Tuesday, November 7
Session: Scheduling
1:30 p.m. to 3 p.m.; Room D-267

An Object-Oriented Job Execution Environment
Lance Smith, San Jose State University, and Rod Fatoohi, NASA Ames

This is a project for developing a distributed job execution environment for highly iterative jobs, where the same binary code is run hundreds of times with incremental changes in the input values for each run. An execution environment is a set of resources on a computing platform that can be made available to run the job and hold the output until it is collected. The system allows for fine-grained job control, timely status notification, and dynamic registration and deregistration of execution platforms, depending on resources available. Several objected-oriented technologies are employed: Java, CORBA, UML, and software design patterns. The environment has been tested using a CFD code, INS2D.

Session: MPI/OPENMP
11:30 p.m. to 3 p.m.; Room D-271/273

A Comparison of Three Programming Models for Adaptive Applications on the Origin2000i
Hongzhang Shan, Jaswinder P. Singh, Princeton University, Leonid Oliker, Lawrence Berkeley National Laboratory, Rupak Biswas, NASA Ames

Adaptive applications have computational workloads and communication patterns that change unpredictably at runtime, requiring load balancing to achieve scalable performance on parallel machines. Efficient parallel implementation of such an adaptive application is therefore a challenging task. This paper compares the performance of and the programming effort required for two major classes of adaptive applications under three leading parallel programming models on an SGI Origin 2000 system -- which supports all three models efficiently. Results indicate that the three models deliver comparable performance. However, the implementations differ significantly beyond merely using explicit messages versus implicit loads/stores, even though the basic parallel algorithms are similar.

Wednesday, November 8

State of the Field Talk
8:30 a.m. to 9:15 a.m.; Ballroom C

COTS Cluster Systems for High-Performance Computing
Thomas Sterling, NASA Jet Propulsion Laboratory, High Performance Computing Group, and California Institute of Technology, Center for Advanced Computing Research

In recent years an alternative strategy to achieving high performance, which overcomes the combined problems of cost and architecture variability as well as stability of single suppliers, has emerged. A product of nearly a decade of applied research on workstation clusters, PC clusters, and non-dedicated LAN-connected user desktop and server facilities for cycle harvesting has yielded a rapidly maturing methodology for aggregating and employing low-to-moderate range computer systems in distributed complexes for both capacity and capability workload processing requirements.

COTS clusters are now having significant impact on the realm of processing once reserved to supercomputing. But not all applications are suitable for such loosely-coupled ensembles, and system software environments are still in evolution.

In this talk, Dr. Sterling will explore the history, methodologies, capabilities, and limitations of COTS clusters. He will examine in detail hardware component capabilities and configuration. Sterling will describe software systems and tools for cluster system programming and management, and will present performance and scaling data on successful applications as well as those demonstrating poorer suitability. Finally, Dr. Sterling will discuss examples of near-term research and technology trends that are likely to determine the future directions and capabilities of the next generation of COTS clusters for high-performance computing.

Session: Applications II
1:30 p.m. to 3 p.m.; Room D 274

Parallel Unsteady Turbo-Pump Simulations For Liquid Rocket Engines
Cetin C. Kiris, ELORET/NASA Ames, Dochan Kwak, NASA Ames, William Chan, ELORET/NASA Ames

This paper reports the progress made toward complete turbo-pump simulation capability for liquid rocket engines. The Space Shuttle Main Engine (SSME) turbo-pump impeller is used as a test case for the performance evaluation of the MPI, hybrid MPI/Open-MP, and MLP versions of the INS3D code. Then, a computational model of a turbo-pump has been developed for the shuttle upgrade program. Relative motion of the grid system for rotor-stator interaction was obtained by employing overset grid techniques. Unsteady computations for SSME turbo-pump, which contains 101 zones with 31 million grid points, are carried on Origin 2000 systems at NASA Ames. The approach taken for these simulations, and the performance of the parallel versions of the code are presented.

