Los Alamos National Laboratory has selected Linux Networx to build two large computing clusters for scientific research that use Opteron processors, underscoring the popularity of clustered supercomputers.
"Lightning," the larger of the two clusters, will contain 2,816 Opteron processors from Advanced Micro Devices and will be capable,
theoretically, of performing 11.26 trillion operations per second (teraops), making it the largest Linux cluster
created to date, according to Linux Networx.
Commissioned out of the National Nuclear Security Administration's Advanced Simulation and Computing program, Lightning will be used to calculate problems related to testing and stockpiling nuclear weapons.
"Orange," the second cluster, will contain 512 Opteron processors and be used as part of the Institutional Computing project, which conducts unclassified research in areas such as antibiotics and other commercially relevant tasks. The system will also feature high-speed Infiniband interconnections from Mellanox.
Although computers such as NEC's Earth Simulator generally provide the highest performance, supercomputers built out of hundreds of standard one- and two-processor servers linked together in
large clusters have become the norm in large computing jobs and laboratory work.
Because they are made out of commercially available components, clusters are generally far cheaper and can be built
fairly quickly. Lightning, for instance, is expected to take less than two months to build.
"Probably 80 percent of the (supercomputer) applications can be run more efficiently on clusters," said Stephen Hill, Linux
Experts and federal agencies in Washington have been debating the use of traditional versus clustered supercomputers. Recent reports out of Washington indicate that the government will take a balanced approach, Hill said.
The advent of clusters has also opened up scientific computing market to more competition. Dell, often chided by large
companies such as IBM for not conducting much independent research, is one of the largest providers of Linux clusters.
A substantial portion of Dell's independent research and development goes into how to optimize large clusters by improving the management software and testing how different interconnects and processors work together, said Reza Rooholamini, director of cluster and operating system engineering for the Dell Enterprise Systems Group.
Salt Lake City-based Linux Networx is a relative newcomer to supercomputing. The 125-person company, which used to be named
Alta Technologies and which specialized in components, entered the clustering market in 1997. It changed its name in 2000 and has landed contracts with a number of the national laboratories. Like Dell, the company tries to differentiate itself through hardware design and system management software.
"It was a big-iron market," Hill said. To continue to expand the market, the company is also opening up training facilities
where application developers can experiment with the company's systems, he said.
The deal with Los Alamos National Laboratory in New Mexico marks the latest in a string of large cluster deals for AMD, which makes the Opteron
chip. In July, IBM announced it would use Opteron servers inside a large cluster for Japan's
National Institute of Advanced Industrial Science and Technology. In June, the Dawning Information Industry of China said it would build a 10-teraop system with the Opteron.
The Opteron chip will also be used in Red Storm, a more traditional supercomputer being created by Cray.