|
Networked Computing for the 21st Century
High End Computing and Computation |
 
|
|
|
|
|||
 |
|
|---|---|
|
|
|
|
|
Overview
|
HECC research and development (R&D) investments provide the foundation for U.S. leadership in high end computing and promote the use of high end computing and computation in government, academia, industry, and in broad societal applications. HECC research explores algorithms for physical, chemical, and biological modeling and simulation in complex systems; information-intensive science and engineering applications; and the advanced concepts in quantum, biological, and optical computing that will keep the U.S. in the forefront of computing breakthroughs for years to come. HECC R&D supports critical Federal government mission needs, including:
Federal investments in four HECC areas, or thrusts, will enable development of distributed, computation-intensive applications to support future U.S. science and engineering research, national security priorities, and economic competitiveness. |
|
|
|
|
Thrust 1
|
System software technology The aim of this thrust is to improve the usability and effectiveness of teraflops-scale systems across a wide range of government, industry, and academic applications, concentrating on medium term technology development (three to five years). Thrust 1 activities address high end architectures, including symmetric multiprocessor systems (SMPs), clusters of SMPs, and a computational grid of distributed homogeneous and heterogeneous clustered systems. Longer-term activities will focus on the system software technology requirements of future generations of high end system architectures. Thrust 1 efforts recognize that Government investments are required, since the market size for high end computing is not large enough for system vendors and independent software vendors (ISVs) to make significant investments in the Federal mission-critical system software technology required for computationally intensive applications. It is appropriate that as Government-developed system software and tools mature they become the property of non-government entities and shared resources throughout appropriate segments of the research community and industry. Thrust 1 investment focus areas are:
|
|
|
|
|
Thrust 2
|
Leading-edge research for future generations of computing Driven by Federal, academic, and commercial applications, Thrust 2 focuses on long range research and technology development to achieve a sustained pflops (1015 floating-point operations per second) computational rate and exabyte (1018 byte) storage. Thrust 2 activities will support U.S. global leadership in high end computing (HEC), ensuring that scientists and engineers, especially those working on Federal missions, will continue to have access to the most powerful computers, and assuring that the research and technology necessary for HEC systems will be available to U.S. industry. Activities requiring petaflops speed and exabyte storage include:
|
|
|
|
|
Thrust 3
|
Incorporating technology into applications Many HECC agencies support mission-driven scientific applications requiring large scale computation-intensive and/or data-intensive operations spanning the full space-time spectrum of scientific problems. Thrust 3 R&D will incorporate the first use of HECC methods into agency applications, and encourage the use of computational science algorithms to solve problems requiring high performance computational facilities, ensuring that key applications execute at full potential. Advances are needed in fast, efficient algorithms for computational science techniques that address problems in very large, sparse matrices, searching, sorting, and pattern matching. Research on algorithms with large amounts of concurrency, fault tolerance, and latency-hiding is crucial for high end computational systems of the future. |
|
|
|
|
Thrust 4
|
Infrastructure for research in HECC The Thrust 4 objective is to ensure a balanced R&D infrastructure with maximum computational strength and bandwidth. Interdependent with LSN activities, this thrust supports the research facilities built on large scale test systems and on large scale, high performance computational grids and networks. |
|
|
|
|
Accomplishments and plans
|
FY 1998 accomplishments and FY 1999 plans Prudent agency investments in coordinated R&D areas will enable development of the distributed, computation-intensive, and data-intensive applications needed to assure future scientific, engineering, and economic competitiveness, and fulfill national security requirements. Select agency R&D efforts, organized by thrust area, include the following:
|
|
|
|
|
Thrust 1
|
System software technology -- FY 1998 accomplishments and FY 1999 plans |
|
Distributed computing |
National Science Foundation (NSF) -supported work in distributed multimedia systems has as objectives the development of:
|
|
Proof carrying code |
NSF-funded researchers at Carnegie-Mellon University have developed methods and implementations that allow highly secure execution of code downloaded from untrusted sources. The problem of intrusion via imported programs has become significant in the Java/Web-based environment of much of today's computing. The new technique is based on downloading with the code a proof that the code satisfies desirable safety characteristics. Theorem-proving software on the receiving site checks the given proof against the code, assuring that the code indeed conforms. If either the code or the proof has been tampered with or otherwise corrupted, the proof will fail, making this method highly resistant to attack. This work capitalizes on research in programming language semantics, theorem proving, compiler techniques, and formal methods. Early products of this research are relatively efficient proof-checking mechanisms and a "certifying" compiler for a significant subset of C++, a compiler that annotates the compiled code with predicates that it guarantees are satisfied by the object. |
|
Scalable software and ease of use |
Defense Advanced Research Projects Agency (DARPA) -supported projects develop scalable software geared toward easing the use of systems by application programmers. In FY 1998, DARPA will:
|
|
Operating system extensions |
In FY 1999, DARPA plans to design instruction set extensions and storage
components to allow Defense applications to specify whether operations
are executed in the central processor or in logic circuits embedded in
the memory hierarchy.
