China now dominates supercomputing. That matters for U.S. national security.
China has, for the seventh time in a row, taken the top prize as the global leader in supercomputing. The Sunway TaihuLight, at the National Supercomputing Center in Wuxi, a city near Shanghai, has clocked a theoretical peak performance of 125.4 petaflops, meaning 1,254 trillion calculations per second. As announced on June 20 on Top500.org, a website considered the industry benchmark for supercomputer rankings, the machine is nearly three times faster than the previous first-place holder, the Chinese Tianhe-2.
U.S. media outlets met the latest rankings news with a flurry of alarmed reporting, evoking the tone, if not the terminology, of the Sputnik era. “China’s New Supercomputer Puts the US Even Further Behind,” read one Wired headline on June 21. “China Beats the U.S. In Supercomputing Prowess,” fretted Fortune. “There is no U.S.-made system that comes close to the performance of China’s new system,” concluded U.S.-based Computer World on June 20. What the reports don’t explain is precisely how supercomputing serves national security interests, which in turn explains why supercomputing advances by a strategic competitor like China arouse such alarm.
The classic trinity of supercomputing applications entails cryptography, the design of nuclear weapons, and weather forecasting, Dag Spicer, senior curator at the Computer History Museum in Mountain View, California, told Foreign Policy. Each of these three operations is considered vital to national security, and each of them requires high-performance computers to simulate complex relationships between exceedingly large numbers of particles, entities, or variables.
Of the classic trinity, cryptography weighs particularly heavily. “Security is only as good as your fastest computer,” noted Hansen Hsu, curator of the Center for Software History at the Computer History Museum. “Security is always an arms race.” Encryption, at its most fundamental level, is premised on creating cryptographic problems so computationally intensive that, while breakable in theory, would require so much time to crack — a year or a century — as to be essentially unbreakable in practice. “But if you could build a computer so powerful that it would change those 100 years into one year,” Hsu said, “or one year into a couple of weeks, then the equation is different.”
Supercomputers have also become central to an ever-increasing range of sciences and industries. “Our science is driven today by how we simulate things,” Jack Dongarra, a world-recognized authority on supercomputing at the University of Tennessee and Oak Ridge National Laboratory, told FP. Automobile and airframe manufacturers are major users of supercomputers, with companies running niche, computationally intense applications that, among other things, simulate car crashes virtually. Another emergent field within supercomputing is protein folding, a critically important but computationally draining simulation process. Among its many other applications, it allows for understanding the myriad configurations and processes by which proteins fold and also mis-fold, helping scientists better to understand diseases like Alzheimer’s and Parkinson’s.
But excelling in such fields is rarely a simple question of which country holds the top slot on Top500.org. As Dongarra put it, “being number one or two doesn’t matter too much.”
Quantity does matter, however, in terms of how many high-performance computers a country possesses, and the portfolio of scientific, military, and industrial projects to which they are collectively dedicated. What’s so impressive about China’s achievements of late is not only speed, Dongarra stressed, but also the sheer number of Chinese-built systems now found somewhere in the Top500.org rankings. With more and more supercomputers being built by and based in China, scientists and industrialists from around the world become more likely to bring their supercomputing needs to the People’s Republic – and not to the United States. (Take the story of genome sequencing, where the Beijing Genomics Institute has become a prime destination for scientists and pharmaceutical companies worldwide.) The question is now whether the global center of gravity for supercomputing will drift towards China, as it already has for genomic mapping.
Significant also is Sunway’s status as the world “greenest” supercomputer.
Sunway is not only the fastest supercomputer ever to be built; it is also the most energy efficient, by far. It operates at a rate of approximately six gigaflops per watt, nearly three times more efficient than the next most efficient system in the top five of the Top500: the Titan supercomputer at Oak Ridge National Laboratory, which operates at a little over two gigaflops per watt.
This has profound implications for the next major challenge in supercomputing, because it opens the door to Exascale computing, which will achieve speeds in excess of one quintillion (one billion billion) floating point operations per second — about 10 times faster than Sunway’s new record. As outlined in a Fall 2010 report by the U.S. Department of Energy Office of Science, advances in exascale computing could drive forward projects ranging from high-precision climate change modeling to the reverse-engineering of the human brain — research fields in which, due to the hypercomplexity of the systems being investigated, even petascale computing remains insufficient.
Exascale computing is in many ways less a question of computing architecture than of energy and heat. “Could we build this machine today?” asks Dongarra. “The answer is yes, but if you did it today, it would be very expensive.” Electricity is one of the main expenses. While petascale computing in the form of Sunway requires approximately 15 megawatts to operate — roughly the amount of energy consumed by 1,400 average American homes — it has been estimated that exascale computing would require somewhere in the order of 200 megawatts of power, on par with the annual consumption of a small town.
As supercomputers become more central to a variety of fields, they will also become more central to national security. There’s little chance China’s supercomputing prowess will subside any time soon, particularly given its historic strides in energy-efficient supercomputing. Now might be a good time for the United States to stop worrying about China’s advances, and start learning from them.
By Tom Mullaney