Edward R. McCracken is Chairman and Chief Executive Officer of Silicon Graphics, Inc., of Mountain View, California, the leading manufacturer of high performance visual computing systems. McCracken has an M.B.A. from Stanford University and a B.S.E.E. from Iowa State University. He serves on the Board of Directors of Tularik, Inc., a privately-held biotechnology company. He also serves as co-chairman of Joint Venture: Silicon Valley Network, Inc., co-chairman of the United States Advisory Council on the National Information Infrastructure, and chairman of PRASAD America, a charitable foundation.
Nearly two decades ago, predictions were made that the computer would replace the wind tunnel. While the prediction has not come true, the dependence on Computational Fluid Dynamics (CFD) in the aircraft design process has increased dramatically. IofNEWT combines the two methodologies, wind tunnels and computations, together with experimental instrumentation, supercomputers, scientific visualization, networking, remote communications, and expert systems, to create a real-time aerodynamic development environment. The resulting synergy produces a capability for providing increased data and its subsequent conversion to understanding and knowledge of aerodynamic phenomena, as well as changing the paradigm for wind tunnel testing. A sub-element of IofNEWT is the Remote Access Wind Tunnel (RAWT) which allows real-time access and interaction with the wind tunnel and supercomputing facilities from remote user sites. The presentation will address the evolution of the IofNEWT program, including RAWT, and implementations accomplished thus far. During the 1994 wind tunnel tests (a subsonic commercial transport and a supersonic commercial transport), experimental results (pressure taps, pressure sensitive paint, oil flows, loads and tunnel conditions) were combined in real-time with numerical simulation data of the wind tunnel configurations, including complex facility effects. The results were presented simultaneously to on-site and off-site users, with off-site interaction capabilities.
A common viewpoint is that high performance computing offers the potential to revolutionize our approach to modern scientific and engineering practice. Unfortunately, in the debate about just how to realize the benefits much of the discussion is narrowly focused on just the hardware. What is missing is a more pragmatic assessment of how to reduce the elapsed time between problem definition and its solution. This presentation will discuss the need for a seamless integration of such elements as algorithm development, problem mapping, high level compiler technologies, documentation, and display of results. Particular attention will be given to the issue of how to capture the potential offered by massively parallel processing and Gbit/s network-based supercomputing. The talk will be illustrated with many practical examples that range from a combinatorial optimization based approach to the design of new chemical plants, computational chemistry, resource scheduling in the presence of uncertainty, and scientific visualization. Several new algorithms will be presented for the solution of large systems of differential equations. The new methods, which require modern compiler technology to ensure optimal problem mapping strategies, are 3-5 orders of magnitude faster than existing techniques. The final component of the presentation will be a discussion of the educational issues and, in particular, the need for more programs in scientific and engineering computation.
Over the last two decades, the techniques of computer modeling and simulation have become increasingly important to the fields of Bioengineering and Medicine. Although biological complexity outstrips the capabilities of even the largest computational systems, and will for some time to come, the computational methodology has taken hold in biology and medicine and has been used successfully to suggest physiologically and clinically important scenarios and results. The sophistication of advanced architectures and parallel programming models have created a corresponding increase in the level of biological complexity that can realistically be modeled. Computational problems in medicine require a researcher to apply diverse skills in confronting problems involving very large data sets, modeling and visualizing complex three-dimensional geometries, numerical analysis, and substantial amounts of large scale computing. This talk will illustrate the application of high performance computing techniques to computationally intensive field problems found in electrophysiology. Large scale bioelectric field problems range in scope from single cells up to models incorporating full human anatomical structures. The focus will be on direct and inverse problems which arise in electrocardiography (ECG) and electroencephalography (EEG). The solutions to these problems have utility in many clinical applications such as defibrillation studies, detection of cardiac arrhythmias, noninvasive electric field imaging, and localization and analysis of spontaneous brain activity. To tackle these problems, my research group has developed 1) software tools to construct, manipulate, and display large scale, three-dimensional geometric models and images, 2) interactive, three-dimensional adaptive finite element programs for elliptic partial differential equations with general boundary conditions and source terms, and 3) complete visualization systems for displaying the results.
Tuesday, November 15
10:30 AM - 12:00 PM
Panelists: Charles Grassl, Cray Research, Inc.; Roger W. Hockney, Independent Consultant;
Walter Kohler, Digital Equipment Corporation; Aad van der Steen, Academic
Computing Centre Utrecht;
Harvey Wasserman, Los Alamos National Laboratory
The proliferation of massively parallel processors (MPPs) has made it very difficult for users to accurately assess which machine is "best" for a specific user application. Simply testing one or more kernels on a new machine is insufficient because slight program modifications can produce large variations in performance and thus could lead users to erroneous conclusions about the behavior of the entire application. Most industrial applications are far too large to be easily or quickly benchmarked. Although there are many existing benchmark suites, can performance with any of those accurately predict the level of efficiency for a specific industrial application? This is the dilemma potential MPP users face. The intent of this panel session is to provide the audience with practical advice with which to evaluate MPP performance and to warn users about the pitfalls of performance metrics. All the panelists have had extensive benchmarking experience. Topics to be discussed include: assessing MPP performance for scientific, engineering, and commercial data processing applications (including database and transaction processing); limitations of existing benchmark suites; user guidelines for performance assessment; performance tradeoffs among modestly and massively parallel computers as well as high-end workstations; deceptions in measuring MPP performance.
