DOCUMENT RESUME

ED 336 083 IR 015 102

TITLE Grand Challenges: High Performance Computing andCommunications. The FY 1992 U.S. Research andDevelopment Program.

INSTITUTION Federal Coordinating Council for Science, Engineering

and Technology, Washington, DC.

PUB DATE 91

NOTE 64p.; Report by the Committee on Physical,Mathematical, and Engineering Sciences to supplementthe President's Fiscal Year 1992 budget. For relateddocuments, see ED 332 693-694, IR 015 103, IR 015105, IR 015 125, and IR 015 204.

AVAILABLE FROM Committee on Physical, Mathematical, and EngineeringSciences, c/o National Science Foundation, Computerand Information Science and Engineering, 1800 GStreet, N.W., Washington, DC 20550.

PUB TYPE Viewpoints (Opinion/Position Papers, Essays, etc.)

(120) -- Reports - Descriptive (141)

EDRS PRICE MF01/PC03 Plus Postage.

DESCRIPTORS *Computer Networks; Computer Software; EconomicOpportunities; Elementary Secondary Education;*Federal Programs; Higher Education; Human Resources;*Information Technology; National Security; *Researchand Development; Science and Society; *TechnologicalAdvancement

IDENTIFIERS *High Performance Computing; *National Research IndEducation Network; Supercomputers

ABSTRACTThis report presents a review of the High Performance

Computing and Communications (HPCC) Program, which has as its goal

the acceleration of the commercial availability and utilization ofthe next generation of high performance computers and networks inorder to: (I) extend U.S. technological leadership in highperformance computing and computer communications; (2) provide widedissemination and application of the technologies both to speed the

pace of innovation and to serve the national econoay, national

security, education, and the global environment; and (3) spur gains

in U.S. productivity and industrial competitiveness by making high

performance computing and networking technologies an integral part of

the design and production process. An executive summary, which opensthe report, is followed by four chapters: (1) Program Goals andOverview (program description, needs and benefits, and programexecution strategy); (2) HPCC Program Components (High PerformanceComputing Systems, Advanced Software Technology and Algorithms, theNational Research and Education Network--NREN--and Basic Research and

Human Resources); (3) Program Development and Agency Budgets (program

planning, evaluation criteria, and agency program descriptions forthe Defense Advanced Research Projects Agency, Department of Energy,National Aeronautics and Space Administration, National ScienceFoundation, NatIonal Institute for Standards and Technology, Nationa/

Ocea:,ic and Atmospheric Administration, Environmental ProtectionAgency, and the National Library of Medicine; and (4) Grand Challenge

and Supporting Technology Case Studies (forecasting severe weather

events, the study of cancer genes, predicting new superconductors,air pollution, aerospace vehicle design, energy rmnservation and

turbulent combustion, microsystems design and pakaging, the earth's

biosphere, high speed networks, and education using the NREN. Eight

figures and a glossary are included. (DB)

U S. DEPARTMENT OF EDUCATIONOffice of Educational Research and Improvement .EDUCATIONAL RESOURCES INFORMATION

iCENTER (ERIC)

This document has been reproduced asreceived from the person or organizationoriginating it

O Minor changes have been made to improvereproduction duality

Points of view or opinions stated in this doc.ment do not necessarily represent offiOERI position or policy

11111111...__.

P

44Wal

s.I 0r '..\,, I %I

I 1

1 ,I 1

..__Att Ah_ III i

;

1 2 3

4,

On the Cover:1. Numerically modelled thunderstorm.2. Computational model of chemical carcinogen binding

with DNA molecule.3. Visualization of structure of superconducting material.4. Simulation of acid rain pollutants over Ohio River basin.5. Aerodynamic characteristics of space vehicle: computer

simulated versus wind tunnel data.6. Photo of wafer and multichip prototypes.7. Numerical simulation of fuel jet.8. Computer image of earth's biosphere components

generated from satellite data.The images used in this report were produced by ongoing scientificprojects in areas of the planned HPCC program. They were selected toillustrate the breadth of subject matter of the HPCC program, and areelaborated upon in Chapter 4. The U.S. map suggests how the NationalResearch and Education Network supports geographically distributedcollaborative research activities.

3

Grand Challenges:High Performance Computing

and Communications

The FY 1992 U.S. Research and Development Program

A Report by the Committee onPhysical, Mathematical, and Engineering Sciences

Federal Coordinating Council for Science,Engineering, and Technology

Office of Science and Technology Policy

To Supplement the President's Fiscal Year 1992 Budget

Office of Science and Technology PolicyFederal Coordinating Council for Science,

Engineering, and TechnologyCommittee on Physical, Mathematical, and

Engineering SciencesActing Chairman

Charles Herzfeld, Department of Defense

MembersBarry Williamson, Department of the InteriorCharles E. Hess, Department of AgricultureRobert White, Department of CommerceWilliam Raub, Department of Health ana Human ServicesJames Decker, Department of EnergyJohn B. Childers, Department of EducationNorine Noonan, Office of Management and BudgetEugene Wong, Office of Science and Technoiogy PolicyArnold Aldrich, National Aeronautics and Space AdministrationErich Bretthauer, Environmental Protection AgencyFredrick M. Bernthal, National Science Foundation

Executive SecretaryJane Stutsman, National Science Foundation

FCCSET DirectorateMaryanne C. Bach, Executive DirectorCharles H. Dickeis, Senior Staff AssociateJean Grace, Exec itive Assistant

High Performance Computing and Communications Working GroupDavid Nelson, Department of Energy, ChairmanBarry Boehm, Defense Advanced Research Projects Agency, CoChairmanCharles N. Brownstein, National Science Foundation, CoChairmanLee Holcomb, National Aeronautics and Space Administration, CoChairmanLawrence E. Brandt, National Science FoundationJohn S. Cavallini, Department of EnergyMelvyn Cirnent, National Science FoundationJack D. Fellows, Office of Management and BudgetStephen M. Griffin, National Science FoundationPaul E. Hunter, National Aeronautics and Space AdministrationGary M. Johnson, Department of EnergyThomas A. Kitchens, Department of EnergyNorman H. Kreisman, Department ofEnergyAlbert T. Landberg, Jr., National Institute of Standards and TechnologyFred Scoresby Long, National Oceanic and Atmospheric AdministrationDaniel R. Masys, National Library of MedicineJoan H. Novak, Environmental Protection AgencyWilliam L. Scherlis, Defense Advanced Research Projects AgencyPaul H. Smith, National Aeronautics and Space AdministrationK. Speierman, National Security AgencyStephen L Squires, Defense Advanced Research Projects AgencyStephen Wolff, National Science Foundation

EXECUTIVE OFFICE OF THE PRESIDENTOFFICE OF SCIENCE AND TECHNOLOGY POLICY

WASHINGTON, D.C. 20506

MEMBERS OF CONGRESS:

I am pleased to forward with this letter "Grand Challenges: High PerformanceComputing and Communications, The FY 1992 U. S. Research and DevelopmentProgram," a report by the Committee on Physical, Mathematical, and EngineeringSciences of the Federal Coordinating Council for Science, Engineering, and Technology,

a supplement to the President's Fiscal Year 1992 Budget.

The report presents an ambitious and well-coordinated research and developmentprogram designed to sustain and extend U.S leadership in all advanced areas of

computing and networking. The program not only provides a far-sighted vision for theunderlying technologies but also gives recognition to the importance of both humanresources and those applications that serve .%-ajor national needs. This is a program ofnational investment that will bring both economic and social dividends.

The program is strategically related to other key components of the President's overallapproach to challenges in science, technology, and education. It provides for the use ofimproved computational and c:.mmunications technologies to contribute to more effectivesolutions of grand challenge problems.

The goal of the Federal High Performance Computing and Communications (HPCP,)Program is to accelerate significantly the commercial availability and utilization of thenext generation of high performance computers and networks. Recent advances offer thepotential for a thousand-fold improvement in useful computing capability and a hundred-fold improvement in available computer communications capability by 1996. These

advances will come through improvements in hardware and software. This increasedcapability will greatly expand the availability of these resources for research andeducation. It is my personal view, moreover, that the successful implementation or thisprogram will lay the foundation for changes in education at all levels.

Several years of effort on the part of senior government, industry, and academil scientistsand managers are reflected in this program. Acting Chairman Charles Herzfeld and hisinteragency committee members, associates, and staff are to be commended on theexcellent work that is manifest in both the program and the report.

$41"1146141D. Allan BromDirector

To obtain a copy of this document send request to:

Committee on Physical, Mathematical, and Engineering Sciencescio National Science FoundationComputer and Information Science and Engineering1800 G Street, N.W.Washington, D.C. 20550

Table of Contents

Page

Executive Summary 2

1, Program Goals and OverviewIntroductionNeeds and BenefitsProgram DescriptionGoalsStrategyProgram Execution Strategy

2. HPCC Program ComponentsHigh Performance Computing SystemsAdvanced Software Technology and AlgorithmsNational Research and Education NetworkBasic Research and Human Resources

5

12

3. Program Development and Agency Budgets 22

Program PlanningEvaluation CriteriaAgency BudgetsAgency Program Descriptions

Defense Advanced Research Projects AgencyDepartment of EnergyNational Aeronautics and Space AdministrationNational Science FoundationNational Institute for Standards and TechnologyNational Oceanic and Atmospheric AdministrationEnvironmental Protection AgencyNational Library of Medicine

4. Grand Challenge and Supporting Technology Case Studies 17

Forecasting Severe Weather EventsCancer GenesPredicting New SuperconductorsAir PollutionAerospace Vehicle DesignEnergy Conservation and 71.irbulent CombustionMicrosystems Design and PackagingEarth's BiosphereHigh Speed NetworksEducation Using the NREN

Glossary 56

List of Figures

Page

Figure 1. The High Performance Computing and 3

Communications Program

Figure 2. Pertormance Requirements for Grand Challenge 7

Problems

Figure 3. Computer System Performance Trends for 13

Grand Challenge Problems

Figure 4. Performance Improvement 15

Figure 5. NREN Applications 18

Figure 6. Evaluation Criteria for the HPCC Program 23

Figure 7. HPCC Budgets by Agency ,Ind Program Component 24

Figure 8. Agency Responsibilities 26

Grand Challenges:High Performance Computing

and Communications

The FY 1992 U.S. 1Zesearch and Development Program

fiNMEMIIF +01111111111, .A=MMMNIMMI.Mffa

Executive Summary

EXECUTIVE SUMMARY

High performance computing and computer communications networksare becoming increasingly important to scientific advancement,economic competition, and national security. The technology is reachingthe point of having a transforming effect on our society, industries, andeducational institutions. The goal of the Federal High PerformanceComputing and Communications (HPCC) Program is to acceleratesignificantly the commercial availability and utilization of the nextgeneration of high performance computers and networks in a mannerconsistent with the Strategic and Integrating Priorities shown in Figure 1.

