CHAMPAIGN, IL — Though still in the development stage, the TeraGrid is already enabling scientific discovery. When complete in 2004, the TeraGrid, http://www.teragrid.org/, will be the world's largest, fastest distributed computing infrastructure for open scientific research. It will offer at least 20 teraflops of computing power as well as data-storage facilities, high-resolution visualization capabilities, and toolkits for grid computing, all integrated and connected through a 40 gigabits-per-second network. NCSA is one of the founding sites of the TeraGrid partnership and has one 2-teraflop Linux cluster and an additional 8-teraflop cluster currently being installed. In addition to NCSA, TeraGrid partners are Argonne National Laboratory, the Center for Advanced Computing Research at the California Institute of Technology, Indiana University, Oak Ridge National Laboratory, Pittsburgh Supercomputing Center, Purdue University, the San Diego Supercomputer Center, and the Texas Advanced Computing Center at the University of Texas at Austin. While the TeraGrid is not yet available to the broad research community, Bruce Loftis, Associate Director of NCSA's Application Technologies Division, tested features of NCSA's TeraGrid systems this summer by running optimization-simulation code for a groundwater remediation study. The code was run on a local grid-enabled desktop workstation. The function evaluations were completely independent and were distributed across the TeraGrid. A single run required about 10,000 CPU hours and required two to three days to complete. TeraGrid machines at NCSA and the San Diego Supercomputer Center were used, as well as other grid-enabled resources at NCSA and the University of Kentucky. "It was very hard to do this," Loftis emphasized, but sufficient runs were completed to produce solid results, enabling graduate student Xiaolin Ren to complete her thesis. Ren studies under Barbara Minsker, an associate professor of civil and environmental engineering at the University of Illinois at Urbana-Champaign who is on a one-year sabbatical at NCSA. Minsker's research focuses on groundwater remediation—the removal of contaminants from groundwater. The optimization research conducted on NCSA's TeraGrid systems was designed to determine whether using an accurate cost function is a better way to choose a cleanup plan than using rough cost estimates. Pump-and-treat remediation is largely a matter of where to place pumps to remove contaminants, at what rate to pump, and how long the pumping should continue. The goal is to achieve an acceptable reduction of contaminants at an affordable cost. Recent studies by the Environmental Protection Agency and the Navy demonstrated that many pump-and-treat systems are not operating as designed, have unreachable or undefined goals, and have not been optimized since their installation. Using the Umatilla Chemical Depot in Oregon as a case study, the simulation completed using TeraGrid systems identified the tradeoffs between costs and benefits of various cleanup designs. For example, the results demonstrated that, beyond a certain threshold, greater spending would result in negligible or nonexistent reduction in the contaminant level. The researchers concluded that optimization could help decision-makers rein in expenses on environmental cleanup projects while still effectively reducing contamination. Minsker said the TeraGrid's capabilities are a boon to her research, which uses computationally intensive genetic algorithms. "It enables us to look at real-world problems that no one has been able to solve before," she said. "With the TeraGrid, we can solve a much bigger problem." A paper based on the research conducted using the TeraGrid—titled "Which Groundwater Remediation Objective is Better, a Realistic One or a Simple One?"—has been submitted to the Journal of Water Resources Planning and Management, http://ojps.aip.org/wro/. "This demonstrates the critical importance of the TeraGrid in enabling scientific discovery," said NCSA director Dan Reed. "We are entering a new era in which computation and simulation tackle scientific questions that previously seemed intractable.