ORNL's powerful tools for scientific discovery
Click here for more photos. |
Speed in supercomputers is an elusive commodity. Only seven years ago, in January 1995, ORNL installed what was then the world's fastest computer. The Intel XP/S 150 was capable of an astounding 150 billion arithmetic operations per second (150 gigaflops) and at $16 million, seemed like a bargain. Today ORNL has three large supercomputers with a total capacity of 5.5 trillion calculations per second (5.5 teraflops). The three massively parallel supercomputers at DOE's Center for Computational Sciences (CCS) at ORNL are dubbed Eagle, Falcon, and Cheetah (the fastest mammal). A single node of ORNL's new IBM Power4 computer, Cheetah, is faster than the Intel machine of 1995.
Eagle is an IBM RS/6000 SP supercomputer and Falcon is a Compaq AlphaServer SC computer. Both were installed and up and running in April 2000. Eagle performs 1 trillion arithmetic operations per second, making it a one-teraflop machine. Falcon has a peak performance of 0.5 tera-flop. Both supercomputers have four central processor units (CPUs) per node, which includes memory, disk drives, and input-output capabilities. Cheetah is an IBM Power4 supercomputer; it arrived at ORNL in late 2001. It provides an amazing 4 teraflops of computing power. Cheetah is the first system to use IBM's new Power4 "Regatta-H" nodes. Each of Cheetah's 24 nodes has 32 processors. The total system has more than a terabyte of memory and 40 terabytes of disk space. A comparison of the capabilities of ORNL's three supercomputers is shown below:
Eagle, which was the tenth-fastest supercomputer in the world in April 2000, is used for running codes that address complex problems in astrophysics, biology, climate prediction, computational chemistry, fusion energy, and materials science. Falcon is used to evaluate code and machine performance (see Evaluating Supercomputer Performance), as well as for large scientific runs in the research areas of astrophysics, computational chemistry, and materials science.
As Ernie Moniz, then Department of Energy undersecretary, said at the June 20, 2000, dedication of ORNL's new supercomputers, these machines are "extraordinary tools for extraordinary science." He noted further that "simulation using teraflop computers will be a tool of scientific discovery. Simulation will play an important role in the bridging from the molecular level to engineering systems to get the needed efficiencies..." to solve energy, environmental, materials, and medical problems. ORNL is providing supercomputing resources to researchers supported by DOE's Scientific Discovery through Advanced Computing Program.
Already calculations performed using ORNL computers and codes have led to the location of disease-causing genes (including the gene that causes the disease suffered by the title character in the movie Lorenzo's Oil). Computer modeling at ORNL helped IBM better understand the use of giant magnetoresistance (GMR) for reading data while the company was developing quarter-size disk drives for digital cameras. GMR simulation at ORNL has also influenced Seagate's new designs of disk drives for desktop computers. Our codes and nodes may help predict which lightweight materials considered candidates for future, highly-efficient cars will likely hold up as well in a crash as today's heavier steel cars; this approach saves money and reduces waste since crunching numbers is much cheaper than crunching cars in real crash tests.
Because ORNL's supercomputers will generate a mind-boggling number of results from a trillion calculations per second, systems must be in place to obtain, store, catalog, retrieve, and transmit over long distances these huge quantities of data. ORNL and its collaborators have developed a state-of-the-art data storage- and-retrieval system called the High Performance Storage System (see Retaining and Retrieving Data More Effectively). In addition, ORNL researchers are devising ways to move large chunks of data (up to 2 terabytes) more quickly and efficiently over high-speed network links (see Networking: Making Faster Connections Among Supercomputers) between ORNL and the National Energy Research Scientific Computing Center (NERSC) at DOE's Lawrence Berkeley National Laboratory in California. A high-speed fiber-optic link 10,000 times faster than today's fastest networks is also being set up to connect ORNL, Atlanta, and the Research Triangle in North Carolina. This network may connect with the first IBM Blue Gene supercomputer now being developed with help from ORNL researchers. When operational in 2005, this 100-teraflop supercomputer will help biologists understand the complex rules by which proteins assume their shapes (which are related to disease), as well as predict future climate as the earth’s atmosphere is loaded with increasing amounts of carbon dioxide, and aid in the design of nanoscale electronic devices.
Climate modelers use ORNL and NERSC supercomputers to simulate the earth’s climate for past, present, and future greenhouse-gas scenarios. Climate models solve complex mathematical equations that describe atmospheric and oceanic circulation, temperature, pressure, and many other variables over the entire earth. These calculations must simulate time in 20-minute increments, and many simulations are for hundreds of years. This large number of calculation results requires supercomputers, as well as high- capacity data storage and networking capabilities.
Because of their expertise and equipment, ORNL and other researchers are complementing theory and experiment as they use CCS's powerful tools for scientific discovery.