The HPCC Program is the result of several years of ef fort on the partof senior government, industry, and academic scientists and managersto design a research agenda to extend U.S. leadership in highperformance computing and networking technologies.

For FY 1992 the HPCC Program proposes to invest $638 million inthe four complementary and coordinated components shown inFigure 1. This investment represents a $149 million, or 30%, increaseover the FY 1991 enacted level.

The HPCC Program is driven by the recognition that unprecedentedcomputational power and capability is needed to investigate andunderstand a wide range of scientific and engineering "grandchallenge" problems. These are fundamental problems whose solutionis critical to national needs. Progress toward solution of these problemsis essential to fulfilling many of the missions of the participatingagencies. Examples of grand challenges addressed include: predictionof weather, climate, and global change; determination of moiecular,atomic, and nuclear structure; understanding turbulence, pollutiondispersion, and combustion systems; mapping the human genome andunderstanding the structure of biological macromolecules; improvingresearch and education communications; understanding the nature ofnew materials; and problems applicable to national security needs.

The HPCC Program nurtures the educational process at all levels byimproving academic research and teaching capabilities. Advancedcomputing and computer communications technologies will acceleratethe research process in all disciplines and enable educators to integratenew knowledge and methodologies directly into course curricula.Students at all levels will be drawn into learning and participating in awide variety of research experiences in all components of this program.

The FY 1992 Program and this document were developed by theHPCC Working Group under the direction of the Committee onPhysical, Mathematical, and Engineering Sciences of the FederalCoordinating Council for Science. Engineering, and Technology.

Figure 1 The High Performance Computingand Communications Program

Coals: Strategic PrioritiesExtend U.S. technological leadership in high performance computing andcomputer communications.

Provide wide dissemination and application of the technologies both to speedthe pace of innovation and to serve the national economy, national security,education, and the global environment.

Spur gains in U.S. productivity and industrial competitiveness by making highperformance computing and networking technologies an integral part of thedesign and production process.

1.1

Strategy: Integrating PrioritiesSupport solutions to important scientific and technical challenges through avigorous R&D effort.

Reduce the uncertainties to industry for R&D and use of this technologythrough increased cooperation between government, industry, and universitiesand by the continued use of government and governmentfunded facilities as aprototype user for early commercial HPCC products.

Support the underlying research, network, and computationa infrastructures onwhich U.S. high performance computing technology is based.

Support the U.S. human resource base to meet the needs of industry,universities, and government.

.7511M11

Program ComponentsHigh Performance Computing Systems

Research for Future Generations of Compiling SystemsSystem Design ToolsAdvanced Prototype SystemsEvaluation of Early Systems

Advanced Software Technology and AlgorithmsSoftware Support for Grand ChallengesSoftware Components and ToolsComputational TechniquesHigh Performance Computing Research Centers

National Research and Education NetworkInteragency Interim NRENGigabits Research and Development

Basic Research and Human ResourcesBasic ResearchRescarch Participation and TrainingInfrastructureEducation, Training, and Curriculum

2

I. PROGRAM GOALS AND OVERVIEW

Introduction

High performance computing (HrC) is emerging as a powerfultechnology for industrial design and manufacturing, scientific research,communications, and information management. A robust U.S. highperformance computing and computer communications capabilitycontributes to leadership in critical technology and national securityareas. Improved computational and communications technologiescontribute to more effective approaches to probbnn solving, new

oducts and services, and enhanced national competitiveness acrossbroad sectors of the economy.

Recent advances offer the potential for a thousandfold improvement inuseful computing capability and a hundred4old improvement inavailable computer communications capability by 1996. Based onseveral years of planning, under the auspices of the FederalCoordinating Council for Science, Engineering, and Technology(FCCSET), Federal agencies and the technical community havedeveloped the Federal High Performance Computing andCommunications (HPCC) Program to realize this potential and to meetthe challenges of advancing computing and associated communicationstechnology and practices. Agencies have realigned and enhanced theirhigh performance computing research and development programs,coordinated their activities with other agencies, and shared commonresources to develop the program presented in this document.

Needs and Benefits

High periormance computing has become a vital enabling force in theconduct of science and engineering research over the past three decades.Computational science and engineering has joined, and in some areasdisptaced, the traditional methods of theory and experiment. Forexample, in the design of commercial aircraft, many engineering issues

are resolved through computer simulation rather than through costlywind tunnel experiments. This trend has been powered by computinghardware and software, computational methodologies and algorithms,availability and access to high performance computing systems andinfrastructure, and the growth of a trained pool of scientists andengineers. This process has been nurtured by Federal investment inadvanced research, agency supercomputer centers, and nationalnetworks through DARPA, DOE, NASA, NSA, and NSF. Thesefacilities have contributed to national mission areas such as energy,

space, health, defense, environment, weather, and basic science and

-513

technology that could not be effectively addressed without the use ofsuch advanced facilities.

High performance computing technology is knowledge and innovationintenswe. Its development and use engages the entire scientific andengineering community. Building upon fundamental research of theearly 1980's, a new computing technology of scalable parallelprocessing computers emerged. By the mid-1990's, this innovativeapproach to high performance computing systems promises to achievesustained performance improvements of a thousandfold compared tocurrent systems.

In a growing number of science and technology fields, progress andproductivity in modem research are increasingly depenoent on the closeinteraction of people located in distant places, sharing and accessingcomputational resources across networks. Although the U.S. is theworld leader in most of the critical aspects of computing technology, thislead is being challenged.

The Federal HPCC Program is a strategic Federal investment in thefrontiers of computing and computer communications technologies andis formulated to satisfy national needs from a variety of perspectivesincluding: technology, science applications, human resources, andtechnology transition. Needs are derived from the agency missions andbased on the underlying science, engineering, and technology baserequired to carry out these missions. Many of these mission needs arerelated to solving very intensive large scale computing problems. Thesefundamental problems often cut across various agencies and missionsand are called grand challenge problems (Figure 2).

The industrial and academic sectors provide major sources ofinnovation, cost effective development, and support of informationtechnologies and their application to grand challenge problems. Asthese technologies are developed, the results support the Federal agencymissions and become available nationally. The program provides fordevelopment of these revolutionary technologies within a framework ofa partnership among government, industry, and academe and allows forrapid transition of laboratory results into new products that will then beapplied within the program.

Figure 2 Performance Requirementsfor Grand Challenge Problems

Computer Performancein Billions of Operationsper Second

1000

100

10

0.1

MUM.

Mal=

Grand Challen esClimate ModelingFluid Turbulence

Pollution DispersionHuman Genome

Ocean CirculationQuantum ChromodynamicsSemiGonductor ModelingSuperconductor Modeling

Combustion SystemsVkiion and Cognition

StructuralBiology

Vehicle Signature PharmaceuticalDesign

48 HourWeather

ChemicalDynamics

IMMO

Airfoil Design

2D PlasmaModeling

ULSI DesignA1172 HourWeather

Estimate of HiggsBoson Mass

3D PlasmaModeling

Speech andNatural Languagp

1980 1990 2000

Program Description

The Program consists of four integrated components representing thekey areas of high performance computing and communications:

High Peiformance Computing Systems (HPCS) the developmentof the underlying technology required for scalable parallelcomputing systems capable of sustaining trillions of operations persecond on large problems.

Advanced Software Technology and Algorithms (ASTA) thedevelopment of generic software technology and algorithms forgrand challenge research applications to realize the performancepotential of high performance computing systems in a networkedenvironment.

National Research and Education Network (NREN) thedevelopment of a national high speed network to providedistributed computing capability to research and educationalinstitutions and to further advanced research on very high speednetworks and applications.

Basic Research and Human Resources (BRHR) support forindividual investigator and multidisciplinary long term researchdrawn from diverse disciplines, including computer science,computer engineering, and computational science and engineering;initiation of activities to significantly increase the pool of trainedpersonnel; and support for efforts leading to accelerated technologytransition.

Advances in high performance computing enable advances in almostevery other science and engineering discipline. There is a complex webof research interdependencies among the four components, and eacharea contributes to progress in other areas. Because of thesedependencies, achieving and maintaining balance between the researchcomponents is a primary goal and the most important priority in thecurrent context and environment. The HPCC Program is designed toprovide balanced support both for technology areas includingcomponents, systems, software, and algorithms, and for applications,infrastructure, and human resources to achieve rapid overall researchprogress and productivity.

The component activities are planned to produce a succession ofintermediate benefits on the way to meeting the long rangeprogrammatic goals. The HPCC Program builds on Federal programsalready in place, providing additional resources in selected areas.Computational science and engineering grand challenges as illustratedin Figure 2 are the focal points for these efforts.

8,

Goals

The goais of the High Performance Computing and CommunicationsProgram are to:

Extend U.S. technological leadership in high performancecomputing and computer communications.

Provide wide dissemination and application of the technologiesboth to speed the pace of innovation and to serve the nationaleconomy, national security, education, and the global environment.

Spur gains in U.S. productivity and industrial competitiveness bymaking high performance computing and networking technologiesan integral part of the design and production process.

These goals will be realized by achieving: computational performanceof one trillion operations per second (1012 ops, or teraops) on a widerange of important applications; development of associated systemsoftware, tools, and improved algorithms for a wide range of problems;a national research network capable of one billion bits per second(109 bits, or gigabits); sufficient production of Ph.D.'s and other trainedprofessionals per year in computational science and engineering toenable effective use and applicAion of these new technologies.

