Supercomputing coming to a closet near you?
by Ashlee Vance, IDG
Los Alamos, NM -- The license plates in New Mexico salute the state as a "Land of Enchantment," and on a warm Friday morning a group of scientists at the Los Alamos National Laboratory made sure the state lived up to its billing.
At a complex surrounded by scrub-covered plateaus, three Los Alamos researchers last week unveiled a potential breakthrough in high-performance computing: a 240-processor, Linux-based cluster dubbed Green Destiny that literally made the mouths of other scientists drop open when they saw the system for the first time.
As with any attempt at something new, however, some of the mouths opened to express excitement, while others opened to express dismay with the project's direction.
Dot coms have died, technology sales are slow, and layoffs continue, but the work being done at Los Alamos for the most part triggered a refreshing rush of enthusiasm for the technophiles present. For a moment, the introduction of a supercomputer that can fit inside a closet countered the gloom hovering over the technology industry.
Even luminaries such as Gordon Bell, often called the father of high- performance computing, and Linus Torvalds, creator of the Linux operating system, celebrated the arrival of this new, unexpected technology.
Supercomputing in Small Spaces Green Destiny was originally conceived not as a supercomputer but as a powerful Web server for the Research and Development in Advanced Network Technology (RADIANT) group at Los Alamos, according to Wu- chun Feng, RADIANT's team leader.
Feng thought RLX Technologies' blade servers--no-nonsense servers stripped down to their core components--would be best suited for handling research documents and contact information. Then Feng, along with fellow RADIANT researchers Michael Warren and Eric Weigle, decided that RLX's innovative server design, coupled with the low power consumption of Transmeta's processors, might be the right combination for what they now call Supercomputing in Small Spaces.
The researchers got RLX to send over its System 324 server, which packs 24 server blades into a chassis only 5.25 inches high. The group then started running Warren's N-body Simulation program on the servers, which is used to learn more about supernovas. Not only did the software run well, but the RLX servers also required none of the maintenance needed by other, larger computers at Los Alamos. Nine months later the System 324 is still running without an interruption.
However, 24 servers were not enough for Feng, so he contacted Chris Hipp, blade server pioneer and RLX's founder. He wanted to see if Los Alamos could get its hands on an entire rack of 240 blades, to test his theory that high- performance computing could be done in a smaller space and for less money than previously imagined.
It was this tower of blade servers--Green Destiny--that sparked the cheers, and a few jeers, from the scientists here.
For many of the Los Alamos scientists, the unveiling of Green Destiny was their first introduction to blade servers, let alone blade servers being used to build a supercomputer.
The slew of expletives and exclamations that followed Feng's description of the system made it clear that the blades had captured the audience's attention. Some murmured "Wow," while others let out multiple shouts of "Jesus!" as their jaws dropped.
A Few Jeers Several scientists here did not share the enthusiasm for Green Destiny, however. Los Alamos, after all, is home to several massive supercomputers that take up entire floors of buildings and require several cooling systems shaped like mini-nuclear reactors to keep them running. These "real" supercomputers handle serious work, and some of the people running them consider Green Destiny a joke. One scientist walked out of Feng's presentation, making his feelings clear.
The controversy stems in part from Feng's decision to use 240 blade servers running on chips designed primarily for notebook computers. With Green Destiny, Feng introduced the notion that "simply doing bigger, faster machines is not good enough anymore."
Feng is the first to admit that it will take a lot of work and a bounty of creativity to lift a blade-based system on a par with current supercomputers, but he also holds to the belief that systems like Green Destiny may be the answer to problems facing supercomputing in the next ten years.
Some of the magic in Green Destiny stems from Feng's decision to go with Transmeta chips, which rely more on software than on high transistor counts to process data. The Transmeta processors consume less energy than similar chips from Intel and Advanced Micro Devices (AMD), Feng said. Intel and AMD derive performance gains largely by following Moore's Law, which says that transistor count doubles on chips roughly every 18 months.
"Currently, our biggest concern is the continued pursuit of Moore's Law and its effect on system reliability," Feng wrote in a research paper. "The continued tracking of Moore's Law will result in the microprocessor of 2010 having over one billion transistors and dissipating over one kilowatt of thermal energy; this is considerably more energy per square centimeter than even a nuclear reactor."
Supercomputing Sans Intel Using Transmeta's chips, Feng's team was able to create a high-performance computer that sits in the hallway of a dusty warehouse where the temperature often exceeds 85 degrees Fahrenheit. Compare that to the "Q" computer, which was also unveiled Friday at Los Alamos, and which calls home a 4043-square-meter computer room supported by special cooling equipment.
Feng does not claim Green Destiny can come close to out-computing Q, but he will say his model of supercomputing in small spaces might be a more practical approach for the future.
As companies like Excite@Home look to sell off hardware to bring in extra money--and as layoffs continue to hit the IT industry to the point that it often seems the end really isn't in sight--Green Destiny suggests that not all is doom and gloom. It hums away in its warehouse corner, causing some to marvel at a new approach to supercomputing and others to scoff at its mere existence.
Last updated: Oct 29, 2010