Leader in number crunching
With new machines, ORNL has one of the world’s largest unclassified scientific computing facilities
In April, ORNL became one of the most powerful unclassified scientific computing facilities in the nation. The IBM RS/6000 SP supercomputer, dubbed the “Eagle,” has been upgraded, and a new Compaq Alphaserver supercomputer, called the “Falcon,” has been installed. The two supercomputers can operate at a theoretical speed of 1.5 teraflops, or more than a trillion calculations per second. That’s 10 times the computational speed of ORNL’s recently dismantled Intel Paragon, which was the fastest supercomputer in the world in early 1995.The IBM SP, which was installed last year, originally operated at 100 gigaflops and then 400 gigaflops, less than half a teraflop. It has been upgraded so that it can operate at more than 1 teraflop. The new Compaq Alphaserver supercomputer can operate at almost half a teraflop. The computational speed and capacity resulting from these changes make ORNL one of the most powerful unclassified computing facilities worldwide.
“ORNL can be a leader and galvanizing force for taking the Southeast to the next level in high-performance computing,” says Thomas Zacharia, director of ORNL’s Computer Science and Mathematics Division, who would like to see ORNL reach 10 teraflops of computing capacity in the next few years. “Computing is one of three top priorities of our new contractor, UT-Battelle. We expect to fulfill this commitment. We have an opportunity to reach out to the core universities of UT-Battelle. The proposed Joint Institute for Computer Science will be a mechanism or vehicle for building these partnerships.”
 |
| The Computer Science and Mathematics Division's Pat Worley checks out the freshly uncrated Compaq Alphaserver. |
While the IBM supercomputer is dedicated to a range of computational science research efforts, the Compaq machine will be focused on computer science projects such as early systems evaluation aimed at developing better tools for computational science researchers.
Computer science will play an increasing role in science, Zacharia believes. Development of step-by-step procedures called algorithms will allow more powerful computers to solve more complex scientific problems through simulations of experiments. “High-performance computing is needed to meet DOE objectives,” he says. “It has become a crucial tool for scientific discovery in climate, bioinformatics, and materials research as well as many other areas.”
One example of a new supercomputer algorithm developed at ORNL is “large-scale time-averaged normal coordinate analysis.” ORNL researchers have used this new algorithm on the new ORNL supercomputers to calculate the vibrational modes of a polymer system of 24,000 atoms, shattering the previous record of 3,000 atoms. They expect to soon extend the number of atoms modeled to over 100,000, which will enable the study of viruses for improved drug design, better understanding of photosynthesis and the design of new nanoscale devices.
One scientific challenge is the prediction of changes in the future global climate as greenhouse gas levels rise. Computing at ORNL will be used to predict changes in the regional climate in the Southeast based on results of global scenarios. For example, scientists will try to predict whether, in the next few decades, East Tennessee will have more droughts, North Carolina will have more hurricanes and Florida will have greater coastal flooding than in the past few decades.
Researchers in the computational biosciences are using ORNL supercomputers for bioinformatics. They are locating and discovering genes in DNA sequences, predicting the structure of proteins encoded by specific genes, and estimating gene functions. The DNA sequences they will be analyzing computationally include human chromosomes 19, 16, and 5, for which draft sequences have been produced by DOE’s Joint Genome Institute, of which ORNL is a part. The new information on genes and gene functions could lead to the development of new drugs.
ORNL researchers are using supercomputers to simulate collisions between future cars made of advanced lightweight materials that are designed to use fuel more efficiently and cleanly. The idea behind these calculations is to determine whether these cars will hold up during crashes as well as the steel cars of today. These and trillions of other numbers will be crunched at one of the nation’s most powerful unclassified computing facilities.—Carolyn Krause