Strategy

The goals will be met through coordinated government, industry, anduniversity collaboration to:

Support solutions to important scientific and technical challengesthrough a vigorous R&D effort.

Reduce the uncertainties to industry for R&D and use of thistechnology through increased cooperation between government,industry, and universities and by the continued use of governmentand governmentfunded facilities as a prototype user for early

commercial HPCC products.

Support the underlying research, network, and computationalinfrastructures on which U.S. high performance computingtechnology is based.

Support the U.S. human resource base to meet the needs ofindustry, universities, and government.

At the program component level the strategy will: exploit and extendscalable parallelism and engage in intensive software development inHPCS; use common requirements of the grand challenges to foster HPC

software progress in ASTA to strengthen HPC software developmentand coordination; evolve from the current Internet network to the NRENusing a series of testbed systems; and strengthen academic activities incomputer science and computational science and engineering as part ofBRHR,

Program Execution Strategy

The strategy is based on the strengths of partnerships among the Federalagencies and other organizations. Major portions of the program will becostshared arid leveraged by the participation of industry anduniversities. This general approach operates well today and providesstrong evidence that the HPCC Program can be successful in the future.The specific elements of the approach are to:

Create a balanced, criticalmass program. The program mustachieve sufficient scope and talance among the components. Atechnology program that created extremely fast processors withoutcomparable memory, inputoutput, and mass storage systemswould not succeed. Neither would a program that created powerfulcomputers without adequate software, network access, and capablepeople. Similarly, a program that created only high performancenetworks would not satisfy the increased performance requirementsneeded for grand challenges, The HPCC Program must operate at asufficient scale and coverage of technology areas that the newtechnologies can be effectively applied to grand challengeproblems with acceptable levels of risk. The HPCC Programachieves balance by the extensive participation of experiencedusers, applications developers, and researchers in the HPCCdisciplines throughout the design, development, andimplementation process.

Build on agency strengths. The strategy builds on agency strengthsby giving appropriate agencies the responsibility to coordinateactivities for areas of demonstrated capability. It also ensures thatthe strengths of the other agencies are included by integrating theirparticipation in various task areas. DOE, NASA, NSF, DOC(NOAA, NIST), and DOD (DARPA, NSA) have decades ofexperience in applying the world's most powerful computers, andthus provide a valuable perspective on high performancecomputing requirements and applications. NSF has thedemonstrated technical arid operational expeTtise needed fordeploying high performance national networks in the researchcommunity, and is uniquely positioned to support basic research in

computer science, computational science, and other scientific areasthat can benefit from high performance computing in

interdisciplinary programs. DARPA, having pioneeredtechnologies for high performance computing systems andmicroelectronic components, has a strong existing researchprogram in teraops computing and gigabit networking, and offersrapid technology transition into commercial products in support ofDOD science, technology, and applications base. EPA, NIH, andNOAA provide complementary HPCC requirements perspectivesand application bases. NIST has extensive experience in HPCsystems and networks instrumentation arid evaluation, arid providesa means for standards development.

Accelerate Technology Transfer. The transfer of technology fromresearch to development and to application can be a very slowprocess due to a number ofbarriers to the use of new technology:high initial cost, inadequate and userunfriendly software andsystems, and lack of standards. The HPCC Program relies uponsubstantial industry participation to overcome these barriers andyield the benefits derived from moving new technologies toindustry. The strategy accelerates technology transition by using aparticipative development process for each of the task areas, and bystimulating the growth of shared knowledge and capabilities. Anexample of this is the creation of networkaccessible repositories ofscalable HPC software and associated user groups to provide usagefeedback and improvement of the HPC software base.

Overcome Barriers. To overcome high costs of creating successivegenerations of high performance computers, the program will

emphasize scalable computer designs. Scalable computing andnetworking technologies enable exploratory use with small, 1 _ sver

cost prototype systems needed to eventually support the acquisitionof larger systems. The development of userfriendly software andsystems as part of the investment in the HPC software base is amajor, integral part of the itrogram. The central role of missionagencies, and the broad academic scientific research communityinvolvement, ensures that the hardware, software, and networkingtechnology developed will be responsive to user needs. A strategyof cooperative government, industry, and university activities isused to manage and coordinate the coupling of these sectors toachieve maximum synergy.

2. HIGH PERFORMANCE COMPUTING AND

COMMUNICATIONS PROGRAM COMPONENTS

Tho HPCC Program is composed of four integrated and coordinatedcomponents that are designed to enhance scientific productivity andsupport long term agency needs. The emerging scientific computingenvironment is that of advanecl workstations with high resolution colordisplays connected to a high speed computer communications networkand high performance computing resources. The regional and nationalnetworks provide a means to gain access to additional high performancecomputing systems and research resources. Realizing the full potentialof these temps computing and gigabits networking systems will requireadvanced software technology and algorithms and people educated andtrained to use these tools and resources in this dynamic field.

High Performance Computing Systems (HPCS)

The HPCS component produces scalable parallel computing systemsthrough the development of prototype systems. The program isdesigned to attack computational science problems by developinginnovative systems that will provide a onehundred to onethousand--fold increase of sustained computing capability over machines thatfollow the more conventional design evolution path (See Figure 3).DARPA will coordinate the research and development effort that willproduce teraops systems.

The program is structured to focus on technological challenges in theearliest stages of the product development cycle. Critical underlyingtechnologies are developed in prototype form along with associateddesign tools. This allows empirical evaluation of alternative solutiens asthe prototype systems mature. Evaluation is performed throughout thedevelopment cycle, with experimental results being fed back into thedesign process to refine successive generations of systems. There is riskinherent in creating new technologies, and each project will be managedaccording to its proximity to commercial introduction. Larger projectswhich are close to yielding commercial products are performed on a costshared basis with industry.

HPCS is composed of the four subcomponents shown below.

Research for Future Generations of Computing Systems. This activityproduces the underlying component, packaging, and scalingconcepts. These projects ensure that the required advancedtechnologies will be available for the new systems while providing afoundation for the more powerful systems that need to follow.

12 2 0

Figure 3 Computer System Performance Trendsfor Grand Challenge Problems

Power and Time to System Development

Computer Performancein Billions of Operationsper Second

moo If-

100

10

0.1

Grand Challenges

-71===1980 1990 2000

1 Teraops = 1000 Billion Operations per Second

System Design Tools. This activity develops computer aided designtools and frameworks for enabling multiple tools to work together toenable design, analysis, simulation, and testing of systemscomponents. The tools will enable rapid prototyping of new systemconcepts.

Advanced Prototype Systems. This activity consists of focuseddevelopment of experimental systems that are designed anddeveloped on a cost shared basis with industcy. The 100 gigaopssystems provide a basis for the teraops systems. New models ofcomputation will be introduced and successive generations ofcomputer systems, along with systems software, will encompassbroader grand challenge domains. Modular technologies will enablea wide variety of system configurations using common components.Systems with 100 gigaops sustained performance will be developedby the early 1990's. The teraops level will be reached by the mid1990's.

Evaluation of Early Systems. Experimental systems will be placed atsites with high levels of expertise to provide feedback to systemsarchitects and software designers. Performance evaluation criteria forsystems and results of evaluation will be made widely available.Because of scalability, early systems can be acquired at smallerscales to evaluate their potential performance. As noted below, theASTA component will support, on a cost shared basis, the acquisitionof large scale systems for experimental use in grand challengeapplications.

Advanced Software Technology and Algorithms (ASTA)

Dramatic improvements in algorithm design and software technologyare essential to achieving sustained teraflops computing systemperformance. Improvements in hardware, especially for scalableparallel systems, must be matched with new and innovative algorithmsand software to enable researchers to expand the boundaries ofcomputational capabilities. In Figure 4, this point is illustrated for broadclasses of scientific computing problems by showing that comparableimprovements in throughput have resulted from advances incomputational methods as from improved hardware technology. In thiscase, hardware improvements yielded a speedup of 1000 over 20 years,while software and algorithms improvements yielded a speedup of3000 over the same period, for an aggregate speedup of more than amillion.

Figure 4

Speed-UpFactor

105

104

102

101

10°

1970

Performance Improvementfor Scientific Computing Problems

Derived from Computational Methods

Speed-UpFactor

05

104

Multi-Grid

Conjugate GradientSuccessive Over-Relaxation

Gauss-Seidel

Sparse Gaussian Elimination

1980 1990

Derived from Supercomputer Hardware

103 Vector Supercomputer

102

10'

10°

1970

2000

1980 1990 2000

The ASTA component of the Federal High Performance C ,.nputing andCommunications Program has three goals:

Enable solution of grand :.hallenge application problems in scienceand engineering.

Improve system userfriendliness, reliability, and softwareproductivity.

Use experience gained on leading edge applications to help guidefuture software efforts.

15

The emphasis is on the development of advanced algorithms andsoftware technology required to address applications problems on thescale of grand challenges. The ASTA component is comprised of foursubcomponents, as shown below.

Software Support for Grand Challenges. The HPCC Program isdesigned to demonstrate new computing technology capabilities byconfronting an expanding number of grand challenge applicationproblems. The goal is to reduce the risks to researchers inherent inadopting innovative high performance computing technologies.Grand challenge application problems will be selected based upontheir scientific importance, the potential for cost sharing with sourcesdirectly concerned with the specific scientific and engineeringapplications, and the potential for leveraging across sectors.

Software Components and Tools. Tile multidisciplinary research teamswill have common needs in softy 'are technology and programmingenvironments, advanced compU chnology, optimization andparallelization tools, interoperabnity and data management,visualization, debugging and analysis, and performancemeasurement. Advances in these generic software areas willminimize the need for specialized researchers to master advanced,complex computing science skills. Coordination of the deVelopmentof advanced software technology and algorithms among theparticipating agencies will be important to ensure effective andefficient use of resources. A particular focus of this element will beto develop advanced software applications that exploit the NRENusing distributed file systems and national software libraries.

