made to take the Grand Challenge downtown.
With an urban setting and traffic, vehicles “have to make decisions fast, so we’ve speeded up the timeframe” in which vehicles must receive sensor data, process it and respond, all without human intervention, Whitaker said. “As usual, we’ve taken it to the nth degree and said we want full autonomy. By [asking for an extreme], we get a lot of the middle ground covered.”
The placement of this year’s contest in a dynamic setting creates challenges unheard of in previous challenges and requires technological advancements that will bring self-guided vehicles to a near reality, participants say.
“This year we have moving objectives and that dynamic interaction is new and very difficult,” said Gary Schmiedel, vice president of the advanced product engineering group at Oshkosh Truck, one of the corporate entrants in this year’s Urban Challenge and one of the teams that successfully completed the 132-mile course in 2005. “This brings us much closer to a real-world application of the technology and means that we have to build a truck that’s as versatile as you or I would be.”
At the level of sophistication that will be required in this year’s contest, “this is really a software competition, not a hardware competition,” said David Stavens, a doctoral candidate at Stanford University who’s working on Stanford’s entry in the Urban Challenge and was a co-creator of Stanley, the modified Volkswagen Touareg sport utility vehicle that won DARPA’s 2005 Grand Challenge for Stanford University.
The Stanford team, consequently, is spending much of its time this year working on probabilistic algorithms and machine learning capabilities and is tackling the problem with help from the Stanford Artificial Intelligence Laboratory, Stavens said. Probabilistic algorthms will help this year’s Stanford entry, Junior, a Volkswagen Passat station wagon, deal with uncertainties along the course, while machine learning will enable the team to program the car with human-like driving skills.
“By driving other roads, you can gain enough knowledge that the robot will be able to handle the Urban Challenge course just fine,” Stavens said. “This is a very rich subset of the skills that you and I would use when we jump in our own cars and go driving, but this type of technology can save our soldiers’ lives in the battlefield and save lives in the civilian world.”
After this year’s challenge, DARPA will evaluate whether the contests have advanced the technology enough to make commercial production of autonomous vehicles for the military feasible and economically practical, Whitaker said. After an experiment along the lines of the challenges, “there’s an intermediate phase before [the military] goes out and starts buying systems. It could also be that we’ll need to see more work on the commercial side,” he said.
Teams build on technologies from past challenges
As the agency that created the Internet and nurtured it through its early years, the Defense Advanced Research Projects Agency has a long history of transferring its technical innovations from military to civilian use. The Grand Challenge will likely prove to be another example.
Although the challenge’s primary goal is developing driverless military vehicles, DARPA has organized the competitions with the expectation that technologies created for them will be applied in the private sector, too.
Many of the corporate Grand Challenge participants, in fact, look at it as an opportunity to test and perfect — in demanding military conditions — technologies they will later adapt for industrial or civilian use.
Velodyne Acoustics, a maker of high-fidelity stereo and home theater equipment, entered the 2005 Grand Challenge and invented laser-based sensors for its vehicle that it has now sold to participants in the 2007 Urban Challenge.
The company also is marketing its invention to prospects in several industries, said Michael Dunbar, Velodyne’s business development manager.
David Hall, the company’s founder, chief executive officer and chief engineer, along with his brother, Bruce, Velodyne’s president, entered a vehicle in the 2005 Grand Challenge as Team DAD (for Digital Audio Drive). While working on the project, they identified shortcomings with the laser-based light, distancing and ranging (Lidar) scanners used alone or in combination with cameras as the eyes in the guidance systems of autonomous vehicles, Dunbar said. Lidar systems available on the market at the time scanned for objects only along a single, fixed line of sight.
In response to those limitations, David Hall, an avid inventor, created his own Lidar scanner consisting of an assembly of 64 lasers spinning at 300 to 900 rotations per second capable of detecting objects anywhere in a 360-degree horizontal field. The Velodyne Lidar assembly produces 1 million data points per second, compared to the 5,000 data points a second of earlier systems.
Velodyne doesn’t have a vehicle in this year’s Urban Challenge but has sold its HDL-64 Lidar scanner to 10 Challenge participants that have included it on their vehicles, either alone or in conjunction with optical sensors, Dunbar said. “Some of the teams can use our sensor and eliminate other types of sensors so [the sensor data] is much easier for them to manipulate,” he said.