|
|
High performance software tools |
DOE computer science and software tools work focuses on software to facilitate the use of high performance systems for solving scientific problems. Efforts range from efficient operating systems and I/O for massively parallel processing (MPP), to frameworks for isolating application codes from the underlying hardware details, to tools for monitoring the performance of scientific applications, and to efforts to improve the management, visualization, and understanding of the results of high end computations. DOE projects include software tools for program development environments and performance monitoring for Advanced Computational Testing and Simulation (ACTS). The goal is to develop an integrated set of algorithms, software tools, and infrastructure that will enable computer simulation to be used in place of experiments when real experiments are too dangerous, expensive, inaccessible, or not politically feasible. In FY 1998, DOE will:
|
|
Earth and Space Sciences (ESS) |
The FY 1998 goal of NASA's ESS scalable testbed is 50 gigaflops sustained. For this effort, NASA will develop pre-competitive prototype systems software that provides high availability and portability demonstrated in a large scale production environment with the objective of eventual commercial availability. In FY 1998, the goal of NASA's 100-250 gigaflops sustained scalable testbed is 100 gigaflops sustained. |
|
MARQUISE |
In FY 1999, NSA will flight test SOLITAIRE, the embedded high performance computer prototype developed under the MARQUISE program, on an Air Force aircraft. The agency will continue research on miniaturized, spray cooled, and embedded diamond power supplies. In FY 1999, NSA expects to demonstrate a 48-Volt to 2.8-Volt @ 200 Amps high-efficiency power converter with a power density of 2 Kwatts/inch3. NSA continues to conduct research in embedded scalable nodes (the follow-on architecture to MARQUISE) for mission scenarios, and has selected three potential repackaging implementations that could make a 100 percent software-compatible, miniaturized high performance computer available six months after the introduction of the commercial computer. 
The SOLITAIRE board contains a complete SGI/CRAY J90 System Element,
which includes four vector processors, network crossbar, I/O, and 1/2
Gbyte of dynamic random access memory (DRAM). There are nine diamond
based die cast multichip modules (MCMs) mounted on one side of the board
and eight die cast MCMs mounted on the reverse side. SOLITAIRE
represents a new achievement in double sided board column grid
attachment, featuring over 25,000 connections between the modules and
the board. This repackaging implementation of the System Element is 80
percent smaller and 75 percent lighter than the commercial product, and
concentrates 500 Watts onto a 10 inch square board.
|
|
Microelectronics in high speed computing |
Related NSA research areas include synthetic diamond packaging technology, all-optical switching, and optoelectronic integrated circuit packaging technology. Other activities include:
Pictured to the left is a 128x128 superconductive crossbar switch. The
switch matrix is mounted on the 4"-diameter MCM and maintained at 4
degrees Kelvin temperature. It is connected to its room-temperature
environment by an eight multichannel electrical cable.
A spray cooled dc/dc converter can supply an impressive 200 W/in3. This
magnitude of miniaturization allows tight integration of the power
converter and electronics in a shared environment. Point of load
conversion eliminates heavy bus bars and rigid cables by distributing
power at high voltage and low current. Spray cooling results in relaxed
device specifications and improved reliability due to reduced operating
junction temperatures and thermal gradient. A 28 vdc/4 vdc spray cooled
converter is pictured to the left.