Panelists: Nora Sabelli, National Science Foundation; Wallace Feurzeig, BBN
Systems and Technologies;
Nancy Roberts, Lesley College; Mitchell Resnick, MIT; Elliot Soloway,
University of Michigan;
Gene Stanley, Boston University
This panel derives from a February 1994 workshop on computer modeling and simulation in science education, under a project supported by the National Science Foundation. Computer modeling can dramatically enliven science education by engaging students in active investigation and providing compelling experiences that enhance scientific insight and understanding. Computational facilities once limited to the science research community are becoming increasingly available for use in pre-college science education. These include networking resources, parallel modeling, visualization, computer-based laboratory probes, and virtual reality environments. However, there is a significant lag between the new technological capabilities and their effective educational implementation. This panel will address the question: what research issues need to be addressed in bringing modeling into the K-12 science curriculum? Discussions will center on several technological and pedagogical issues viewed as key and central for guiding educational modeling research. These include the relationship between the modeling methods and tools used in science research and those used for modeling in science education, the relationship between learning to use models and learning to design and build models, and the relationship between computer visualization of a model's output behavior and computer visualization of a model's structure and component processes as it runs.
Wednesday, November 16
8:30 AM - 12:00 PM
Moderators: Richard Allen, Sandia National Laboratories; Barb Helland, Ames Laboratory
This panel will focus on issues facing large and small post secondary educational institutions as they implement computational science courses, curricula, and degree programs. A survey of existing undergraduate and graduate programs will presented. Consideration will then be given to the interface among high school, undergraduate, and graduate programs in computational science. Broad issues concerning the implementation of these programs in small colleges as well as major universities will be addressed, including: the creation of multidisciplinary programs; student preparation and prerequisites; and access to hardware and software tools. Finally, the panel members will present their model of a graduate degree program in computational science for audience critique and discussion.
Moderator: Donald A. B. Lindberg, National Coordination Office for
HPCC
Organizer: Sally Howe, National Coordination Office for HPCC
Panelists: Representatives of Federal Agencies on the High Performance
Computing, Communications,
and Information Technologies (HPCCIT) Committee
Organizer: Henry Shay, Lawrence Livermore National Laboratory
Panelists: S. Pappert, MIT; L. Sandberg, Pacific Bell; D. Schutzer, Citibank;
J. Tennenbaum, Enterprise Integration Technologies Corporation
As is evident from the high level of participation of National Information Infrastructure (NII) projects at SC'94, there is a great deal of activity in NII throughout the country. A report of the National Research Council, "Realizing the Information Future," has, however, criticized the administration's approach to NII for having inadequate technical depth. On the other hand, other critics claim that NII is too "top-down." This panel discussion seeks to examine the evolving impact of NII in a variety of application domains of general public interest: health care, education, finance, etc. In terms of these areas, what is NII's grade at this time?
Moderator: Wayne Pfeiffer, San Diego Supercomputer Center
Panelists: Rich Hirsh, NSF; Steven Hotovy, Cornell Theory Center; Nick
Nystrom, Pittsburgh Supercomputing Center; Paul Smith, NASA; Dave Rudy, NASA
Langley Research Center; Thomas Sterling, USRA CESDIS
The Joint NSF-NASA Initiative in Evaluation (JNNIE: pronounced "genie") is an important collaborative study among nine NASA and NSF high performance computing research centers to evaluate the capabilities and deficiencies of the major commercial scalable parallel computing systems. By employing a common set of metrics, sharing high performance computing resources, and investigating a wide array of scientific applications, JNNIE investigators are developing results that permit meaningful comparisons to be made among parallel systems and programming methodologies. JNNIE findings go beyond benchmarking to reveal the factors contributing to observed behavior of parallel systems. The focus of this year and a half long effort is on performance, sources of performance degradation, scalability, and ease-of-use. Among the important results presented will be architecture features and software technology that support or impede parallel computing effectiveness. The purpose of the panel session at Supercomputing '94 is to announce the findings of this study to the supercomputing community. The panel session will present the experimental methodology, measurements, analysis, and conclusions. Also, recommendations of near term directions in the evolution of commercial scalable parallel computing systems will be given. An extended period for discussion will be provided to permit questions and views to be expressed by attendees.
Organizer: Ernest F. Brickell, Sandia National Laboratories
The escrowed encryption program which resulted in the Clipper chip has raised intense debate between privacy advocates, industry leaders, and law enforcement. This session will bring together advocates and adversaries of key escrowing to present the status of the existing program, discuss the issues, and to project future issues such as international key escrowing.
As the Nation's Information Superhighway becomes a reality, it is clear that its character will ultimately be determined by uses which are in some cases substantially different from the scientific and engineering applications which pioneered the Net. To help us better understand these opportunities and the requirements which come along with them, panelists representing relatively new and innovative applications of Internet, such as community computing, education, network transactions, and multimedia, together with some more traditional users, will present their views on the future of the Net.
Tuesday, November 15
10:30 AM - 12:00 PM
Moderator: Milissa M. Benincasa, Rome Laboratory
High Performance Computing (HPC) is currently being used to solve problems in the scientific and information processing areas. However, the lack of software support has deterred many users from utilizing parallel computers. One such group of users are real-time embedded system developers. Very little attention has been paid to the needs of real-time systems utilizing high performance computers. Both the Department of Defense (DoD) and the commercial sector can benefit from the use of parallel computer real-time systems. For example, the Global Surveillance and Communications thrust is one area in the DoD that could benefit from utilizing high performance computers. In the commercial sector multi-media has requirements for real-time processing. This roundtable session will focus on addressing the issues and requirements related to applying high performance computing to real-time embedded systems development. In particular, this roundtable session will address the following questions:
1. Can existing sequential methods and techniques employed in building real-time systems be extended to support real-time systems that will utilize HPC? Will this transition from sequential computing to HPC be evolutionary or revolutionary?
