November 4, 2002

BERKELEY, Calif. — Lawrence Berkeley National Laboratory, a leader in advancing scientific computing for the U.S. Department of Energy, will share its expertise with the high-performance computing and networking community at the SC2002 conference to be held Nov. 16-22 in Baltimore.

Berkeley Lab scientists and users of DOE's National Energy Research Scientific Computing (NERSC) Center located at the Lab will instruct in three tutorials, give one plenary talk, present 10 technical papers, conduct two "birds-of-a-feather" sessions and moderate one panel discussion. Additionally, LBNL staff will give talks in exhibit booths for NASA, LBNL and DOE's Scientific Discovery through Advanced Computing (SciDAC) program.

Here is a brief description of LBNL contributions to the conference program.

Plenary Speaker

Julian Borrill, a computational cosmologist in the Lab's Scientific Computing Group, will give a plenary talk on "Computing the Cosmos: from Big Bang to Black Holes" starting at 9:15 a.m. Wednesday, Nov. 20. Borrill has been a member of several large-scale international collaborations researching the nature of the universe, from how it was formed to how it may eventually end. "Our ultimate goal is to be able to simulate the entire history of the universe, from the creation of spacetime in the Big Bang to its local negation in a black hole singularity," he said.

Tutorials

On Monday, Nov. 18, Vern Paxson, Stephen Lau, James Rothfuss and Bill Kramer, all of Berkeley Lab, will present a full-day tutorial on "Trends in Computer Security for Open Scientific Facilities." The session will address the issue of balancing the collaborative needs of an open scientific research facility while simultaneously protecting a site from attackers, and examine the challenges that lie ahead in high performance computing security.

David Bailey, the chief technologist for NERSC, will help lead a one-day tutorial about "Tools and Methods for Performance Modeling and Prediction" on Sunday, Nov. 17. This tutorial presents a practical framework for understanding the performance of high-performance computing applications that is being developed by the Performance Evaluation Research Center (PERC), a DOE SciDAC Center with several lab and university members and affiliates.

John Shalf of the Lab's Computational Research Division will help lead a full-day tutorial on "Developing HPC Scientific and Engineering Applications: From the Laptop to the Grid" on Monday, Nov. 18. The session will cover strategies and techniques for development of portable, high-performance, parallel scientific and engineering applications.

Panel Discussion

Horst Simon, director of the NERSC Center Division, will moderate a panel discussion of "The 40 Tflop/s Earth Simulator System: Its Impact on the Future Development of Supercomputing" from 10:30 a.m. to noon Friday, Nov. 22. The international panel will discuss the future of supercomputer design, performance, applications and computational science.

Technical Papers

Tuesday, November 19

Ekow J. Otoo, Frank Olken and Arie Shoshani of the Scientific Data Management Group will present a paper on "Disk Cache Replacement Algorithm for Storage Resource Managers in Data Grids." The paper addresses the problem of cache replacement policies for Storage Resource Managers (SRMs) that are used in Data Grids.

Yun He and Chris Ding, both of the Scientific Computing Group, co-authored a paper entitled, "MPI and OpenMP Paradigms on Cluster of SMP Architectures: The Vacancy Tracking Algorithm for Multi-Dimensional Array Transposition." The paper investigates remapping multi-dimensional arrays on cluster of SMP architectures under OpenMP, MPI, and hybrid paradigms.

Chief Technologist David Bailey and Xiaoye S. Li of the Lab's Scientific Computing Group are co-authors of a paper on "High Performance Computing Meets Experimental Mathematics." The session will describe some novel applications of high-performance computing in a discipline now known as "experimental mathematics."

LBNL's Ji Qiang and Robert Ryne are among the co-authors of a paper on "Advanced Visualization Technology for Terascale Particle Accelerator Simulations." This paper presents two new hardware-assisted rendering techniques developed for interactive visualization of the terascale data generated from numerical modeling of next-generation accelerator designs.

Brian Tierney of LBNL's Distributed Systems Department is co-author of a paper on "A TCP Tuning Daemon." The paper describes a technique for making greater use of available bandwidth to improve the performance of high-performance distributed applications.

Wednesday, November 20

Laura Grigori and Xiaoye S. Li present their paper on "A New Scheduling Algorithm for Parallel Sparse LU Factorization with Static Pivoting." They present a static scheduling algorithm for parallel sparse LU factorization with static pivoting. Experimental results and comparisons with SuperLU_DIST are reported after applying this algorithm on real world application matrices on an IBM SP RS/6000 distributed memory machine.

James W. Demmel and Katherine A. Yelick, who have joint appointments at Berkeley Lab and the University of California, Berkeley, are among a group of co-authors of a paper entitled "Performance Optimizations and Bounds for Sparse Matrix-Vector Multiply." The paper considers performance tuning, by code and data structure reorganization, of sparse matrix-vector multiply (SpMxV), one of the most important computational kernels in scientific applications.

Esmond Ng of the Lab's Scientific Computing Group is co-author of a paper on "A New Data-Mapping Scheme for Latency-Tolerant Distributed Sparse Triangular Solution." This paper concerns latency-tolerant schemes for the efficient parallel solution of sparse triangular linear systems on distributed memory multiprocessors.

Thursday, November 21

Jason Lee, Dan Gunter, Martin Stoufer and Brian Tierney of the Distributed Systems Department are authors of a paper describing "Monitoring Data Archives for Grid Environments." To determine the source of performance problems in high-performance distributed systems, detailed end-to-end monitoring data from applications, networks, operating systems, and hardware must be correlated across time and space. This paper describes a relational monitoring data archive that allows researchers to view and compare this very detailed monitoring data from a variety of angles.

Brian Tierney is also one of 13 authors of a paper describing "Giggle: A Framework for Constructing Scalable Replica Location Services." In wide area computing systems, it is often desirable to create remote read-only copies (replicas) of files to reduce access latency, improve data locality, and/or increase robustness, scalability and performance for distributed applications. This paper defines a replica location service (RLS) as a system that maintains and provides access to information about the physical locations of copies.

Birds-of-a-Feather Sessions

Leroy Drummond and Osni Marques of the Scientific Computing Group have organized a Tuesday, Nov. 19, session to discuss "ACTS Tools Certification. " The ACTS Collection brings together general-purpose computational tool development projects supported by DOE. The organizers are working on implementing a peer-reviewed certification that will push the frontiers of the tools forward.

Erich Strohmaier of the Future Technologies Group and a founding member of the TOP500 Supercomputers project has organized a Birds-of-a-Feather session about the latest TOP500 list, which will be released at SC2002. Various experts will present detailed analyses of the TOP500 and discuss the changes in the HPC marketplace during the last year.