INTERCONNECTS

Lonestar, the new Cray/Dell Linux cluster supercomputer at the Texas Advanced Computing Center (TACC) at The University of Texas at Austin (UT Austin), went into full production on Monday, January 5, 2004, after a two-month "friendly user" period. Within a week, users were producing interesting results. FOURTH MOST POWERFUL COMPUTER IN ACADEMIC RESEARCH — Lonestar, the new Cray/Dell Linux cluster supercomputer at the Texas Advanced Computing Center (TACC) at The University of Texas at Austin (UT Austin), went into full production on Monday, January 5, 2004, after a two-month "friendly user" period. Within a week, users were producing interesting results. Lonestar is the fourth most powerful supercomputer in the U.S. academic community. "Researchers at UT Austin, and across the nation, now have access to another tremendously powerful tool for scientific discovery," said TACC Director Dr. Jay Boisseau. Computational scientists at UT Austin and other U.S. universities are now using the resource for research in disciplines including astronomy, genomics and proteomics, engineering, fluid dynamics, geosciences, materials science, molecular biology, physics, and nanotechnology. Half of the computer cycles will be allocated directly to UT Austin researchers, while the other half will be allocated through the National Partnership for Advanced Computational Infrastructure (NPACI), of which TACC is a member. In the fall of 2004, Lonestar will become one of the resources contributed by UT Austin to the extended TeraGrid. Friendly User Experience Among the friendly users were Dr. Bhagawan R. Sahu, a physicist and postdoctoral researcher in the group of UT Austin physics professor Leonard Kleinman; John Peterson, a research associate with UT Austin aerospace engineering professor Graham F. Carey; and Victor Calo, a research associate with UT Austin aerospace engineering professor Thomas J.R. Hughes. Sahu, Peterson, and Calo work with very different kinds of codes, and all three report good experience using Lonestar. Sahu performed test calculations on lithium boron hydride, a crystalline material that is a leading candidate for the storage of hydrogen, with 18.51 weight-percent of hydrogen, highest among the known hydrides. Its light weight should make it suitable for on-board hydrogen-propelled automobile and space applications. On Lonestar, he tested a pseudopotential-based density-functional version of VASP, the Vienna Ab-initio Simulation Package, to predict the efficiency of hydrogen evolution from the crystal under various conditions. Sahu also tested the full-potential-based density-functional code WIEN2k, studying the electronic structure of molybdenum nitride, a hard material used for coatings. "I found Lonestar excellent for these density-functional theory calculations," Sahu said, "and my test codes easily ran twice as fast as they had on Tejas, TACC's IBM IA-32 cluster." Peterson worked on fluid-flow simulations with applications to viscous fluids in small-aspect-ratio containers. Such flows occur, for example, in the "burning" of a compact disk (CD), where the laser's heat creates a surface-tension-driven fluid flow that results in a tiny pit, which is the recorded bit. The code is an open-source finite-element solver for the incompressible Navier-Stokes equations, using the libMesh library (available at http://libmesh.sourceforge.net). Peterson's problem representation comprised 16,000 elements. "On Lonestar, I got results back so fast that I was able to double the number of timesteps and get a much better realization of the flow," Peterson said. Calo's research examined the transition to turbulence of a fluid running over a flat plate, induced by high levels of free-stream turbulence, with applications to turbomachinery design, aerodynamics, and electronic cooling devices. "The code we use is a variational multiscale large-eddy simulator, which is computationally intensive and very scalable to increasing numbers of processors," Calo said, "which was why my experience with Lonestar was so good. The hardware seems extremely reliable, and the fast hardware controllers with low latency were critical in maintaining the performance as I used more processors." Lonestar The current Lonestar configuration is a machine that can deliver 3.67 teraflops (trillions of floating-point operations per second). It consists of 600 Xeon processors (rated at 3.06 gigahertz) within 282 Dell dual-processor PowerEdge 1750 compute nodes and 18 PowerEdge 2650 I/O and front-end server nodes. The total memory of the system is 636 gigabytes, and a Myrinet-2000 switch fabric interconnects the processors, running point-to-point at 250 megabytes/second. Lonestar has access to 39 terabytes of storage, a portion of which will be configured as a high-performance parallel file system. Cray Inc. integrated the system using Cray/Rx, a version of the Rocks cluster management software developed by NPACI. An upgrade to Lonestar in three months will feature an additional 200 processors. "These will be the fastest Xeons available," said Chris Hempel, TACC Associate Director for Resources and Services, "and they will help address the most challenging problems in science and engineering."