Computational Techniques. The focus of the HPCC Program on scalableparallel computing systems dictates that significant advances incomputational techniques will be needed. The design and theory ofalgorithms are as important as hardware or networking improvementsin reaching teraflops computational performance. Research incomputational techniques will include parallel algorithms, numericaland mathematical analyses, parallel languages, computational models,and program refinement techniques. Higher level languages will bedeveloped to enable computational scientists to work directly at thelevel of the abstract computational problem being addressed and tomore easily explore specific implementation approaches.

High Perfbrmance Computing Research Centers. High performancecomputing research centers and testbed facilities will provide a largenumber of researchers access via the NREN to both conventional andinnovative supercomputing architectures. Many of these centers willintroduce innovative architectures and will be the focus oftechnology transfer activities and form the base for training newgenerations of computer and computational scientists.

National Research and Education Network (NREN)

The NREN component of the HPCC Program dramatically expands andenhances the U.S. portion of an existing worldwide infrastructure ofinterconnected computer networks called the Internet. A substantialfraction of the domestic Internet is supported and loosely coordinated byFederal agencies, principally DARPA, DOE, NASA, and NSF.

Collaboration among scientists is an important and integral facet of theU.S. research environment. It can be greatly improved by increasing thelevel of network connectivity and by introducing new capabilities intothe existing infrastructure. The NREN design will not only addressbroad network connectivity, but will also provide the basis for necessaryhigher level capabilities and services.

Many educational institutions, government laboratories, and industrialresearch facilities are currently connected to the Internet. Yet, it stillfalls short of a widespread, uniform, and high performance nationalinfrastructure. In order to satisfy the HPCC Program goals, the NRENmust not merely provide network access to research and educationalinstitutions at all levels and locations, it must also deliver newcapabilities. Some of these, such as distance learning, may initially beextensions of current technology. All capabilities will benefit from, andmany will be enabled by, a program of research into very high speedtechnology. This technology is needed to support access to digitallibraries, large scale distributed computing resources, as well as toperform computationally intensive applications that require real timevisualization of modeling and simulation results, rapid interrogation andretrieval of scientific data from specialized data bases, remote control ofexperiments and simulations, and teleconferencing.

In addition to serving the needs of the scientific and researchcommunities, the NREN will provide valuable experience necessary forthe successful development of a broader, privatelyoperated nationalinformation infrastructure. Such an infrastructure would allowconsumers, businesses, and schools and government at all levels to sharequality information and entertainment when and where they want it at a

reasonable cost.

Applications conducted over a computer network vary in their flow ofinformation from steady, as in a bulk file transfer, to "bursty," as inhumanomputer interaction via keyboard. Similarly, some applicationscan be carried out at relatively low communication rates, while others bytheir nature require high speeds. A number of scientific networkingapplications are characterized on the graph of speed versus burliness inFigure 5. Traffic seen in the early days of networks appears near thebottom of the chart. More advanced applications are furthest from the

origin and require more sophisticated protocols and networkcapabilities. This chart illustrates that a gigabit network is needed notonly to carry the aggregation of low speed traffic, but also toaccommodate high speed uses.

Figure 5

BandwidthPeak Rate

1010

109

109

107

106

106

ce

103

102

101

100

NREN Applicationsby Bandwidth and Traffic Characteristics

CompositeImaging

InteractiveVisualization

VideoTeleconferenc

TextFile

Transfer

CollaborationTechnology

DistributedComputing

ImageTransfer

MultiMediaDatabaseAccess

MultiMediaMail

ElectronicMail

CharacterData

Transfer

Steady BurstyTrafficRequirementsfor Bandwidth

The vision of the NREN is of an interconnection of the nation'seducational infrastructure to its knowledge and information centers. Inthis system, elementary schools, high schools, two and four yearcolleges, and universities will be linked with research centers and

18

laboratories so that all may share access to: libraries, databases, anddiverse scientific instruments such as supercornputers, telescopes, andparticle accelerators. The NREN enables communication and fosterscollaboration among and within these communities. By reducing thetraditional impediments of geographical isotation, the NREN improvesthe quality and raises the level of education nationwide. The NRENcontributes to the success of the Basic Research and Human Resourcescomponent of the High Performance Computing and CommunicationsProgram. By making u ique scientific and informational resourcesaccessible beyond their physical locations, it permits widespreadparticipation in the HPCC Program by scientists, university researchers,and students, and it enables tne development of large scale distributedcomputing resources.

Interagency Interim NREN. The NSF will coordinate the Interim NRENactivities by upgrading its backbone network, by assisting regionalnetworks tn upgrade facilities, capacity, and bandwidth, and byinterconrhating the backbone networks of other agencies. Asignificant effort in its implementation will be the development anddeployment of safeguards to enhance security and control overaccess to the connected computer systems, and that of the networkcomponents themselves in order to minimize vulnerability to carelessor malicious attack. Coordination among participating agencies andthe non Federal networking community will be expanded through thecreation of a National Networking Council.

Gigabit Research and Development. DARPA will coordinate theresearch and development effort that will culminate in initialdeployment of gigabits per second capability. Coordination ofresearch efforts on very high speed switches, protocols, andcomputer interfaces will be necessary.

Basic Research and Human Resources (BRHR)

This component addresses long term national needs for more skilledpersonnel, enhancement of education and training, and materials andcurriculum development in the high performance computing science andengineering areas. The NREN and ASTA components include supportfor research in the large scale project environment. The BRHRcomponent is designed to encourage investigator initiated, long termresearch on experimental projects that will maintain the flow ofinnovative ideas and talented people into high performance computingareas.

Drawing the best and brightest of our Nation's youth into scientific t ndtechnological careers is a formidable challenge that will have profoundeffects on overall U.S. scientific and technological competitiveness.This component of the program will establish industry, university, andgovernment partnerships to improve the training and utilization ofpersonnel and to expand the base of research and developmentpersonnel in high performance computing science, technology, andapplications.

The BRHR component of the HPCC Program has five goals:

Improve the flow of human resources into high performancecomputing.

Improve the university infrastructure to stay at the leading edge.

Expand collaboration and resource sharing among the Federal,academic, and industrial sectors.

Facilitate multidisciplinary research on high performancecomputing and communications.

Create a critical mass of users by building a base with commonsystems, tools, and interfaces.

The BRHR component is organized across the four subcomponentsshown below.

Basic Research. These acti dies support increased participation ofindividual investigators in the conduct of innovativemultidisciplinary research in computer science, computerengineering, and computational science and engineering related tohigh performance computing. The strategy is to increase the numberof multidisciplinary awards across all disciplines wherecomputational methods are critical to achieving advances orscientific breakthrcugils. Program activities include: research onscientific algorithm development for highly parallel computers;

20 2

generic computational algorithm development; scaling techniques;

prediction techniques for concurrent systems; resource managementstrategies for highly parallel distributed systems; fault tolerantstrategies for parallel and distributed systems; and research onheterogeneous software environments.

Research Participation and Training. These activities address thehuman resources pipeline in the computer and computationalsciences, at postdoctoral (training and retraining), graduate, and

undergraduate levels. Program activities include: workshops and

seminars; postdoctoral fellowships in computational science and

engineering; career training in medical informatics through grants to

young investigators; institutional training and postdoctoralprograms; knowledge transfer exchange programs at nationallaboratories, centers, universities, and industry; and softwaredissemination through national databases and libraries.

Infrastructure. These activities will improve university and governmentfacilities for computer science, computer engineering, andcomputational science and engineering research related to high

performance computing. Program activities include: improvement of

equipment in computer science, computer engineering, andcomputational science and engineering academic departments,centers and institutions; development of scientific databases; and

distribution of integrated system building kits and toolsets.

Education, Training and Curriculi4m. These activities will expand and

initiate activities to improve undergraduate and precollegeeducation and training opportunitks in high performance computing

and computational science and engineering for both students and

educators. The introduction of associated curriculum and training

materials at all levels is an integral part of this effort. Program

activities include: bringing people to national centers for training,

technology transfer, and educational experiences; using professional

scientists and engineers to provide curriculum development materials

and instruction for high school students in the context of high school

supercomputer programs, supercomputer user workshops, summer

institutes, and career development informatics for health sciences.

3. PROGRAM DEVELOPMENT AND AGENCYBUDGETS

Program Planning

Leadership for the HPCC Program is provided by the Office of Scienceand Technology Policy (OSTP), through the Federal CoordinatingCouncil for ScienLe, Engineering, and Technology (FCCSET) Committeeon Physical. Mathematical, and Engineering Sciences (PMES).The membership of the PMES includes senior executives of manyFederal agencies.

The planning process for the HPCC Program was coordinated by aPMES working group tnrough information exchange, the commondevelopment of interagency initiatives, and the review of individualagency HPCC proposals and budgets.

Evaluation Criteria

Each participating agency HPCC contribution was reviewed againstformal evaluation criteria during the planning and budget process. Areview of participating agencies was performed using these evaluationcriteria to develop the FY 1992 agency requests for the Program. Theevaluation criteria are shown in Figure 6.

Agency Budgets

Over the last three years of the planning process of this initiative, theparticipating agencies have mutually adjusted their activities within thebase to achieve greater efficiency in addressing the goals of the HPCCProgram. For FY 1992, $638 million is being proposed, a $149 millionor 30 percent increase over the FY 1991 enacted level. The budget isshown in Figure 7.

The funds proposed from Federal sources are not intended to carry outthe entire HPCC Program. Portions of this program will be cost sharedby organizations from the participating sectors. The funding estimatesare based on analyses of the practical experience of prior computing andcomputer networking programs. These estimates were then reviewed asoutlined above. Cost sharing will occur with various U.S. industrial anduniversity partners to a large extent in the HPCS and the NRENcomponents. Cost sharing will occur in the ASTA component withagency programs and other computational applications programs, forexample, in specific grand challenge areas, via multidisciplinarycollaborations. This component also includes deployment of highperformance systems to HPCC research centers. The close coupling of

support in the Program will result in significant leverage, acceleratetechnology transfer, stimulate U.S. industry and markets, and enable thesolution of computationally intensive applications. In addition, althoughHPCC is not intended to include classified programs, the technologyproduced will have an important impact in these natielal security areas.Figure 7 illustrates the relative levels of investment in the four programcomponents.

