APPLICATIONS

The five partners in the National Science Foundation’s TeraGrid project will use SC2002, this year’s edition of the annual high-performance networking and computing conference, to demonstrate a variety of applications on a prototype TeraGrid infrastructure. The demonstrations will showcase applications that are poised to take advantage of the computational capability of the TeraGrid, which will be deployed in 2003 as the world's largest, fastest, distributed infrastructure for open scientific research. With the theme "From Terabytes to Insights," SC2002 will be held November 16-22 at the Baltimore Convention Center, featuring the most advanced network in the history of the conference. The TeraGrid demonstrations will involve the Center for Advanced Computing Research (CACR) at the California Institute of Technology (booth R3417), the National Partnership for Advanced Computational Infrastructure (NPACI, booth R1134), the National Computational Science Alliance (Alliance, booth R1249), Argonne National Laboratory (booth R1671), and the Pittsburgh Supercomputing Center (PSC, booth R1850). The partners will showcase applications running on a 32-node IBM Linux cluster with Intel® Itanium® 2 processors, more than 40 terabytes of local disk storage provided by StorageTek, and high-end visualization systems spread among the booths. The booths will be connected by 10-gigabit-per-second network links, and each partner exhibit will have separate high-speed connections to SCinet, the SC2002 network provider. When completed, the TeraGrid will include 20 teraflops of computing power, facilities capable of managing and storing nearly one petabyte of data, high-resolution visualization environments, and toolkits for grid computing. The TeraGrid infrastructure will be distributed among five sites: the National Center for Supercomputing Applications (the lead site for the Alliance), the San Diego Supercomputer Center (lead site for NPACI), CACR, Argonne, and PSC. TeraGrid components will be tightly integrated and connected through a network that will operate at 40 gigabits per second—the world’s fastest research network. The TeraGrid demonstration schedule at SC2002 is as follows: MONDAY, NOVEMBER 18 7 p.m., NPACI booth R1134 A TeraGrid-based, volume-rendering demonstration will show the NPACI Scalable Visualization Toolkits being used with grid-based technology to create scientific animations of 3-D data. During the demonstration, an animation of a flight path will be created, a visualization request will be submitted to the TeraGrid, and the resulting animation will be displayed. Multiple visualizations will use data sets in astronomical, oceanographic, and medical applications. This demonstration will be repeated at noon, November 19, in the NPACI booth. (http://www.npaci.edu/sc2002/cgi-bin/sc2002_schedule.cgi) 8 p.m., Alliance booth R1249 NAMD is a parallel, object-oriented molecular dynamics code designed for high-performance simulation of large biomolecular systems (http://www.ks.uiuc.edu/Research/namd/). A finalist for the 2000 and 2002 Gordon Bell Awards, NAMD uses data-driven execution and measurement-based load balancing to scale scientifically relevant simulations to over 1,000 processors and 50 time steps per second. VMD is a molecular visualization program for displaying, animating, and analyzing large biomolecular systems using OpenGL 3-D graphics and built-in Tcl and Python scripting (http://www.ks.uiuc.edu/Reseach/vmd/). During the demonstration, VMD will run on the Alliance tiled display wall and will display the state of the live NAMD simulation running on the Itanium 2-based Linux cluster. NAMD and VMD are supported by NIH and developed by the Theoretical Biophysics group the University of Illinois' Beckman Institute. TUESDAY, NOVEMBER 19 Noon, Alliance booth R1249 Researchers will report on a series of parallel numerical simulations of small bubbles seeded in a turbulent channel flow at average volume fractions of up to 14 percent. These results show that even for relatively large bubbles, an initial transient drag reduction can occur as bubbles disperse into the flow. The simulations involve up to 21,600 bubbles per run; previous simulations involved only up to 500 bubbles in laminar flow. A new parallel scheme for computing the bubbles' dynamics in conjunction with new algorithmic developments allowed for such dramatic advances. Flow visualizations will be shown on the tiled display wall. 1 p.m., Center for Advanced Computing Research booth R3417 Caltech researchers will demonstrate on-demand processing of astronomical images using TeraGrid hardware. The Virtual Sky website (www.virtualsky.org) will be used as a portal to select an area at optical wavelengths to be mosaicked, and the Itanium-based cluster at the CACR booth will be used to re-project and construct the desired image from the 10-terabyte 2MASS infrared data set. The resulting infrared image can then be federated directly with the optical image. (This demonstration will be repeated at 1 p.m., November 20, in the NPACI booth.) WEDNESDAY, NOVEMBER 20 Noon, Alliance booth R1249 The use of ensembles—or simulation suites based on different initial conditions, models or model parameterizations, and boundary conditions—has become popular in forecasting. Ensemble simulations are composited to provide probability forecasts. This demonstration will highlight how the MEAD expedition will use the TeraGrid to execute simulation suites and associated data-mining/machine-learning and visualization to study hurricanes and other severe storms. Using grid-based portals, researchers will be able to launch of hundreds of individual WRF (Weather Research and Forecasting), Regional Ocean Modeling System (ROMS), or coupled WRF/ROMS simulations on the grid in either ensemble or parameter mode. The resulting large volumes of data will then be made available through the MEAD portal for further study and for education. Simulations will be displayed on the tiled display wall. 1 p.m., Pittsburgh Supercomputing Center booth R1850 The TeraGrid environment will facilitate high throughput access to mass storage at the partner sites. Output from computation on a system at the Pittsburgh Supercomputing Center (PSC) can be transparently archived at the San Diego Supercomputer Center (SDSC) as well as at PSC. This demonstration will illustrate existing technologies that allow Storage Area Networks to extend and combine over Wide Area Network links, such as the TeraGrid network backbone. A server at the PSC booth will use a hardware solution (Akara Systems) to access Fibre Channel tape drives at SDSC at line speeds. At the same time, the PSC server will also use a software solution (SAM-FS Remote) combined with PSC-developed software to write to tape drives at PSC. 2 p.m., Argonne National Laboratory booth R1671 Argonne scientists will demonstrate an MPI application, the MM5 regional climate model code modified at ANL for long climate simulations, which will be run across multiple booths at SC2002 using the Globus Toolkit and MPICH-G2. The demonstration will use MM5 to perform a high-resolution regional climate simulation over the United States. At last year’s conference, MPICH-G2 used TCP for all its messaging, and that demonstration did show as proof of concept the ability to use Globus and MPICH-G2 to run this application on TeraGrid machines. This year, MPICH-G2 will use Myrinet for improved intra-cluster messaging.