|
|
UPC |
NSA's center for Computing Sciences (CCS), in partnership with the University of California at Berkeley and Lawrence Livermore National Laboratory, is developing UPC, a language that combines features of the CCS-developed AC with features of Split-C and PCP, two other parallel C languages. UPC represents an attempt to develop a base language that could become a standard for explicitly-parallel C. UPC extends AC in three ways:
|
| Quantum Computing |
In FY 1998 and 1999, NSA will continue fundamental exploratory studies in quantum computation theory and experimentation, including algorithms and complexity theory, quantum error correction techniques, quantum decoherence processes, and small scale laboratory implementations. |
|
Performance measurement and testing technologies |
NIST continues developing application-level measurement techniques for local area network (LAN) -clustered workstations, including physical performance monitors and software tools. These include the acquisition of performance information from the operating system, communication libraries, and their associated daemons, all of which have an impact on concurrent applications, but today provide little, if any, feedback to the user. Since the kernel and daemons operate in parallel with the application, data acquisition and resolution are more difficult. NIST will investigate the Linux kernel and message passing libraries such as MPI to provide performance information and develop techniques that allow users to acquire information selectively. Other work includes investigations into optical tape subsystem designs and testing methods for digital video disk (DVD) media. |
|
Parallelizing environmental modeling software |
A framework for parallelizing a broad range of environmental modeling codes has been developed by EPA-supported researchers at Iowa State University. The key feature is class-specific automatic parallelization where high-level knowledge about the specific class is used to arrive at efficient parallel codes. This approach combines expert system and compiler technologies to automate tedious and time-consuming aspects of hand coding. The current system (PA-1) can parallelize 3-D time matching explicit finite difference codes. One person used PA-1 to parallelize the Mesoscale Meteorology model (MM5) in two weeks, as compared to three years for manual parallelization. A scalable 2-D Regional Acid Deposition Model was parallelized in three weeks versus 10 months for manual parallelization of a simpler one-dimensional model. |
|
HPC modernization |
Although not formally part of the CIC R&D crosscut, the Department of Defense (DoD) HPC Modernization Program, which is represented on the HECC Working Group, will provide advanced hardware, computing tools, and training to DoD researchers using the latest technology to support the warfighter. Incorporating high performance computing into the system design process will allow the U.S. to maintain its technological supremacy in weapons systems design into the foreseeable future, and the use of high performance computing in the early stages of system acquisition will decrease the total life-cycle costs of new warfighting support systems. |
|
Common High Performance Computing Software Support Initiative (CHSSI) |
Also part of the DoD HPC Modernization Program, CHSSI is an application software development program designed to provide DoD computational scientists and engineers with software for scalable computing systems. These products will improve DoD science and technology and developmental test and evaluation computation. Core applications focus on improved 21st century military capability. |
|
|
|
|
Thrust 2
|
Leading edge research for future generations computing -- FY 1998 accomplishments and FY 1999 plans |
|
Computational models |
In computational models, DARPA will conduct R&D in new classes of computing technologies that may offer performance/cost/size/weight/power improvements beyond the limits of today's semiconductor-based computing. |
|
Intelligent systems technology |
The focus of DARPA's intelligent systems technology effort is on advanced techniques for knowledge representation, reasoning, and machine learning, which enable computer understanding of spoken and written language and images. Intelligent systems technology research includes:
|
| Scalable computing |
This R&D focuses on hardware and software technologies leading to a
secure scalable computing and communications technology base for systems
configured over a wide performance range, from mobile handheld devices,
to desktop workstations, to large scale distributed systems. In FY 1998,
DARPA will demonstrate highly efficient, parallel nodes;
auto-parallelization of file I/O for scalable systems; first node-level
performance of ultra-low-power systems; performance of novel backplane
networks supporting security; and hardware-accelerated, distributed,
shared-memory performance on workstation clusters.
|
|
Ultrascale computing |
In FY 1998, DARPA will:
|
|
Global scale system software |
In FY 1998, DARPA will:
|
|
Information survivability |
DARPA research in information survivability will focus on:
|
|
Multi-threaded architecture (MTA) |
In FY 1998, as part of its work in evaluating prototype computing
systems, DOE will evaluate the performance of a suite of algorithms and
DOE-critical applications on the first Tera Computer system, which is
used in evaluating the MTA that is critical to all proposed petaflops
architectures. In FY 1999, DOE will evaluate software requirements for
petaflops systems and new architectural developments.