Figure 6 Evaluation Criteriafor the

Federal High Performance Computingand Communications Program

RelevancelContribution. The research must significantly contrthute to the

overall goals and strategy of the Federal High Performat.,:e Computing andCommunications (HPCC) Program, including computing, software,networking, and basic research, to enable solution of the grand challenges.

Technicalacientific Merit. The proposed agency program must be

technically/scientifically sound and of high quality, and must be the product

of a documented technical/scientific planning and review process.

Readiness. A clear agency planning process must be evident, and the

organization must have demonstrated capability to ca ry out the program.

Timeliness. The proposed work must be technically/scientifically timely for

one or more of the HPCC program components.

Linkages. The responsible organization must have established policies,programs, and activities promoting effective technical and scientificconnections among government, industry, and academic sectors.

Costs. The identified resources must be adequate, represent an appropriate

share of the total available HPCC resources (e.g., a balance among programcomponents), promote prospects for joint funding, and address longterm

resource implications.

Enhancements to Existing Program Research. Existing agency HPCCprograms will receive adequate support before new initiatives are funded.

Agency Approval. The proposed program or activity must havepolicylevel approval by the submitting agency.

Figure 7High Performance Computing and Communications

Budgets by Agency and.Program Component(Dollars in Millions)

Agency

BaseFY 1991

HPCCFY 1992

FY 1992 HPCC Component ,

HPCS ASTA NREN BRHR

DARPA 183.0 232.2 103.3 38.5 32.9 57.5

DOE 65.0 93.0 15.0 58.0 12.0 8.0

NASA 54.0 72.4 14.2 49.8 7.4 1.0

NSF 169.0 213.0 24.0 103.0 32.7 53.3

DOC/NIST 2.1 2.9 0.3 0.6 2.0 0.0

DOC/NOAA 1.4 2.5 0.0 1.8 0.7 0.0

EPA 1.4 5.2 0.0 4.5 0.0 0.7

NIII/NLM 13.5 17.1 0.0 8.9 4.2 4.0

Total 489.4 638.3 156.8 265.1 91.9 124.5

FY 1992ComponentFundingComparison

HPCC High Performance Computing and Communications

Components of HPCC:HPCS - High Performance Computing SystemsASTA - Advanced Software Technology and AlgorithmsNREN National Research and Education NetworkBRHR - Basic Research and Human Resources

Agency Program Descriptions

The agency responsibilities under the Federal HPCC Program areoutlined in Figure 8. Several agencies have been assigned acoordinating respon:iibily in specific technical areas:

DOE and NASA will coordinate activities in HPC system evaluation,testbed development, and applications software capabilities.

NASA will coordinate the accumulation of and access to the HPCsoftware base. This will be facilitated by a wide area file systemtechnology that is currently being deployed for early experimental useby DARPA which will be extended to include the NREN as it maturesand is deployed by NSF

DARPA will coordinate activities in the development of scalableparallel HPC systems, including their basic units of replication,system modules, and the necessary associated systems software.DARPA will also coordinate activities in gigabit network technologyresearch.

NSF will coordinate activities for the broad deployment of theNational Research and Education Network working with all agencieswith mission specific requirements.

NSF will coordinate activities in basic research and human resourcedevelopment, while each agency retains its role as required toaccomplish thei: own missions.

NIST will coordinate activities in HPC system instrumentation,evaluation, and in standards issues.

Each agency participates in all of the identified activities to ensum thai,

the resulting capabilities are a good match to user needs. DOCNOAA,HHS/NIH/NLM, and EPA bring distinctive applications areas of broad

interest and network user bases.

Figure 8

HPCC Program: Agency Responsibilities: HPCS, ASTAActivity

AgencyHigh PerformanceComputing Systems

Advanced SoftwareTechnology & Algorithms

DARPA

Technology developmentand coordination forTeraops systems

Technology developmentfor parallel algorithmsand systems software

.

DOE

Technology developmentSystems evaluation

Energy applications centersEnergy computation researchSoftware tools

NASA

Aeronautics and spaceapplication testbeds

. Systems development forspace flight

Software coordinationResearch in:

Aerospace computationsInformation management

NSF

Basic architecture research Basic research in:Software tools, databases

. Grand challengesComputer access

DOC/NIST

DOC/NOAA

.

Ocean and atmosphericcomputation research

Software tools

EPA

Research in environmentalcomputatiors, databases,and application testbeds

.....NM/NLM

Medical application testbeds. Software tools. Medical computation

research

HPCC Program: Agency Responsibilities: NREN, BRHR

National Research andEducation Network

Basic Research andHuman Resources

Activity

Agency

Technology developmentand coordination for

, gigabits networks

University programs

DARPA

Energy deployment missionfacilitius

Gigabits applicationsresearch

. Basic research andeducation programs

DOE

Aerospace deploymentmission facilities

. Research initiation anduniversity block grants

NASA

Facilities coordination anddeployment

Gigabits research

. Programs in:. Basic research. Education. Training / curricula

NSF

High speed networkresearch and standards DOC/

NIST

Ocean and atmosphericmission facilities DOC/

NOAA

States environmentalmission assimilation

. Technology transfer toStates

University programs EPA

. .

Medical mission facilities University programsNIH/NLM

Defense Advanced Research Projects Agency (DARPA'l

DARPA, as the DOD lead agency for advanced technology research,will focus on developing the high performance computing andnetworking technologies needed for the Defense and overall HPCCPrograms. DARPA programs have produced both the computing andnetworking foundation for the HPCC Program, including the firstgeneration of scalable parallel computing systems and large scalecomputer networks, and the associated system software and supportingtechnologies. DARPA has worked with industry to pioneer theapplication of these new technologies within Defense and on acooperative basis with other Federal agencies.

The DARPA HPCC Program builds upon the DARPA StrategicComputing Program. As the HPCC Program builds up, the StrategcComputing Program integrates its results with Defense specific needssuch as embedded systems, accelerators of specific problem domains,and grand challenges probLms related to defense. DARPA will continuethis mode of executing the HPCC Program, cooperating with variousdefense organizations, and working closely with AFOSR, ARO, ONR,Defense Service Laboratories, NS.A, other Defense organizations, andother Federal agencies as appropriate.

High Performance Computing Systems will be produced in the fourmain subareas: Research for Future Generations of Computing Systems,System Design Tools, Advanced Prototype Systems, and Evaluation ofEarly Systems. Systems capable of sustaining 100 gigaops for largeproblems will available for deployment by late 1993 and the teraopssystems will be available by 1996.

In Advanced Software Technology and Algorithms, DARPA projectswill produce scalable libraries for Defense problem domains andprogramming and analysis tools for scalable parallel and distributedheteropneous systems in a workstation/server configuration that will beopen to the integration of embedded systems and accelerators.

For the National Research and Education Network component, highperformance networking technologies will be produced to satisfy thegigabit technology needs of the NREN and to provide a dual usetechnology base for Defense. This networking technology includesdevelopment of new protocols and switch and transmissiontechnologies, and it will be capable of supporting a wide range ofadvanced network services.

The Basic Research and Human Resources component will focus Onfundamental scientific issues in these three technology developmentareas in cooperation with other basic research programs and provide forsmaller individual investigator projects to complement the largerprojects. In addition, the relevant and related basic research will also beintegrated into the largek.i,:ojects as it matures.

Department of Energy (DOE)

The Department of Energy will participate in all components of theHigh Performance Computing and Communications Program. In thearea of High Performance Computing Systems, the DOE will be an earlycustomer of small versions of systems with advanced architectures andwill evaluate these systems on energy related applications. The DOEwill consider cooperative development of advanced systems between itsnational laboratories and vendors, especially integration of very highspeed computer and networking hardware with software systems. TheDOE will support research and development on algorithms and systemssoftware for parallel computing systems.

The Advanced Software Technology and Algorithms effort will includeresearch and development of: parallel algorithms for grand challengeapplications, software and tools for early prototypes of 100 gigaflopsand teraflops systems, prototype computational science programmingenvironments that meet standards and are transportable, and support forhigh performance computing research centers to facilitate the transitionfrom research on parallel machines into the applications and theprogramming environments. The DOE will fund several grandchallenge collaborations, initiate a software component and toolsprogram with strong industrial participation, and initiate an applicationsdriven computational research program. The DOE will evaluateproposals and make research awards related to grand challenges inglobal climate change, molecular biology, human genome research,materials and chemical sciences, combustion research, wasteremediation, fusion energy, and other areas within its mission.

The DOE will participate in the cooperative interagency NationalResearch and Education Network. The Energy Sciences Net (ESNet)will be incorporated into the NREN and will provide quality networkaccess to the energy research facilities. ESNet will maintaincompatibility and will be upgraded in concert with NREN. Gigabitnetwork support technology will be developed for DOE applicationsdistributed across multiple energy research centers at the nationallaboratories and universities.

The DOE's Basic Research and Human Resources activities willinclude: stimulating research in computational science, expandingtraining programs at the national laboratories for high school teachersand college students in computing techniques, initiation of a high schoolsupercomputer access program, and provision of fellowships incomputational science with internship at national laboratories.

National Aeronautics and Space Administration (NASA)

NASA's HPCC Program participates in all four components of theFederal Program, through a vertically integrated program focused onNASA's grand challenges in: Computational Aerosciences, Earth andSpace Sciences, and Remote Exploration Lid Experimentation.

The goal of NASA's program is to accelerate the development andapplication of high performance computing technologies to meetNASA's science and engineering requirements. In cooperation with theother Federal agencies, NASA's program will deploy teraflops computercapabilities essential for computational design of integrated aerospacevehicle systems and for predicting long term global change, and willenable the development of massively parallel techniques for spacebomeapplications.

NASA's program is focused on bringing together interdisciplinary teamsof computer scientists and computational physicists to develop thesetechnologies within three vertically integrated projects that are unique toNASA's missions. These technologies include applications algorithmsand programs, systems software, peripherals, networking, and the actualhigh performance computing hardware. NASA will deve139 a suite ofsoftware tools to enhance productivity. These include: load balancingtools, run time optimizers, monitors, parallelization tools, as well as datamanagement and visualization tools.

