NCSA NEWS

Prototype TeraGrid One of the Stars of SC2001

released December 11, 2001

Contact
Karen Green
NCSA Public Information Officer
kareng@ncsa.uiuc.edu
217.265.0748 phone
217.244.7396 fax

CHAMPAIGN, IL — The TeraGrid has yet to be built, but visitors to the SC2001 exhibit floor had the chance to glimpse its potentials through demonstrations presented by the four TeraGrid partners during the conference exhibits Nov. 13-15 at the Denver Convention Center.

Funded by the National Science Foundation, TeraGrid is a multi-year effort to build and deploy the world's largest, fastest, most comprehensive, distributed infrastructure for open scientific research. At SC2001 TeraGrid demonstrations were presented at four booths on the exhibit floor—the National Computational Science Alliance (Alliance), the National Partnership for Advanced Computational Infrastructure (NPACI), the Center for Advanced Computing Research (CACR) at the California Institute of Technology, and Argonne National Laboratory. A prototype, 20-gigabit/second network connected the Alliance and NPACI booths and all four booths featured Linux clusters running on Intel Itanium processors.

The Alliance is led by the National Center for Supercomputing Applications (NCSA) at the University of Illinois at Urbana-Champaign. NPACI is led by the San Diego Supercomputer Center (SDSC) at the University of California, San Diego. NCSA, SDSC, Caltech, and Argonne are the four TeraGrid partners.

"We were able to show the participants in the SC conference that the TeraGrid is more than an exciting concept," said Rob Pennington, head of NCSA's computing and communications division and coordinator of NCSA's TeraGrid efforts. "Linux clusters running on Itanium processors are here to stay and are able to handle the most sophisticated scientific codes. The fast networking makes it conceivable to use distributed machines as one system, and the grid computing tools—such as the Globus Toolkit—simplify jobs and make it possible to share applications, data, and scientific instruments."

The Globus Toolkit™, developed jointly under the leadership of Argonne and the University of Southern California Information Sciences Institute, will provide the core of grid services necessary to integrate the components of the TeraGrid into a single resource. At SC2001 a dozen major companies, including Compaq, IBM, Sun Microsystems, and Microsoft, announced their adoption of Globus as the basis for their grid efforts.

"In addition to large-scale clusters, the TeraGrid project will deploy a distributed data center," said Mike Vildibill, head of SDSC's high-end computing and communications division and SDSC site lead for the TeraGrid project. "Our ability to successfully run real applications across a distributed testbed is an important milestone." SDSC's data-oriented installation at SC2001 included an Itanium Linux cluster attached to a 7 terabyte Sun Microsystems storage area network.

A number of applications ran November 14 on mini TeraGrid clusters across all four booths in a series of demonstrations coordinated in the Argonne booth. Many of these applications used MPICH-G2 (http://www.globus.org/mpi/), a grid-enabled implementation of the MPI standard developed at Northern Illinois University and Argonne. Equipment provided by Nortel Networks and Cisco Systems provided 20 gigabit/second network connections among the clusters and 1 gigabit/second connections among cluster nodes and into SCinet, the SC network.

"We had only three days to work with and we were able to get all these applications up and running on the prototype TeraGrid," said Sandra Bittner, a network and systems engineer at Argonne, who coordinated many of the collaborative demonstrations from the Argonne booth. "Some people told us this would be impossible to do, but we did it."

Applications that ran successfully on Linux clusters in the TeraGrid partners' booths included:

• Virtual shock physics Test Facility (VTF), a code developed at CACR for computing 3D dynamic response of a variety of materials impacted by strong shock waves. The code was developed with funding from the Department of Energy's Accelerated Strategic Computing Initiative and allows researchers to simulate strong shock and detonation waves colliding with fluid and solid targets. Another VTF demonstration featured simulations of shock and detonation waves hitting both fluid and solid targets and ran on 36 Itanium processors in the Caltech, Alliance, and Argonne booths. A third demonstration ran in the Caltech booth only, and a final demonstration used 48 Itanium processors in the Alliance and Caltech booths.



• SEQUEST, an application used to identify proteins by comparing the mass spectrometer output to the appropriate protein or DNA database. The demonstration ran across 44 nodes of the mini TeraGrid and examined the mass spectrometer output of a malaria parasite called Plasmodium falciparum. Results from this research will contribute to malaria vaccine studies being performed by Daniel Carucci at the Naval Medical Research Center in Silver Spring, MD, and John Yates and Laurence Florens at the Scripps Research Institute in La Jolla, CA. The parallel version of the SEQUEST code was developed by SDSC computational scientists Giri Chukkapalli, Amitava Majumdar, and Bob Sinkovits.



• NAMD, a molecular dynamics code developed by the Theoretical Biophysics Group at the University of Illinois, Urbana-Champaign. The demonstration involved atomic level simulations that examined how the protein MscL (for mechanosensitive channel of large conductance) responds to pressure changes. Understanding the behavior of MscL could provide insight into the biomolecular causes of our sense of touch.



• WhyWhere, an SDSC toolkit used for geographic prediction and modeling. The WhyWhere demonstration by SDSC researcher David Stockwell combined a massive database of environmental and satellite data, efficient image-processing algorithms, and grid-based cluster computing into a search and mapping system for predicting and explaining biodiversity data. Stockwell ran the demo from a laptop and the wireless Internet at the convention center. From SDSC, researcher Dong Ju Choi also performed benchmarking runs with WhyWhere on as many as 50 processors of the TeraGrid prototype.



• Remote High-Resolution Movie Playback, a tool developed at Argonne's Futures Laboratory to play back prerendered scientific visualizations. The movie playback tool exploited the network bandwidth among show floor sites, consuming as much as 1 gigabit/second bandwidth to play movies with a resolution of 2,629 x 1,448 pixels. The application makes use of grid-enable MPICH, and consists of two jobs—a display and a data server job—that utilize message passing interface 2.0 (MPI 2.0).



• Remote Distributed Visualization, a visualization application developed by Argonne's Futures Laboratory to aid in understanding large datasets. This application allows researchers to visualize simulations that typically produce more than a terabyte of data. The data was divided into problems that could be handled by individual cluster nodes and reassembled as a visualization on Argonne's uMural2 tiled display wall.



• Volume rendering, using the AppLeS Parameter Sweep Template (APST) to distribute computations. SDSC researchers Henri Casanova and Steve Cutchin demonstrated the use of APST on the prototype TeraGrid for a 3D volume-rendering application on a data set of colliding galaxies. During the APST demonstration, Casanova and Cutchin used as many as 26 nodes in the NPACI and Alliance booths and employed both processors of each node. (Images: http://grail.sdsc.edu/projects/apst/demos.html).



• TRACER, an advective code used to study the movement of weighless tracers through airflow associated with storms and hurricanes. The demonstration used data from Bob Wilhelmson's group. Wilhelmson is an NCSA researcher and professor of atmospheric science at the University of Illinois. Paul Woodward, director of the Laboratory of Computational Science and Engineering (LCSE) at the University of Minnesota, helped to optimize the code for Itanium systems. The demonstration used TRACER to study how air rotates, rises, and sinks within a simulation of Hurricane Opal. The demo ran on a 32-node cluster in the Alliance booth and resulting animations were displayed in real time using a volume rendering package from David Porter at LCSE.

"We had good performance numbers on a 32-node cluster—even with only two days to get things up and running and to work out any problems," said Porter. "As we move to bigger cluster systems we will be able to look at more and more complicated flow problems. This is just a start."

 

Releases Archive