2. Is system software support (operating systems, languages, libraries) currently available for real-time system development on HPCs? If not, what is required?
3. Can commercial HPCs be utilized for real-time systems or are custom HPC systems required? If custom HPC systems are required, why?
Moderators: David Martin, Iowa State University; Thomas Marchioro, Ames Laboratory
The ever increasing need for mathematical/computational literacy within our society and among members of the work force has generated enormous pressure to revise and improve the teaching of related subjects throughout the curriculum, particularly at the undergraduate level. The Calculus Reform movement is perhaps the best-known example. This Roundtable will be an informal and open discussion of the salient issues confronting efforts to improve and expand the teaching of computational science as a problem oriented, interdisciplinary approach to scientific investigation.
Among the questions to be considered are:
1. How can faculty/researchers obtain the recognition necessary to further excellence in teaching in the mathematical/computational sciences?
2. What sort of educational resources -- both hardware and software -- are needed to teach computational science at the undergraduate level? Are traditional procedural languages sufficient? Are PCs enough? Are massively parallel platforms needed?
3. How can electronic educational materials be distributed in an efficient way? Can they be made interactive in nature? How should such materials be tied to the World Wide Web and the growing "Information Superhighway?"
4. Can these teaching materials be adapted for use by teachers' colleges? And/or by industry for (re)training of staff?
5. How should efforts to improve undergraduate education be integrated with current programs aimed at the high school level? Specifically, what new kinds of preparation might be needed from K-12, freshman math, etc., to improve mathematical/computational literacy throughout the curriculum?
6. What is the role of the national laboratories and high performance computing centers in education at the undergraduate level?
Moderators: Neil Madsen, Lawrence Livermore National Laboratory;
Mark Christon, Lawrence Livermore National Laboratory;
John DeFord, MacNeal-Schwendler Corporation
Generation of the complex input meshes for large-scale scientific applications is probably the greatest impediment to the use of both the current and next generations of high performance computers for numerical simulation. Industry and research leaders will describe the current state of the art for generation of meshes for large-scale scientific problems. This will be followed by a panel and general audience discussion about the issues surrounding building meshes on the order of 10**7-10**9 grid points in terms of algorithms and architectures.
Moderator: Diane T. Rover, Michigan State University
This roundtable will assemble leading practitioners and researchers from all sectors for a discussion on the fertile area of performance evaluation tools and on the issue of tool integration and interoperability. Performance evaluation spans a range of activities, including modeling and simulation, measurement and monitoring, visualization, benchmarking, optimization, and prediction. Moreover, each type of parallel system has various levels or perspectives at which evaluation can be done, ranging from the application to the machine. On top of all this are different groups of users, each of which has its own interests, questions, and needs. With such an immense "playing field," the techniques and tools that have been developed for the performance evaluation of parallel and distributed computing systems are manifold. On the one hand, this is appealing, since no single technique or tool will be sufficient. On the other hand, this hinders the development of well-defined methodologies for obtaining optimal performance and answering users' performance questions. One of the reasons for this is the lack of integration in the form of environments and frameworks. Often it's not clear how one tool relates to the others and how it could be used in combination with others. Several standards, such as for performance trace data formats and profiling interfaces, have been evolving, and portable software (e.g., communication and math libraries, benchmark programs) is emerging. However, other information processing areas in which tools have been in abundance have more fully addressed the issue of interoperability; for example, the electronic design automation (EDA) and software development (CASE) industries. Environments or frameworks have been the result. This roundtable will be a forum for assessing whether this area is mature enough for prescribed frameworks to be successfully implemented and, moreover, to be consistently adhered to. The participants will address the advantages and disadvantages, practicalities, recent history, and future of tool integration in high performance computing. These will be debated in the context of factors such as industrial cooperation, academic involvement, current software development platforms, infrastructure for tool integration, levels of integration, and utility to users.
Moderator: Scott E. Fahlman, Carnegie Mellon University
Biomedical imaging takes many forms. The familiar X-rays and CAT scans are now being joined by such technologies as magnetic resonance imaging (MRI) and advanced forms of light microscopy that can observe living cells and tissues as they grow and change. These new technologies place unusual demands upon the computers that have become an integral part of the imaging system, and they present some exciting challenges. We want to enhance, analyze, and display the images in something approaching real-time, which requires computational power at the supercomputer level and beyond. In addition to raw computing cycles, these applications stress the acquisition and display hardware, communication channels, storage systems, operating systems (with the need for real-time response), memory capacity, and our ability to build the necessary software. Ultimately, these computing facilities must be packaged into a stand-alone instrument that can be used in a clinical or laboratory setting by people not trained in computer science.
This roundtable will present three well-known experts in different areas of biomedical image processing. Each will describe the nature of the problems she or he works on and the computational demands of this work, both present and future. Our goal will be to familiarize the audience with the challenges posed by this exciting field and to identify which computational problems and approaches cut across all of these applications.