NASA's role includes coordinating the Advanced Software Technologyand Algorithm component for the Federal program; acquiringexperimental hardware for testbeds in computational aerosciences, earthand space systems sciences, and remote exploration andexperimentation: and supporting the development of the NationalResearch and Education Network. To encourage vigorous research intothe underlying theory and concepts of high performance computing,NASA will foster interactions among academia, industry, and nationallaboratories and will strengthen the basic research in high performancecomputing at the NASA centers and research institutes and inuniversities.

NASA's considerable expertise in experimental parallel computertestbeds and small, scalable testbed systems will be used to demonstratehigh performance computing technologies as a step toward fullscalecomputational capabilities. A key to successful exploitation ofmassively pai.allel computing power will be the blending ofapplicationdriven and architecturedriven computer systems to mosteffectively meet NASA's needs.

National Science Foundation (NSF)

The NSF HPCC Program impacts the activities of all science andengineering disciplines by providing computing and networkinginfrastructure support and by developing enablinv, technologies foradvanced computing and communications platforms and paradigms.

In the area of High PerPrmance Computing Sy-tems, research will beinitiated on new architectures and systems optimized for specificresearch applications. New tools for systems level automated designand component packaging will be supported to permit the design ofapplicationspecific devices and systems.

In Advanced Software Technology and Algorithms, research will beinitiated on scientific database technology and implementation ofprototype networked databases and associated software. Advancedapplications tools for research computing environments will besupplied to enable nationwide access to the full complement of parallelmachines for research on grand challenge problems. Areas of researchfocus will include numeric and symbolic computing, algorithmdevelopment, optimization of applications software for new parallelcomputers, scientific visualization, automated programming tools, andnew methods of scientific and technical information exchanges.

NSF coordination of the National Research and Education Networkactivities will accelerate the harmonizing of multiple agency networksand protocols into a single NREN. The number of nodes will beincreased to expand distributed information resources, and to increaseredundancy, capacity, control, and security. Midlevel nets will beassisted to upgrade facilities and service very high bandwidthrequirements. Support will be provided for research on new protocolsfor gigabit networks, switch and transmission technology, routing,congestion, and flow control. The exploration of pricing mechanismsfor network services and neiwork applications and structured transitionto commercial service will be initiated.

In Basic Research and Human Resources, support will includemultidisciplinary research and university infrastructure for computerscience, computer engineering, and computational science andengineering. To increase the human resource pool in computinghardware and software systems areas, support will be expanded forgraduate and postdoctoral positions. Curriculum improvement, teachertraining, and support for centers to provide education and training in theuse of experimental and parallel supercomputers are integral parts ofthis component.

31

National Institute for Standards and Technology (DOC/NIST)

The NIST research program is directed toward developing performance.

monitoring tools and promoting "open systems" software. NIST has

proposed to augment its current HPC research program by promoting

the commercialization of protocol and security mechanisms for medium

speed networks.

The goal of NIST's activities in high performance computing is to

develop hardware performance monitoring tools, promote "ornsystem" software, and support a classification system for inde%ing and

distributing scientific software so that industry and the research

community can effectively exploit the power of future generations of

high performance computers. In support of The National Research and

Education Network, NIST will develop and speed commercialization of

network protocols and security mechanisms that can achieve the desired

gigabit speeds on future versions of the network. NIST will participate

in the HPCC Program by:

Developing hardware monitoring methods leading to load

characterization and performance measurement techniques for

ultrahigh--speed systems which will be made available on a

publicly accessible database.

Conducting research on new protocols and related securityprimitives necessary to sustain gigabit network speeds and

standards to provide interoperability, common user interfaces to

systems, and enhanced security for the network.

Establishing a network testbed at NIST instrumented to collect

performance data and test new network protocols, management

routines, and security mechanisms for gigabit networks.

4

32

National Oceanic and Atmospheric Administration(DOC/NOAA)

The National Oceanic and Atmospheric Administration operational andresearch programs are directed toward weather prediction, oceansciences, the Climate and Global Change Program, and the CoastalOceans Program, together with data management activities for all

agency programs. The HPCC Program will allow extensivedevelopment of new forecast models, studies in computational fluiddynamics, and the incorporation of evolving computer architectures andnetworks into the systems that carry out agency missions.

The NOAA High Performance Computing Program is focused in twocomponents:

Advanced Software Technology and Algorithms. This componentprovides support for: grand challenges in atmospheric and oceanicsciences; development of advanced numerical models to simulate thegeneral circulation of the oceans and atmosphere in support of NOAAmissions and the activities of collaborating research groups:development of new computational methods for solving atmospheric,oceanic, and related problems on new computer architectures; datamanagement R&D; support for basic research in suategies, techniques,and tools required for the management and analysis of largescaledistributed scientific databases and distributed data handling, includingquality control; and devclopment of algorithms for massively paralelprocessors, together with their standardized software componentlibraries and tools for the solution of oceanic and atmospheric analysisand forecasting problems.

NOAA will acquire, install, and operate advanced computationalfacilities for the evaluation of near operational prototype computershaving massively parallel architectures and will dcvelop algorithmsappropriate for these architectures.

The National Research and Education Network component providesnetworking in support for NOAA's climate and global change researchcommunity and a wide range of agency missions in oceanography,weather prediction, and environmental sciences research. NOAA willevaluate advanced network protocols and hardware technologies relatedto its missions.

Environmental Protection Agency (EPA)

The EPA research program is directed toward the advancement anddissemination of computational techniques and software tools whichform the core of ecosystem, atmospheric chemistry and dynamicsmodels. The models extend the computational capability ofenvironmental assessment tools to handle multipollutant interactionsand optimization of control strAtegies.

The Advanced Software Technology and Algorithms component will,vf:lop new approaches for coupling numerical equation solvers to the

rformance features of emerging computer architectures includingspecific techniques for solution of partial differential equations onmassively parallel architectures. These solvers will be designed tooperate on a flexible, generic grid system decoupled from specificapplications to provide a testbed for evaluation and optimization. Theresulting solution framework can serve as the computational engine fora variety of interdisciplinary applications.

Advancement of scientific visualization techniques can provideexceptional benefits in increasing the pace at which environmentalproblems are solved. EPA's program includes basic research to improveuser interfaces and computational algorithms related to image rendering

and manipulation of geometric images. The main focus is onsimultaneous representations of multiple data sets and realtimevisualization in a heterogeneous distributed processing environment.Related research on storage and access techniques for massiveenvironmental data bases across multiple architectures will beconducted.

The Basic Research and Human Resources component includes theestablishment of technology transfer centers to propagate the use ofHPCC technology to State and local environmental groups, and Federalmanagers for optimization of pollution control and prevention strategies.A main goal of a Technology Transfer Center is to providenonsophisticated users with training and gu'i..4ance in the application of

a variety high performance computing tools to solve importantenvironmental challenges. The program also supports crossdisciplinecareer training through universities and other institutions to ensure acontinuing base of technicl professionals knowledgeable in the use of

high performance computing technology for environmental problem

solving.

National Institutes of Health / National Library of Medicine(HHS/NIH/NLM)

The HHSINIH program includes molecular biology computing, creation.and transmission of digital electronic images, the linking of academichealth centers via computer networks, the creation of advanced methodsto retrieve information from life sciences databases, and training in

biomedical computer sciences. The HPCC program will complementthe Human Genorne Project by providing new methods for computerbased analysis of normal and disease genes.

The Advanced Software Th.hnology and Algorithms cornponent willdevelop advanced software technology and algorithms in two areas ofimportance in biomedical research and education: biotechnologycornputin!-i and digital images. In the area of biotechnology computing,the program will support development of molecular sequencecomparison algorithms, new database methods, and algorithms topredict biological structure and function from genetic code. Thebiomedical images area will support new methods for representing,linking, and rendering images of biological structure

The National Research and Education Network component has two subcomponents: connections among academic medical centers and theirgrowing array of computerized information sources: and development ofintelligent gateways that link conceptually related databanks in the lifesciences. The academic medical centers of the country are confrontedwith a growirw array of disconnected computerbased informationsources, ranging from patient records, xrays and laboratory systems, tobasic research tools such as protein and DNA sequence databanks.Development of advanced software systems capable of representing andlinking these dissimilar data types, and communicating them amongcenters for research and health care, is the goal of this part of the

program. The focus of the Gateways HPCC program is the building ofsystems that are capable of translating a user 's request into multiplecomputerbased vocabularies, selecting appropriate databases from awide range of widelyavailable resources, and retrieving information ina manner that does not require users to understand the structure andsyntax of the systems being queried.

The Basic Research and Human Resources component addresses theneed to train biomedical researchers and health care providers in the useof advanced computing and network communications to aid in theirwork. The successful predoctoral and postdoctoral grants program for

career training in medical informatics administered by the NationalInstitutes of Health will be expanded.

4. GRAND CHALLENGE AND SUPPORTINGTECHNOLOGY CASE STUDIES

This section incorporates examples of high performance computing andcomputer communications technologies and several illustrative grandchallenge applications in computational science and engineering, thatare presented on the cover of this report. The examples were chosen to

illustrate the diversity and significance of application areas that havebeen addressed to date. The list of grand challenges is too long to allow

an example from each area, thus lack of representation of certain areasdoes not imply lack of importance. A more complete description of thegrand challenges can be found in the Federal High PerformanceComputing Program report issued in September 1989.

Forecasting Severe Weather Events

Making accurate predictions about the behavior of the atmosphere is ofcntical importance to reducing losses due to storms, hurricanes, floods,and other weather related phenomena. Current operational small scaleweather prediction models have been constrained by two broaddeficiencies: inadequate size of the data sets necessary to define thedetailed atmospheric structures and insufficient computer resources tosupport high resolution. During the next five years, major progress willbe made on the first deficiency by using a combination of newgroundbased and remoteobserving systems. This leaves inadequatecomputer power as the primary stumbling block preventing highresolution models from playing operation.c.l roles in our national weatherprediction efforts.

