It resembles something a hacker—okay, an ambitious hacker—may put together in his garage. Even so, Sandia's Cplant Cluster was speedy enough to take the 44th position on the 1999 list of the 500 fastest computers in the world.
The cluster is made up of 600 computers located at Sandia National Laboratories in Albuquerque, N.M. Individually, each device can solve benchmark problems at a rate of 600 million operations per second. Combined, their 580 nodes can take on those same problems at a pace of 232.6 billion operations per second.
According to project software developer Rolf Riesen, its creators merely used off-the-shelf hardware. Then, they wrote their own device driver for fast communication between nodes. After that, the project team used Sandia's own utilities to tie the nodes together.
These utilities have an impressive pedigree, too. They were based on Sandia's ASCI Red machine. You may know this behemoth as Teraflops, the fastest computer in the world, able to solve problems at rates exceeding a trillion operations per second. Many of this reigning champion's engineers contributed to the Cplant Cluster's design.
Riesen says that the cluster's scalability—the ability to add more units—was a key to its success. A previous version with 400 nodes was ranked at number 92 in a worldwide LINPAC test, which gauges the speed and accuracy of machines processing similar series of operations. This adaptability also will be important, as many companies are likely to turn their resources away from developing massively parallel supercomputers. High-power setups such as the Cplant are likely to become more common.
For more information, check out Sandia's web site at www.sandia.gov.