Moderator: Bill Boas, Essential Communications
We will survey the standard interconnects, protocols and architectures for gigabit networking that are commercially ready for use. The applications discussed are cluster computation, storage management, and digital media distribution. We will cover ATM, SONET, FCS, and HIPPI. The goal for the roundtable is to share technology, product, implementation and operation experience with people who have installed and operated gigabit networks at the conference and at their home sites.
Moderators: Matthew T. O'Keefe, University of Minnesota; Milton Halem, NASA Goddard Space Flight Center
Toward the end of the decade, teraflop computing, exabyte storage, and gigabyte networks will be commonplace technologies. Mass storage requirements at leading supercomputer centers such as NASA Goddard and the National Security Agency are already beyond what current systems can support, and these new technologies will stress mass storage management systems even more. In addition, the storage requirements of the "typical" supercomputer user continue to expand as computer performance increases and storage costs decrease. So what does the future hold for mass storage in high performance computing? This roundtable session will bring together participants from industry, government and academia to discuss mass storage topics, including:
-- industry direction in mass storage technologies, including disk, tape and optical media
-- integration of new technologies, including new networking standards such as Fibre Channel, HIPPI and ATM
-- distributed and network attached mass storage systems
-- security issues
-- migration from existing mass storage systems to newer technology
-- low-cost, department-level mass storage systems for supercomputer users.
Wednesday, November 16
8:30 AM - 10:00 AM
Computation of Ocean Flows Using Unmodeled Conservation Equations
M. Beddhu, M.-Y. Jiang, L. Taylor, D. Whitfield, Mississippi State
University
Global Ocean Modeling on Parallel Computers
R. Smith, Los Alamos National Laboratory
Thursday, November 17
8:30 AM - 10:00 AM
The winner of the Sid Fernbach Award is scheduled to present a lecture during this session.
Chair: Michael Heath, University of Illinois
Tight Binding Molecular Dynamics
S. Goedecker, Cornell Theory Center; L. Colombo, University of Milan
A 65+ Gflops Unstructured Finite Element Simulation of Chemically Reacting
Flows on the Intel Paragon
J. Shadid, S. Hutchinson, H. Moffat, B. Hendrickson, R. Leland, Sandia National
Laboratories;
G. Hennigan, New Mexico State University
Applications of Boundary Element Methods on the Intel Paragon
D. Womble, D. Greenberg, S. Wheat, R. Benner, Sandia National Laboratories; M.
Ingber, University of New Mexico; G. Henry, S. Gupta, Intel Supercomputer
Systems Division
Development and Achievement of NAL Numerical Wind Tunnel (NWT) for CFD
Computations
H. Miyoshi, Foundation for Promotion of Material Science and Technology of
Japan; M. Fukuda, T. Iwamiya,
T. Nakamura, M. Tuchiya, M. Yoshida, K. Yamamoto, Y. Yamamoto, S. Ogawa, Y.
Matsuo, T. Yamane, National Aerospace Laboratory; M. Takamura, M. Ikeda, S.
Okada, Y. Sakamoto, T. Kitamura, H. Hatama, Fujitsu Limited;
M. Kishimoto, Fujitsu Laboratories Limited
Tuesday, November 15
3:30 PM - 5:00 PM
Chair: Bruce Loftis, North Carolina Supercomputing Center
A Scalable High-Performance Environment for Fluid Flow Analysis on Unstructured
Grids
D. Banerjee, T. Tysinger, Fluent Inc.
A Fixed Time Performance Evaluation of Parallel CFD Applications
A. Baker, Iowa State University; S. Ying, Ames Laboratory
A Parallel Iterative Linear Solver for Solving Irregular Grid Semiconductor
Device Matrices
E. Tomacruz, J. Sanghavi, A. Sangiovanni-Vincentelli, University of California
- Berkeley
Chair: Richard Draper, Supercomputing Research Center
A High Performance Parallel Algorithm for 1-D FFT
R. Agarwal, F. Gustavson, M. Zubair, IBM T. J. Watson Research Center
Control Strategies for Parallel Mixed Integer Branch and Bound
J. Eckstein, Thinking Machines Corporation
EXTENT: A Portable Programming Environment for Designing and
Implementing High-Performance Block-Recursive Algorithms
D. Dai, S. Gupta, S. Kaushik, T.-H. Lin, J. Lu, R. Singh, D. Zhou, C.-H. Huang,
P. Sadayappan, Ohio State University; R. Johnson, St. Cloud State
University
Chair: Maya Gokhale, Supercomputing Research Center
An Experimental APL Compiler for a Distributed Memory Parallel Machine
W.-M. Ching, A. Katz, IBM T. J. Watson Research Center
A Library-Based Approach to Portable, Parallel, Object-Oriented Programming:
Interface, Implementation, and Application
S. Parkes, J. Chandy, P. Banerjee, University of Illinois
Efficient Parallel Global Garbage Collection on Massively Parallel Computers
T. Kamada, S. Matsuoka, A. Yonezawa, The University of Tokyo
Chair: Matthew T. O'Keefe, University of Minnesota
SLICC: A Low Latency Interface for Collective Communications on Parallel
Machines
A. Knies, G. Adams, Purdue University; F. Barriuso, W. Harrod, Cray Research,
Inc.