The highresolution, operational weather prediction models of the futurewill represent a new generation of numerical formulations. The primarydifferences will be in the treatment of vertical motions and small scaleprocesses which in the past have been considered to be "subgrid scale."This means that physically significant events, such as convection, mustbe treated as parameterized processes and fine scale details in, forexample, thunderstorm evolution that can affect the sprroundings cannotbe directly addressed by the model. Likewise, fine scale observations oflocal importance, such as mlisture gradients around wet areas, whichcan provide local forcing, currently cannot be incorporated adequately.

To meet this scientific and national need, new technologies need to bedeveloped and incorporated into existing facilities so that a less than 5km resolution model can be operational before the end of ele century.This would allow updated forecasts on a 6hourly or shorter basis. Suchmodels have already shown significant advances in the accuracy ofpredicting a wide variety of weather events, from severe thunderstormsto lake effect snowfalls in research applications. Each reduction of themodel resolution by onehalf requires an increase of computer power ofalmost an order of magnitude, as well as comparable increases insupporting memory, mass storage and networks. For example, a 5 kmmodel could require up to a 20 teraflops computer system to meetoperational schedules using this model.

a )0 *C;rita

4:41 .4I o

e

Researchers have used supercomputers to model thunderstorms numerically.Wind, temperature, pressure, and other variables are calculated every fewseconds at several hundred thousand locations in the area of the developirmstorm. Mathematical equations are then used to simulate the storm's evolution.In this graphic, particles which are released near the ground at regular intervals

are colored orange when rising and Nue when sinking. Yellow signifies paths

of individual particles.

4f;39

Cancer Genes

DNA is the blueprint of life, the molecular thread in the nucleus of eachof our cells which guides the assembly of molecules and complete living

organisms. When the DNA code is altered by mutation, serious diseasescan result, such as cancer; this phenomenon was known to scientists

studying animal tumor's in the 1970's. They isolatcd cancercausinggenes, called "oncogenes" from animal tumors, and later found thatsimilar genes existed in normal human DNA. This was a profoundmystery. Why would we be carrying the seeds of our own destruction in

our genetic blueprint?

In 1984, two separate research groups used a computerized searchingalgorithm to compare a newly discovered oncogene to all known genes.

To their astonishment, the cancer causing gene matched a normal geneinvolved in growth and development. Suddenly, it became clear that

cancer might be caused by a normal growth gene being switched on at

the wrong time. This fundamental and unexpected insight was an early

example of a field that is now known as Computational Biology, thescience of using computers , store and analyze data from complexmolecules in living cells.

The databases used in 1984 for these comparisons contained information

about several thousand molecular units; now they contain over 30million. Moreover, the current multiagency genome research programsof NIH, NSF, and DOE will acquire data on tens uf billions of molecularunits, ranging from simple organisms to human beings. The bestcomputer algorithms for determining the similarity of molecules require

time proportional to the length of the molecules being compared; if the

methods used to analyze oncogenes in 1984 were applied to the three

billion base pairs of the human genome, they would require hundreds of

years of computer time on today's fastest supercomputers. New

computer designs and software methods will be essential to cope with

the explosive growth of molecular data. Functioning as intellectual

amplifiers to detect similarities and differences in molecules whose size

and complexity are too vast for the unaided human mind, the computer

systems to be developed by the HPCC program will be a critical tool for

the life sciences in the 1990's, and the health care systems of the 21st

Century.

40

MIP

411419411.44etaorlot

4004,40.040" ma.

ada4ft e.°

a alal wait.111

i

Two views are shown for the binding of a chemical carcinogen to a short stretch

of DNA. The stick model on the left shows the overall structure of theDNAcarcinogen complex. The space filling model on the right shows the

actual intimate binding of the carcinogen, hiding deep within the DNA. This

chemical contact leads to mutations in the DNA code, and ultimately to a tumor.

Predicting New Superconductors

In 1988, the world was excited by the discovery that a particularyttriumbariumopperoxide compound superconducts at atemperature of 93° Kelvin, still very cold, but much warmer than anyprevious superconductor. This discovery sparked a worldwide effort toexpand research to discover new superconducting materials. Theeconomic benefit of a hightemperature superconductor is beyondcalculation, portending the development of, for example, much moreefficient power transmission and lightweight, powerful magnets torevolutionize the transportation business. Advanced computing is acentral part of the arsenal of research tools which will be necessary toreach that payoff.

Despite these early successes, many questions remain before it isunderstood how some materials superconduct when very similarcompounds do not. This understanding will be critical to predicting newsuperconductors, which might work at even higher temperatures, be lessexpensive, carry more current, or be more amenable to manufacturingprocesses. Increasing progress in all these areas is required before theimpact of these new materials will be telt.

The solution of physical models requires intensive calculations tounderstand the material structure. High performance computing canshorten the discovery process by allowing the development accuratesimulations to point experimenters in the most promising directions. Forexample, most of the groups looking for new superconthictors are tryingvarious copperoxide combinations. Researchers are using highperformance computers to explore the possibility of variouscombinations of elements that may lead to new superconductingmaterials.

4:

42

Researchers have used visualization techniques to investigate the structure ofmaterials which are thought to be involved in the uperconducting mechanism.In this graphic, barium cations (green) and yttrium cations (-,,ilver) are shownwith their associated oxygen defects. The copper atoms (blue) and oxygenatoms (red) are found in two types of Cu04 environments, one of which isdepicted in yellow, the other in light blue.

43 (i

Air Pollution

The ability of the atmosphere to absorb and to cleanse itself of pollutantcontaminants was taken for granted until the 1950's when "killer fogs"in London, England and Donora, Pennsylvania caused the deaths ofhundreds of people. Since then, technological advances in controllingsource emissions have reversed the trend of steadily increasingpollution, even in the face of continued industrial growth. However,reduction strategies of individual pollutant types do not always producethe desired results. In fact, these simple solutions can even make airquality worse due to complex chemical interactions of the remainingairborne contaminants. Pollutants may travel long distances fromindustrial centers to sensitive areas where they are deposited intransformed products such as ozone and acid rain. These complexitiesof pollutant transport and transformation are costly and difficult to studyexperimentally, therefore, numerical models of the atmosphere havebeen developed to assess the effects of manmade emissions on airquality.

The new Clean Air Act mandates the use of numerical models todemonstrate the effectiveness of proposed regulatory control strategies.The potential cost to society to implement these proposed controls isestimated to reach tens of billions of dollars. Current models have beenuseful in evaluating alternative control strategies, but do not yet have thecapability to produce optimum solutions. Present computing limitationson existing supercomputers force simplifications in the scientificdescriptions of chemical and physical processes, and slow examinationof alternatives. Control strategies for each pollutant are oftendetermined independently with little evaluation of multiple pollutantinteractions. Remedial solutions determined for a particular scale ofspace and time are not readily extendable to other scales of pollutantdispersal.

High performance computing will enable multiscale numericalexplorations with crosspollutant interactions to be performed in atimely manner so as tolv useful to legislated control and preventionrequirements. Advanced computer designs and software methods willalso enable cost optimization of pollutant control tradeoffs. Improvedvisualization techniques will enhance the interpretation and evaluationof massive amounts of environmental measurement and computersimulation data. High performance computing models will lead to abetter understanding of the actions needed to minimize pollutant damageto materials and environmental damage to crops while making our airsafer to breathe for future generations.

MIj,

44

Using a supercomputer, atmospheric researchers have simulated the transport,chemical transformation, and surface deposition of sulfur compoundsresponsible for acid rain, Visualization specialists h v depicted the movementof sulfur compounds from major sources in the Ohio river valley to sensitivelakes in the Adirondacks. The yellow cloud represents high concentrations ofsulfur compounds. Several sensitive aquatic regions are outlined in green and atypical wind flow pattern is presented in red.

45

Aerospace Vehicle Design

Being able to predict the aerodynamic characteristics of inflightvehicles is important for designers. Reproducing such flight regimes inwind tunnels is time consuming, costly, and in some cases impossible.Computational aerodynamics simulations are less costly and much fasterthan complex wind tunnel tests and are able to simulate manyinaccessible flight regimes. These capabilities will be particularlyimportant in the design of cupersonic and hypersonic aircraft to serveinternational markets.

Computational aerosciences directly contributes to maintaining U.S.preeminence in aerospace science and engineering disciplines. Thecomputational technology developed in such computationalaerodynamics problems will directly transfer to the U.S. aerospaceindustry and aircraft manufacturers. Other potential beneficiaries are indiverse fields where fluid dynamics is an important design aspect suchas automobile manufacture, ship design, and medicine (e.g.,heart/cardiovascular flow simulation).

Massively parallel computing systems and advanced parallel softwaretechnology and algorithms will enable the development and validationof multidisciplinary, coupled models. These models will allow thenumerical simulation and design optimization of complete aerospacevehicle systems throughout the flight envelope.

Allirli,11111111,1111IL

Shown here is a comparison between supercomputer simulation data andactual wind tunnel model data for the pressure distribution along theintegrated space shuttle, solid rocket boosters and external tank flying atMach 1.55 Note the excellent agreement between the two.

/

Energy Conservation and Turbulent Combustion

For the foreseeable future, 90% of the energy needs of the United Stateswill be met by the combustion .f-c fossil fuels. Two of the largest usesare for electrical power generation and in automobiles. Computationalmodels offer a means of improving the design and efficiency of internalcombustion engines.

Automobile engines are most efficient when run at high temperatures,but increased temperatures also lead to increased nitrogen oxideemissions. The burning of alternative fuels such as methanol iscomplicated by the emission of carcinogens. The effects of theemissions are bfluenced by local climatic conditions, making itnecessary to consider the total system of fuel, engine, and atmospherewhen seeking better designs. Our environment is too delicate andcannot be used as a testbed, so the atmospheric effects must besimulated.

A full three dimensional engine design code L_ been developed andimplemented on a supercomputer. The code is designed to handle themost complex engines, such as the stratified charge and the twostrokeengines. Tin code represents an approximation to reality, because not allof the physical phenomena are well understood. Moreover, even if theywere, the limited capabilities of existing computers would not allow thisdetailed information to be included.