|
|
Non-binary optical storage systems |
Today's optical disks represent data in a binary format -- zeroes and ones. NSF-funded researchers at the Georgia Institute of Technology are developing optical systems to store data in "M-ary" format, where M can be larger than two. Working with Optex Communication Corporation, researchers helped design the signal processing components of an optical memory in which each stored symbol can take on one of M=6 different values, resulting in greater information density per square inch in optical recording. This research is funded by a Presidential Early Career Award for Science and Engineering (PECASE). |
|
Multi-user detection in CDMA communication systems |
Code division multiple access (CDMA) permits multiple users to be assigned the same frequency in a digital cellular communication system: each user gets a unique signature sequence that can be recognized even when there is interference from other users. In this NSF-funded work, multi-user detection is a new approach to implementing CDMA that exploits the fact that the base station knows the signature sequences for all users and can use that knowledge to better combat interference. Optimal multi-user detection is prohibitively complex to implement today. However, NSF-funded researchers at the New Jersey Institute of Technology, North Carolina State University, Ohio State University, Rice University, and the University of Colorado are developing low-complexity algorithms and associated hardware. The benefit will be the ability to handle more users per cell at higher data rates. |
|
Knowledge Networking |
NSF will continue to support a broad academic research program in computing systems, including "Knowledge Networking," an initiative focused on the next generation of interconnected networks and associated database and collaborative technologies. |
|
HTMT |
NSA's Hybrid Technology Multithreaded (HTMT) architecture effort is a collaborative project among a dozen research groups (Caltech/JPL, University of Delaware, SUNY Stony Brook, University of Notre Dame, Princeton University plus an association of other government and industry labs). The objective is to define a very high performance computer system that can reach a petaflop level of performance in significantly less time than a conventional approach. HTMT will be based on a unique multithreaded execution model and will use a combination of leading-edge technologies, including superconducting technology, high-speed VLSI semiconductor technology, optical interconnect, and storage technology. The system is expected to be used in several critical high performance applications of national strategic importance over the next decade. An initial study of the original HTMT concept was funded by NSF and NASA in 1996-1997. A comprehensive design study of HTMT is now sponsored by DARPA, NSA and NASA, and is managed through Caltech and JPL. The research being performed at University of Delaware is focused on the program execution and architecture model of HTMT. |
|
High speed circuits |
NSA is conducting research to determine whether electronics can handle the 100 Gb/s serial data rates of fiber optic links. Uses include Tb/sec data transfer. |
|
RES, JACKAL, and LOTS |
RES, a system developed in 1991-1992 with NSA HPCC funds, was designed to harness "idle" workstations. It is seeing increasingly widespread use and is the subject of continuing enhancement. JACKAL is a cliche-based software reverse-engineering research system (where a "cliche" is an "interesting" piece of code). Input to JACKAL is a high-level abstract language translation of source code (when available) or decompiled code. JACKAL uses tree-matching and string-matching algorithms to identify cliches. Archival store for I/O is a major element in all computing systems. NSA is developing LOTS, an optical tape system with ten times the storage per tape and at least ten times the input data rate compared with conventional large stores. |
|
|
|
|
Thrust 3
|
Incorporation of technology into real applications -- FY 1998 accomplishments and FY 1999 plans |
|
High performance simulation of multiphysics problems |
NSF will support the simulation of physically realizable systems, giving physical scientists the ability to predict properties and function from fundamental structure. R&D in this area could lead, for example, to new lightweight, high strength biomimetic materials, or could enable the design of chemicals to immobilize viruses. This NSF-funded collaborative effort involves computer science and computational mathematics researchers at the University of Colorado-Boulder and the University of Colorado-Denver. The engineering activities are driven by the simulation of problems encountered in aerospace, civil, and mechanical engineering. The problem areas selected include at least two mechanical fields; a common basis in modeling, problem decomposition, and parallel simulation methods; a need for extensive scientific visualization; and the requirement for a common overarching software/hardware infrastructure. The research in computational mathematics will focus on the theoretical foundation of multiphysics simulation and the design of new algorithms in interface conditions, fast nonlinear solvers, and staggered multi-field timestepping. The computer science research will focus on the development of a Problem Solving Environment (PSE) for applications problems that is suitable for the parallel implementation of large scale multiphysics simulations. The primary goal is to provide transparent, portable, and scalable access to all of the resources needed for executing large parallel applications on heterogeneous and/or distributed systems, including interprocessor communication, visualization, debugging, and performance monitoring. |
|
Multi-model multi-domain computational methods |