Run-time and Compile-time Support for Adaptive Irregular Problems
S. Sharma, B. Moon, Y.-S. Hwang, R. Das, J. Saltz, University of Maryland; R.
Ponnusamy, Syracuse University
InterCom: A High-Performance Collective Communication Library
M. Barnett, University of Idaho; S. Gupta, D. Payne, Intel Corporation; L.
Shuler, R. van de Geijn, J. Watts, The University of Texas at Austin
Chair: Robert M. Panoff, National Center for Supercomputing Applications
Distributed Exploratorium for High Performance Computational Techniques
S. Iyer, A. Beck, U. Ravaioli, J. Terstriep, Beckman Institute
Are Expectations for Parallelism Too High? A Survey of Potential Parallel Users
C. Cook, C. Pancake, R. Walpole, Oregon State University
The Development and Operation of Edinburgh Parallel Computing Centre's Summer
Scholarship Programme
G. Wilson, N. MacDonald, C. Thornborrow, University of Edinburgh
Chair: Alfred Brenner, Institute for Defense Analyses
Distributed Network Computing over Local ATM Networks
M. Lin, J. Hsieh, D. Du, J. MacDonald, University of Minnesota; J. Thomas,
Minnesota Supercomputer Center, Inc.
Design and Implementation of Multicast Operations for ATM-Based High
Performance Computing
C. Huang, E. Kasten, P. McKinley, Michigan State University
Performance of I/O Subsystems for High-Speed Networks: Case Study of a Fibre
Channel Network
M. Lin, J. Hsieh, D. Du, J. MacDonald, University of Minnesota
Chair: Dave Dougherty, University of Vermont
A Parallel Gauss-Seidel Algorithm for Sparse Power Systems Matrices
D. Koester, S. Ranka, G. Fox, Syracuse University
Efficient Implementation of the Multigrid Preconditioned Conjugate
Gradient Method on Distributed Memory Machines
O. Tatebe, Y. Oyanagi, The University of Tokyo
A Parallel Formulation of Interior Point Algorithms
G. Karypis, A. Gupta, V. Kumar, University of Minnesota
Chair: John Levesque, Applied Parallel Research, Inc.
Affinity Scheduling of Unbalanced Workloads
S. Subramaniam, Sybase, Inc.; D. Eager, University of Saskatchewan
Non-contiguous Processor Allocation Algorithms for Distributed Memory
Multicomputers
W. Liu, V. Lo, K. Windisch, University of Oregon
Generalized Multiprocessor Scheduling for Directed Acylic Graphs
G. Srinivasa Prasanna, AT&T Bell Laboratories; B. Musicus, BBN Inc.
Chair: Dean P. McCullough, Department of Defense
Scalability of the Cedar System
S. Turner, A. Viedenbaum, University of Illinois at Urbana-Champaign
Cache Performance in Vector Supercomputers
L. Kontothanassis, M. Scott, University of Rochester; R. Sugumar, G. Faanes, J.
Smith, Cray Research, Inc.
The Performance of the Cedar Multistate Switching Network
J. Torrellas, Z. Zhang, University of Illinois at Urbana-Champaign
Chair: Dale Haidvogel, Rutgers University
Adaptive Data Parallel Methods for Ecosystem Monitoring
C. Turner, Oasis Research Center, Inc.; J. Turner, Pacific Southwest Forest and Range Experiment Station
PARAMICS: Moving Vehicles on the Connection Machine
G. Cameron, B. Wylie, D. McArthur, The University of Edinburgh
A Distributed, Parallel, Interactive Volume Rendering Package
J. Rowlan, G. Lent, N. Gokhale, Argonne National Laboratory
Chair: Kenichi Miura, Fujitsu America, Inc.