By setting its own benchmarks for supercomputing systems, DOD gets better performance — and might change how HPC systems are procured
Twice a year, work being done by the world’s fastest supercomputers comes to a screeching halt so the systems can run a benchmark called Linpack to determine how fast they are, at least in relation to one another. Linpack — which measures how many trillions of floating-point operations per second the machine is capable of executing — is the benchmark used to rank the fastest supercomputers in the world, in the twice-annual Top 500 List.
As an exercise in flexing muscle, Linpack is about as useful as any other benchmark. But as a tool for judging supercomputing systems in a procurement process, it is limited at best. The Defense Department, through its High Performance Computing Modernization Program, is shaking up the supercomputing world by applying a more disciplined approach to purchasing big iron.
Instead of using a generic benchmark to compare models, the program issues a set of metrics that carefully codifies its own workload. Program leaders then ask vendors to respond with the best — yet most cost-effective — systems they can provide to execute such a workload.
“We don’t specify how big the machine is,” said Cray Henry, head of the program. “We will run a sample problem of a fixed size, and call the result our target time. We then put a bid on the street and say we want you to build a machine that will run this twice as fast.” It is up to the vendor to figure out how that machine should achieve those results.
Sounds simple, but in the field of supercomputers, this common-sense approach is rather radical.
“It’s a well-oiled process,” agreed Alison Ryan, vice president of business development at SGI. She said that for vendors, “this kind of procurement is actually difficult. It takes a lot of nontrivial work. It’s easier to do a procurement based on Linpack.” But in the end, the work is worthwhile for both DOD and the vendor, because “it’s actually getting the right equipment for your users.”
“They’ve done a great job on the program in institutionalizing the [request for proposal] process,” said Peter Ungaro, chief executive officer at supercomputer company Cray.
DOD created HPCMP in 1994 as a way to pool resources for supercomputing power. Instead of having each of the services buy supercomputers for its own big jobs, the services could collectively buy an array of machines that could handle a wider variety of tasks, including large tasks.
On the rise
Today, the program has an annual budget of about $250 million, including $50 million for procuring two new supercomputers. Eight HPCMP shared-resource centers, which house the systems, tackle about 600 projects submitted by 4,600 users from the military services, academia and industry.
As of December 2006, the program had control of machines that could do a total of 315.5 teraflops, and that number grows by a quarter each year, as the oldest machines are replaced or augmented by newer technologies.
And over the years, the program has developed a painstakingly thorough process of specifying what kind of systems it needs.
What about HPCMP is so different? It defines its users’ workload, rather than use a set of generic performance goals.
Henry said that most of the workloads on the program’s systems can fall into one of about 10 categories, such as computational fluid dynamics, structural mechanics, chemistry and materials science, climate modeling and simulation, and electromagnetics. Each job has a unique performance characteristic and can be best run on a unique combination of processors, memory, interconnects and software. “This is better because it gauges true workload,” Ryan said.
To quantify these types of jobs, HPCMP came up with a computer program called the linear optimizer, which calculates the overall system performance for handling each of these jobs. It weights each job by how often it is executed. It also factors in the price of each system and existing systems that can already execute those tasks.
Once numbers have been generated for each proposed system, the program takes usability into consideration. Henry admitted that is hard to quantify, but it includes factors such as what sorts of third-party software is available for the platform and what sorts of compilers, debuggers and other development tools are available.
Once these performance and usability numbers are calculated, they are weighted against the past performance of the vendors. From there, the answer of which system may be the right one may be obvious — or it may come down to a narrow choice between a handful of systems.
“It’s not often they need the same type of system year after year,” Ungaro said.
Although DOD generally is well- represented on the twice-annual list of the world’s fastest computers — it had 11 in the June 2007 Top 100 ranking, for instance — the true beneficiaries are the researchers who can use the machines. The biggest benefit? “Time to solution,” Henry said.
DOD might need to know the performance characteristics of an airplane fuselage. Using a very accurate simulation saves money and time from testing actual fuselages.
“Typically, the kind of equations we’re trying to solve require from dozens to thousands of differential calculations,” Henry said. And each equation “can require a tremendous number of iterations.”
Imagine executing a single problem a million or even tens of millions of times at once, with each execution involving thousands of calculations. That’s the size of the job these systems usually handle.
DOD has many problems to test against. Programs track toxic releases of gas spread across an environment. They help develop better algorithms for tracking targets on the ground from moving radars. They speed development of missiles. In one example, supercomputing shortened the development time of the Hellfire missile to just 13 months, allowing it to be deployed in Iraq two years earlier than otherwise would have been possible.
By providing the fastest computing power available, the program in its modest way can assure the Defense Department stays ahead of the enemy.