Thursday, November 9
Session: Data Grid
10:30 a.m.; Room D-267

Computing and Data Grids for Science and Engineering
William E. Johnston, NASA Ames and Lawrence Berkeley National Laboratory; Dennis Gannon, NASA Ames and University of Indiana; Bill Nitzberg, Veridian Systems, PBS Products; Leigh Ann Tanner, Bill Thigpen, Alex Woo, NASA Ames

We use the term "grid" to refer to a software system providing uniform and location independent access to geographically and organizationally dispersed, heterogeneous resource that are persistent and supported. While, in general, grids will provide the infrastructure to support a wide range of services in the scientific environment (e.g. collaboration and remote instrument control), in this paper we focus on services for high performance computing and data handling. We describe the services and architecture of NASA's Information Power Grid ("IPG") -- an early example of a large-scale grid --and some of the issues that have come up in its implementation.

Session: Gordon Bell II
10:30 a.m. to 12 p.m.

High Performance Reactive Fluid Flow Simulations Using Adaptive Mesh Refinement on Thousands of Processors
A.C. Calder, University of Chicago, B.C. Curtis, Lawrence Livermore National Laboratory, L.J. Dursi, B. Fryxell, University of Chicago, G. Henry, Intel Corporation, P. MacNeice, K. Olson, NASA Goddard, P. Ricker, R. Rosner, F.X. Timmes, University of Chicago, H.M. Tufo, University of Chicago/Argonne National Laboratory, J.W. Turan, M. Zingale, University of Chicago

We present simulations and performance results of nuclear burning fronts in supernovae on the largest domain and at the finest spatial resolution studied to date. These simulations were performed on the Intel ASCI-Red machine at Sandia National Laboratories using FLASH, a code developed at the Center for Astrophysical Thermonuclear Flashes at the University of Chicago. FLASH is a modular, adaptive mesh, parallel simulation code capable of handling compressible, reactive fluid flows in astrophysical environments. We describe the key algorithms and their implementation, as well as the optimizations required to achieve sustained performance of 238 GFLOPS on 6,420 processors of ASCI-Red in 64 bit arithmetic.

Panel Discussions

Thursday, November 9
3:30 p.m. to 5 p.m.

Petaflops Around the Corner: When? How? Is it Meaningful?
Moderator: Neil Pundit, Sandia National Laboratories
Panelists: Marc Snir, IBM Research, Bill Camp, Sandia National Laboratories, Thomas Sterling, NASA JPL / Caltech, Paul Messina, DOE HQ, Rick Stevens, Argonne National Laboratory, Pete Beckman, Turbolabs

Teraflops-capable machines are being deployed now, and a petaflops-capable computing complex is just around the corner. This panel will address questions about the necessity of computers with this level of power, the competing design issues to attain a petaflop, and whether applications can actually make good use of such an extreme capability by the end of this decade. Expert panelists represent competing design schemes, extreme application drivers, and dissenting opinions about the need for this level of power in this decade.

Friday, November 10
10:30 a.m. to Noon

Megacomputers
Moderator: Larry Smarr, University of California, San Diego
Panelists: Andrew Chien, Entropia, Ian Foster, Argonne National Laboratory, Thomas Sterling, JPL, David Anderson, United Devices Andrew Grimshaw, University of Virginia

We are nearing a major discontinuity in the evolution of high-performance computing, creating a third era of supercomputing. The first era, which ended with the Cray 1, sought performance by building the fastest single processor possible. The second era, starting with the Illiac IV, sought performance through tens to thousands of identical processors. The third era couples very large numbers of heterogeneous computers on networks to create a virtual supercomputer. Questions to be addressed by the panel include: What is the software architecture model? What are the classes of applications most likely to move to this new computing fabric? And what new computer science research frontiers are important in this more "biological" effervescent style of architecture?

Research Gems