Full Scale Experiments of Nonnumeric Search Problems on the
80-Processor EM-4 Distributed-Memory Multiprocessor
M. Sato, Electrotechnical Laboratory; A. Sohn, New Jersey Institute of
Technology
Implementation of a Portable and Reproducible Parallel Pseudorandom Number
Generator
D. Pryor, S. Cuccaro, M. Mascagni, M. Robinson, Supercomputing Research Center
Parallel Probabilistic Inference for Medical Diagnostic Networks
A. Kozlov, J. Singh, Stanford University
Chair: Piyush Mehrotra, ICASE
Communication and Memory Requirements as the Basis for Mapping Task and Data
Parallel Programs
J. Subhlok, D. O'Hallaron, T. Gross, P. Dinda, J. Webb, Carnegie Mellon
University
Global Arrays: A Portable "Shared-Memory" Programming Model for Distributed
Memory Computers
J. Nieplocha, R. Harrison, R. Littlefield, Battelle-Pacific Northwest
Laboratories
On the Design of Chant: A Talking Threads Package
M. Haines, D. Cronk, P. Mehrotra, NASA Langley Research Center
Chair: Stephen Wheat, Sandia National Laboratories
Scheduling of Unstructured Communication on the Intel iPSC/860
J.-C. Wang, University of Illinois at Urbana-Champaign
Tolerating Node Failures in Cache Only Memory Architectures
A. Gefflaut, C. Morin, M. Banatre, IRISA/INRIA
Application-Specific Protocols for User-Level Shared Memory
B. Falsafi, A. Lebeck, S. Mukherjee, S. Reinhardt, I. Schoinas, M. Hill, J.
Larus, A. Rogers, D. Wood,
University of Wisconsin-Madison
Chair: Ron Minnich, Supercomputing Research Center
Adaptive Load Migration Systems for PVM
J. Casas, R. Konuru, S. Otto, R. Prouty, J. Walpole, Oregon Graduate Institute
of Science and Technology
Performance Evaluation of Three Distributed Computing Environments for
Scientific Applications
R. Fatoohi, S. Weeratunga, NASA Ames Research Center
NAS Experiences with a Prototype Cluster of Workstations
K. Castagnera, W. Kramer, J. Musch, M. Smith, R. Williams, D. Yip, NASA Ames
Research Center; D. Cheng,
R. Fatoohi, E. Hook, C. Manning, C. Niggley, W. Saphir, I. Stockdale, Computer Sciences Corporation; D. Sheppard, S. Welch, Sterling Federal Systems