For example, over 400 chemical processes of hydrocarbon and nitrogenchemistry are known to occur in internal combustion engines, yet onlyten or less of the most significant reactions are used in simulations, inorder to allow the calculation to run in a few hours on today's largesupercomputers. Since the 400 hydrocarbonnitrogen reactions areknown, the real problem could be addressed with better algorithmsrunning on a machine 10,000 times more powerful, a teraflops machine.This computational technology is needed by many private industrialengine firms as well as universities and government laboratories.

Researchers have used high performance computers to accomplish thenumerical simulation of the properties of a conical fuel jet. The colors of theparticles within the jet indicate the droplet size. The smallest particles are blue,intermediate size particles are shown in yellow, while the particles with thelargest diameters are depicted in red. The smaller light blue particles around thejet represent the swirling air surrounding the spray.

Microsystems Design and Packaging

Since the invention of the integrated circuit ii 1958, the number oftransistors fabricated on a microchip has doubled every two years,providing a medium to incorporate everincreasing complex electronicdesign into chips, components, and packages. By anaiogy, thecomplexity of detail incorporated in a single integrated circuit 1 cmsquare is equivalent to representing the map features, at a city blocklevel, of the entire Eastern United States. A similar analogy for a 5 cmsquare module densely parked with a collection of chips would be theequivalent of a map of North and South America. Determining theinterconnecting paths, selecting the right modules, testing the interfaces,and choosing the mix of technologies are part of the design process tobuild the scalable components for workstations to supercomputers.

In the computing world, scalable architectures based on the 1-2 milliontransistor custom structures of today will evolve in the mid 1990's tosystem approaches exploiting 10milliontransistor chips, standardcomponent parts, and special interface electronics, all combined inoptimally designed modules adhering to standard interfaces. Systemclock speeds will continue to improve, and the diversity of microchiptechnologies combined in a single module will allow unprecedentedflexibility for designers.

Managing this complexity explosion would be overwhelming withoutthe use of computationally based approaches that enable teams ofdesigners to systematically reduce the time to develop such systems.Today, computational tools enable complex microcircuits to bedeveloped on first pass, at the same time that packaging technologiessuch as multichip modules can be demonstrated. High performancecomputing applied to the technology design process will enable theexploration of design alternatives and rapid exploitation of newtechnologies.

IMO.1S 4

a a

a

1

The figure shows both current and future components of the innovation designcycle. In the lower left of the photograph is a multiproject wafer using anindustry/academia/government supported prototyping capability Themultiproject wafer shown is 4 inches in diameter, shares resources with 82

projects, ard represents approximately 200 million transistors distributed over thewafer. Proven experimental prototypes are used as components of larger modules,interconnected with advanced technology. In the upper right hand corner is anexample of an advanced interconnect module of 36 diverse microchips, packedtogether 1.5 inches on a side. Advanced computationally intensiw; Jesign toolsare essential to realize these high petfoimance components.

5 1

Earth's Biosphere

The Earth's biosphere is a complex physical system. There are amultitude of phenomena which can change the state of the biosphere on

a local, regional, and global scale. In order to predict the directions andconsequences of changes in the state and condition of the biosphere,detailed scien.jic models are required, which in turn are constructedfrom massive amounts of experimental and computational data.

Experimental information is derived from satellite and ground basedsensors. By the late 1990's, the sensors will have the resolution required

to provide data in support of much more accurate and informed decisionmaking on issues such as pollution and global warming. Because thesensors will generate terabytes of data each day, which will be combinedwith local data sets on the ecosystem, major improvements in thecapability for collecting, analyzing, distributing, and archiving data arenecessary.

The effort of constructing valid scientific models which describe thedynamics and underlying processes of the biosphere will involveinterdisciplinary teams of experts from the geophysical, life, physical,computer, and computational sciences. They will work together toconstruct computational models which will validate our empiricalunderstanding of the biosphere, and help predict how worldwideactivities affect the global ecosystem. Computer and computationalscientists will develop the advanced software technology and algorithmsfor handling massive amounts of data and working with high resolution,coupled models of the Earth's atmosphericbiosphericoceanicinteractions. Efforts in software development and experimentation willpredict how local current conditions may impact future globalconditions allowing the linking of ecosystem models to global climatemodels. The result of this collaborative effort will be a much deeperunderstanding of our environment and the impact of human activitiffupon it.

me.. 4. .

VP'

/Aar-r.

A global view of the Earth's ocean chlorophyll and land vegetation is depictedhere. The image was derived from accumulated satellite data. Generating thisimage required over 2 terabytes ( 1012 bytes) of raw data. Data such as these arethe baseline -snapshot" of our current biosphere and arc Ital to understanding

short and long term component interactions of the Earth's biosphere for suchpurposes as crop productivity improvement, fishing and weather prediction.

53 (; I(

High Speed Networks

The development of more powerful computers feeds both the demandfor, as well as the growth of, more powerful data communicationscapability. As computing technology progresses, greater demands areplaced on network performance as researchers conceive of new tasksand modalities of use that require even higher performance.

Current developments in large scale scientific computing are leading totruly distributed computing, allowing a given job to be executed onseveral different machines communicating partial results amongthemselves, sharing in different facets of calculations, and jointlyassembling a final result for output. Such intermachine communicationis inherently capable of taking place at speeds that stress local areanetwork technology and are a hundred or more times faster than arepossible on today's long haul networks. As an example, some of thenation's leading radiology departments have committed to digitaltransfer of radiology images on broadband local networks operating atspeeds of 100 to 1000 million bits per second.

In order for the NREN to support these and other demandingapplications, not yet contemplated, a substantial directed research anddevelopment effort is needed in the areas of protocols (the formalstructure of inter-computer communications), high speed computerinterfaces for computers, and network equipment, such as switches.Multi-gigabit networks represent a change in kind, not just in degree,from today's networks. For example, consider that in a coast-to-coastcommunication at three gigabits per second there are at any instant "inflight" nearly nine megabytes of data, which is mc-e than the memory ofmost personal computers and workstations.

Some gigabit research has already begun, and several experimemalfacilities have been established in a productive collaboration ofacademic, industrial, and governmental organizations, but HPCCsupport will be needed to carry the rest arch program to the stage thatcommer -ial providers can use the technology to install and operate amulti-gigabit NREN.

Education Using the NREN

Computers increasingly fill important niches in all phases of the learningprocess, providing flexible instruments for interactive instruction andstudent based learning experiences. Their use in education bothprovides the skills needed to function in our increasingly technologyintensive world, and aids teaching and learning of all science andengineering topics. The development of fir. National Research andEducation Network will accelerate and transfer the technology ofcomputer communications to the needs of educators and studcrits. Theresult will be to empower them to share resources and ideas on anational scale.

In a recent project, classes in many locations chose a day to measure thelength of the sun's shadow from a vertical measuring stick on the schoolgrounds at 12 nuon. Each class consulted maps and geography books tofind its school's latitude, and sent the results to a shared database locatedin the U.S. and Canada. All schools then received the database ofmeasurements from around the world, and each class used the completedatabase to calculate the curvature of the earth, and from that, the earth' sdiameter. A normally dry recitation facts became an engagingproblem solving exercise because the students themselves derived theanswer from their shared measurements. Along the way they learnedgeography, geometry, statistics, and how to collect and share data overcomputer networks.

This project, implemented by the use of the network, is a learninglaboratory without walls, similar to the way research scientists takeadvantage of high speed digital networks to conduct shared research thatis "distance independent." The HPCC Program will face the challengeof "scalingup" today's Internet, making it "userfriendly" andimproving its services so that it is readily accessible to all U.S.educational institutions.

Glossary

ASTAAdvanced Software Technology and Algorithms

Bitcronym for binary

BRHRBa. ic Research and Human Resources

ByteA group of adjacent binary digi6 operated upon as a unit (usually connotes

a group of eight bits)

CADAcronym for "computer aided design".

Computer EngineeringThe creative application ofengineering principles and methods to thedesign and development of hardware and software systems.

Computer ScienceThe systematic study of computing systems and computation. The body

of knowledge resulting from this discipline contains theories forunderstanding computing systems and methods; design methodology,algorithms, and tools: methods for the testing of concepts; methods ofanalysis and verification; and knowledge representation and

implementation.

Computational Science and EngineeringThe systematic application of computing systems and computationaisolution techniques to mathematical models formulated to describe and

simulate phenomena of scientific and engineering interest.

flopsAcronym for "floating point operations per second". The term "floatingpoint" refers to that format of numbers which is most commonly used forscientific calculation. Flops is used as a measure of a computing system's

speed of performing basic arithmetic operations such as adding,subtracting, multiplying, or dividing two numbers.

Giga109 or billions of ... (e.g.: gigabits)

Grand ChallengeA Grand Challenge is a fundamental problem in science and engineering,

with broad economic end scientific impact, whose solution could beadvanced by applying high performance computing techniques and

resources.

56 13 3

HPCCHigh Performance Computing and Communications

HPCSHigh Performance Computing Systems

High Performance ComputingHigh performance computing encompasses advanced computing,communications, and information technologies, including scientificworkstations, supercomputer systems, high speed networks, specialpurpose and experimental systems, the new generation of large scaleparallel systems, and applications and systems software with allcomponents well integrated and linked over a high speed national network.

Mega-106 or millions of ... (e.g.: megaflops)

NetworkComputer communications technologies that link multiple computers toshare information and resources across geographically dispersed locations.

NRENNational Resear,:h and Education Network

opsAcronym for "operations per second". Ops is used as a rating of the speedof computer systems and components. In this report ops is generally takento mean the usual integer or floating point operations depending on whatfunctional units are included in a particular system configuration.

parallel ProcessingSimultaneous processing by more than one processing unit on a single

application.

Peta10 15 or thousands of trillions of ... (e.g.: petabytes)

SupercomputerAt any given time, that class of generalpurpose computers that are bothfaster than their commercial competitors and have sufficient centralmemory to store the problem sets for which they are designed. Computermemory, throughput, computational rates, and other related computercapabilities contribute to pelf _mance. ConsequenCy, a quantitativemeasure of computer power in largescale scientific processing does notexist and a precise definition of supercomputers is difficult to formulate.

Tera-1012 or trillions of ... (e.g.: teraops)

57

G4