Chair: Michael Strayer, Oak Ridge National Laboratory
Truly Distribution-Independent Algorithms for the N-body Problem
S. Aluru, J. Gustafson, Ames Laboratory; G. Prabhu, Iowa State University
GRAPE-4: A One-Tflops Special-Purpose Computer for Astrophysical N-body
Problems
J. Makino, M. Taiji, T. Ebisuzaki, D. Sugimoto, The University of Tokyo
Scalable Parallel Formulations of the Barnes-Hut Algorithm for N-body
Simulations
A. Grama, V. Kumar, A. Sameh, University of Minnesota
Chair: Robert W. Leland, Sandia National Laboratories
Genetic Algorithms for Graph Partitioning and Incremental Graph Partitioning
H. Maini, K. Mehrotra, C. Mohan, S. Ranka, Syracuse University
Parallel Incremental Graph Partitioning Using Linear Programming
C.-W. Ou, S. Ranka, Syracuse University
Efficient Communication Algorithms for Pipeline Multicomputers
K. Kee, S. Hariri, Syracuse University
Chair: Thomas L. Sterling, USRA CESDIS
Architecture of the VPP500 Parallel Supercomputer
T. Utsumi, M. Ikeda, M. Takamura, Fujitsu Limited
Sunder: A Programmable Hardware Prefetch Architecture for Numerical Loops
T.-C. Chiueh, State University of New York at Stony Brook
Efficient and Scalable Cache Coherence Schemes for Shared Memory Hypercube
Multiprocessors
A. Kumar, P. Mannava, L. Bhuyan, Texas A&M University
Chair: Lori Pollock, University of Delaware
Expressing Cross-Loop Dependencies through Hyperplane Data Dependence Analysis
A. Zaafrani, M. Ito, University of British Columbia
An Efficient Algorithm for the Run-time Parallelization of DOACROSS Loops
D.-K. Chen, J. Torrellas, P.-C. Yew, University of Illinois at Urbana-Champaign
The Range Test: A Dependence Test for Symbolic, Non-linear Expressions
W. Blume, R. Eigenmann, University of Illinois at Urbana-Champaign
Chair: Chris Johnson, University of Utah
Orientation Determination in the 3D Reconstruction of Icosahedral Viruses using
a Parallel Computer
C. Johnson, N. Weisenfeld, B. Trus, J. Conway, R. Martino, A. Steven, National
Institutes of Health
Factors Controlling Generation and Propagation of Pacemaker
Potentials in Network Models of Mammalian SA Node
R. Winslow, Y.-C. Lai, The Johns Hopkins University; D. Cai, University of
Minnesota
Parallel Protein Structure Determination from Uncertain Data
C. Chen, J. Singh, W. Poland, R. Altman, Stanford University
Chair: Allen D. Malony, University of Oregon
Applications Performance Under OSF/1 AD and SUNMOS on Intel Paragon XP/S-15
S. Saini, H. Simon, NASA Ames Research Center
Paging Tradeoffs in Distributed Shared-Memory Multiprocessors
D. Burger, R. Hyder, B. Miller, D. Wood, University of Wisconsin-Madison
Parallel Performance Prediction Using Lost Cycles Analysis
M. Crovella, T. LeBlanc, University of Rochester
Chair: Dannie Durand, Bellcore
Using High Speed Networks to Enable Distributed Parallel Image Server Systems
B. Tierney, B. Johnston, T. Chen, H. Herzog, G. Hoo, G. Jin, J. Lee, Lawrence
Berkeley Laboratory
Reducing the Variance of Point to Point Transfers in the IBM 9076 Parallel
Computer
R. Mraz, IBM T. J. Watson Research Center
Time and/or Space Sharing in a Workstation Cluster Environment
S. Turner, L. Ni, B. Cheng, Michigan State University
Chair: David Schneider, Cornell Theory Center
Dynamic File-Access Characteristics of a Production Parallel Scientific
Workload
D. Kotz, N. Nieuwejaar, Dartmouth College
An Efficient Abstract Interface for Multidimensional Array I/O
K. Seamons, M. Winslett, University of Illinois
Dynamic I/O Characterization of I/O Intensive Scientific Applications
B. Pasquale, G. Polyzos, University of California, San Diego
Chair: John Nickolls, MasPar Computer Corporation
Fault-Tolerant Routing with Non-Adaptive Wormhole Algorithms in Mesh
Networks
R. Boppana, The University of Texas at San Antonio; S. Chalasani, University of
Wisconsin-Madison
Optimal Software Multicast in Wormhole-Routed Multistage Networks
H. Xu, Y.-D. Gui, L. Ni, Michigan State University
Design and Evaluation of a DAMQ Multiprocessor Network Switch with
Self-Compacting Buffers
J. Park, B. O'Krafka, S. Vassiliadis, IBM Corporation; J. Delgado-Frias, SUNY
Binghamton
Chair: Diane Rover, Michigan State University
A Portable Debugger for Parallel and Distributed Systems
D. Cheng, R. Hood, NASA Ames Research Center
The D Editor: A New Interactive Parallel Programming Tool
S. Hiranandani, K. Kennedy, S. Warren, Rice University; C.-W. Tseng, Stanford
University
Interpreting the Performance of HPF/Fortran 90D
M. Parashar, S. Hariri, T. Haupt, G. Fox, Syracuse University
Chair: Brian Malloy, Clemson University
Enabling Unimodular Transformations
R. Sass, M. Mutka, Michigan State University
Improving the Performance of DSM Systems via Compiler Involvement
R. Mirchandaney, Shell Oil Company; S. Hiranandani, A. Sethi, Rice University
A Compiler-Directed Cache Coherence Scheme with Improved Intertask
Locality
L. Choi, P.-C. Yew, University of Illinois at Urbana-Champaign
Chair: David Bailey, NASA Ames Research Center
Improved Load Distribution in Parallel Sparse Cholesky Factorization
E. Rothenberg, Intel Supercomputer Systems; R. Schreiber, Research Institute
for Advanced Computer Science
A Scalable Parallel Algorithm for Sparse Matrix Factorization
A. Gupta, V. Kumar, University of Minnesota
A High Performance Linear Equation Solver on the VPP500 Parallel
Supercomputer
M. Nakanishi, H. Ina, Fujitsu Limited; K. Miura, Fujitsu America, Inc.
Research Exhibits Conference participants can take the opportunity to see working demonstrations of research projects by visiting the Research Exhibits on the main exhibit floor. These exhibits are, in general, highly interactive, and participants will be able to discuss the projects with their developers one-on-one. As part of the full conference exhibit display, the Research Exhibits will be open during all exhibition hours, including the opening reception on Monday evening. The location and a description of each of the research exhibits will be included in the Final Program.
Posters The poster sessions at Supercomputing '94 will allow participants to see and discuss late-breaking research results with authors. Technical poster presentations will include both traditional posters as well as video-display posters in two sessions. The first session will be held on Tuesday, 3:30 PM to 5:00 PM, while the second will be held on Tuesday, 5:30 PM to 7:00 PM, with a reception. Abstracts of all posters will be available in the Final Program. Posters from the K-12 education poster session on Sunday night will be on display Monday and Tuesday.
Heterogeneous Computing Challenge A number of research exhibitors will be participating in the Heterogeneous Computing Challenge, a competition designed to demonstrate the effective use of heterogeneous computing technology in solving real application problems. During the conference, participants in the challenge will be presenting their demonstrations in real-time using SCinet '94 to access machines on the exhibit floor as well as at remote sites. These demonstrations will be judged and awards will be given for Third Place, Second Place, and Best Overall Challenge. A technical session in which the participant groups will present an overview of their challenge entry will be held Thursday, 10:30 AM to 12:00 PM.
SuperViz '94 SuperViz '94 will showcase state-of-the-art visualizations created exclusively by the high performance computing community. Entries will span the many areas of supercomputer applications and will include visualizations produced using high performance computing technology. A SuperViz '94 video will be made available to all participants and will be shown on a large screen during the conference on both Tuesday, 10:30 AM to 5:00 PM, and Wednesday, 8:30 AM to 5:00 PM.
Proceedings Supercomputing '94, in cooperation with the IEEE Computer Society, will offer an electronic version of the table of contents of the proceedings along with a complete set of abstracts, in addition to the traditional hard copy proceedings. Abstracts of all technical papers, posters, and research exhibits will be included. Full text versions of all technical papers received in electronic form will be included as well. The database will be accessible over the Internet prior to the meeting at the URL:
http://info.computer.org/p3/SC94Home.html
Details on how to access the database will be provided in the registration packet and demonstrations will be given during the conference.
University Education Activities An undergraduate education roundtable and a technical paper session on university education are both scheduled on Tuesday. There will also be a panel session on Wednesday morning discussing undergraduate and graduate education. These sessions will address issues currently being debated by educators and administrators who are implementing or planning to implement computational science or engineering programs. Further information is provided on the following pages.
K-12 Education Activities A three-day "mini-conference" for K-12 educators is scheduled for Sunday, Monday and Tuesday, November 13-15. The K-12 activities include hands-on sessions and demonstrations with Internet access, a poster display session and reception, and selected paper presentations by teachers. In addition, a panel related to K-12 education will be held on Tuesday. Further information is provided on the following pages.
High School Day On Wednesday, November 16, high school students and their teachers from the Washington, D.C. area will visit Supercomputing '94. Special sessions and a tour through the exhibit hall will be provided. High School Day is being organized by Mary Ellen Verona, Montgomery Blair High School, Silver Spring, Maryland.
For further information on the Education Program, please contact:
Moderators: David Martin, Iowa State University; Thomas Marchioro, Ames Laboratory
The ever increasing need for mathematical/computational literacy within our society and among members of the work force has generated enormous pressure to revise and improve the teaching of related subjects throughout the curriculum, particularly at the undergraduate level. The Calculus Reform movement is perhaps the best-known example. This Roundtable will be an informal and open discussion of the salient issues confronting efforts to improve and expand the teaching of computational science as a problem oriented, interdisciplinary approach to scientific investigation.
Among the questions to be considered are:
1. How can faculty/researchers obtain the recognition necessary to further excellence in teaching in the mathematical/computational sciences?
2. What sort of educational resources -- both hardware and software -- are needed to teach computational science at the undergraduate level? Are traditional procedural languages sufficient? Are PCs enough? Are massively parallel platforms needed?
3. How can electronic educational materials be distributed in an efficient way? Can they be made interactive in nature? How should such materials be tied to the World Wide Web and the growing "Information Superhighway?"
4. Can these teaching materials be adapted for use by teachers' colleges? And/or by industry for (re)training of staff?
5. How should efforts to improve undergraduate education be integrated with current programs aimed at the high school level? Specifically, what new kinds of preparation might be needed from K-12, freshman math, etc., to improve mathematical/computational literacy throughout the curriculum?
6. What are the roles of the national laboratories and high performance computing centers in education at the undergraduate level?
Chair: Robert M. Panoff, National Center for Supercomputing Applications
Distributed Exploratorium for High Performance Computational Techniques
S. Iyer, A. Beck, U. Ravaioli, J. Terstriep, Beckman Institute
Are Expectations for Parallelism Too High? A Survey of Potential Parallel Users
C. Cook, C. Pancake, R. Walpole, Oregon State University
The Development and Operation of Edinburgh Parallel Computing Centre's Summer
Scholarship Programme
G. Wilson, N. MacDonald, C. Thornborrow, University of Edinburgh
Wednesday, November 16
8:30 AM - 12:00 PM
Moderators: Richard Allen, Sandia National Laboratories; Barb Helland, Ames Laboratory
This panel will focus on issues facing large and small post secondary educational institutions as they implement computational science courses, curricula, and degree programs. A survey of existing undergraduate and graduate programs will presented. Consideration will then be given to the interface among high school, undergraduate, and graduate programs in computational science. Broad issues concerning the implementation of these programs in small colleges as well as major universities will be addressed, including: the creation of multidisciplinary programs; student preparation and prerequisites; and access to hardware and software tools. Finally, the panel members will present their model of a graduate degree program in computational science for audience critique and discussion.
Organizers: Sharon Carruth, Ann Hernandez, Nichols Research Corporation
K-12 educators may choose up to five two-hour Hands-on sessions from the list below. Personal computers will be connected to the Internet and attendees will be able to gain beginning and advanced knowledge in computational science and high performance computing related to areas that they can take back to their classrooms. Additional time for open access to the Hands-on laboratory will be available on Monday from 3:30 PM to 6:00 PM and Tuesday from 3:30 PM to 5:00 PM.
Internet: Navigating the Information Superhighway
Mary Jo Gray, Caroline Gershner, DeValls Bluff High School, DeValls Bluff, AR
Internet: Advanced Resources and Tools
Jane Jones, J.O. Johnson High School, Huntsville, AL; Gary Harper, Andalusia
High School, Andaluasia, AL
Ray Tracing & Climate Modeling
Brian Lindow, Lawrence Livermore National Laboratory
Advanced Ray Tracing & Climate Modeling
Brian Lindow, Lawrence Livermore National Laboratory
The Weather Underground: Interactive Weather on the Information
Superhighway
Perry J. Samson, University of Michigan; Parker Pennington, Michigan Public
Schools
Exploring the Internet with Mosaic
Jill Snyder, Albuquerque High School, Albuquerque, NM; Gina Sullivan, Bob Jones
High School, Madison, AL
Creating Mosaic Documents
Dave Bergandine, University High School, Champaign, IL
Ground Truth Studies (GTS) A New Tool for Understanding Global Environmental
Change
Greg Cox, University of Alabama in Huntsville
Using Visualization to Enhance Middle and High School Science Studies
Curtis Peters, Kelly Gobble, Summertown School, Summertown, TN
Network Hardware and Technology in Local Area Networks
Pat Burns, Greg Redder, Colorado State University
Organizer: Barbara Summers, Oak Ridge National Laboratory
Posters will be displayed presenting various programs in K-12 computational science and high performance computing. All teachers attending Supercomputing '94 who attended the education program of Supercomputing '93 or '92 will display posters on what their school has accomplished in the past one or two years. Other programs or teachers who wish to display a poster at the session are invited to apply.
Posters will remain on display through Tuesday afternoon.
Chair: Jane Jones, J.O. Johnson High School
Collaborative Methods of Communication and Visualization
Don Thatcher, North High School, Des Moines, IA; Edna Gentry, J.O. Johnson High
School, Huntsville, AL
Moderator: Nora Sabelli, National Science Foundation
Panelists: Nora Sabelli, National Science Foundation; Wallace Feurzeig, BBN Systems and Technologies; Nancy Roberts, Lesley College; Mitchell Resnick, MIT; Elliot Soloway, University of Michigan; Gene Stanley, Boston University
This panel derives from a February 1994 workshop on computer modeling and simulation in science education, under a project supported by the National Science Foundation. Computer modeling can dramatically enliven science education by engaging students in active investigation and providing compelling experiences that enhance scientific insight and understanding. Computational facilities once limited to the science research community are becoming increasingly available for use in pre-college science education. These include networking resources, parallel modeling, visualization, computer-based laboratory probes, and virtual reality environments. However, there is a significant lag between the new technological capabilities and their effective educational implementation. This panel will address the question: what research issues need to be addressed in bringing modeling into the K-12 science curriculum?
Discussions will center on several technological and pedagogical issues viewed as key and central for guiding educational modeling research. These include the relationship between the modeling methods and tools used in science research and those used for modeling in science education, the relationship between learning to use models and learning to design and build models, and the relationship between computer visualization of a model's output behavior and computer visualization of a model's structure and component processes as it runs.