A collaboration among Old Dominion University, University of Colorado-Boulder, Notre Dame University, New York University (NYU)-Courant Institute, Boeing Company, and Argonne National Laboratory, this NSF-funded project will develop a software laboratory for aerodynamics and acoustics problems through domain decomposition methods of task-parallel (involving multiple models, such as a viscous boundary layer and an inviscid free stream) and data-parallel (involving multiple subdomains, for additional concurrency) types. The project unifies several pairwise collaborations among numerical analysts, parallel algorithmicists, engineers, and computer scientists into a vertically integrated demonstration of a Grand Challenge computational philosophy; namely, that the solution to Grand Challenge problems is most feasible and cost-effective when software permits users to conveniently adapt the fidelity of the computational model and the strength of the local solver from region to region within an overall implicit algorithmic framework. |
|
Smart antennas |
The explosion of the cellular communications industry between 1995 and 1997 prompted communications companies to explore the use of antenna arrays at cellular base stations as a means to improve both the channel capacity and quality of wireless communications -- cellular telephony, personal communications services (PCS), and wireless local area networks -- by exploiting the enhanced interference rejection capability they provide. It is predicted that by the year 2005, all cellular and PCS base stations will employ some form of smart antenna technology, and many companies will demonstrate working systems. An example is ArrayComm's IntelliCell Base Station Array, which demonstrated uplink and downlink capacity improvements. NSF funded the development of this array and ultimately the creation of this company. ArrayComm provides hundreds of such arrays for Japan's personal communication systems. The world market for base station antennas for digital wireless applications is projected to grow at an unprecedented rate. Cellular carriers in the U.S. may need 15,000 new cell sites over the next decade to upgrade their services and meet anticipated demand. PCS services may require an additional 100,000 sites. NSF continues to fund work in this area at Brigham Young University, the University of Texas, and Purdue University. Researchers have built smart antenna systems to test uplink/downlink scenarios in real environments. Further experiments include channel measurements in troublesome environments, such as highly reverberant areas, in order to determine when and where fades occur, and to determine the means to combat them electronically. The goal is to uncover performance gaps between existing commercial systems and theoretical optimal performance and improve compression algorithms. |
|
Geometric shape analysis applied to molecular biology |
The "alpha shape" approach is used to describe geometric surfaces and volumes. In an NSF-funded multi-year computational geometry project at the University of Illinois, a molecular biology research group has applied the alpha-shape approach to study building blocks of light harvesting proteins. The approach is proving uniquely able to identify and simplify protein shapes without losing the topological characteristics of surfaces involved in aggregation. |
|
NASA advanced supercomputing applications |
NASA supports R&D in advanced supercomputing applications. Research areas include the following:
In these snapshots of simulated neutron star mergers, the lighter color
signifies higher matter density and the darker color, lower density.
Embodying some gravitational effects, simulations show that
gravitational radiation losses influence instabilities during the
merger. Only full relativistic calculations, now underway through
NASA-supported R&D, can predict the gravitational wave signal and
probe the possibility of resulting black holes.
|
|
Computational Aerosciences (CAS) |
CAS supports NASA aeronautics programs and is driven by the needs of the aeronautics industry. Its goal is to accelerate the development and availability of high-performance computing technology to the U.S. aerospace community, to facilitate the adoption and use of this technology by the U.S. aerospace industry, and to hasten the emergence of commercial hardware and software markets. The CAS project is developing computational techniques and applications to solve Grand Challenges, which include multidisciplinary simulations of complete aircraft throughout their flight envelope. These efforts will be structured with near-term, intermediate, and long-term milestones that, at each stage, provide "end-user" benefits. These milestones will be undertaken in close cooperation with academia and the aerospace industry. CAS works with the Numerical Aerodynamic Simulation (NAS) program and the NASA Research and Education Network (NREN) to ensure that the aerospace community has access to the NREN and the CAS testbeds. The CAS project is also encouraging joint development of new metrics that address affordability and cycle time. Research on high speed civil transport (airframe and engine) and high-performance aircraft (airframe and simple engine) is ongoing. 
NASA is involved in ongoing research on high speed civil transport and
high performance aircraft. Images generated by high performance
computing technology, such as the one pictured to the right, illustrate
the wake systems associated with each blade of an aircraft engine. Such
illustrations help researchers to better understand blade-vortex
interaction.
|
|
DOE Grand Challenge applications |
In support of its Grand Challenge applications, DOE R&D covers topics that include raw computational efficiency in quantum chromodynamics software, massive data requirements of relativistic heavy ion collider experiments, Accelerated Strategic Computing Initiative (ASCI) applications, and visualization for global change modeling. DOE also conducts R&D in collaborative tools and information surety techniques. In FY 1998, DOE will coordinate its DOE 2000, Grand Challenge, ASCI, and ACTS R&D. |
|
Molecular structure prediction and simulations |
In FY 1998, NIH organizations National Library of Medicine (NLM), National center for Research Resources (NCRR), Division of Computer Research and Technology (DCRT), National Cancer Institute (NCI), and National Institute of General Medical Sciences (NIGMS) will refine new methods for ab initio structure prediction for use in the pharmaceutical industry. The Institutes will improve computational technologies for larger simulations of protein, DNA, and membrane complexes in water environments. (Note: As of February, 1998, DCRT's activities became part of NIH's center for Information Technology -- [CIT].) NCRR will provide parallel computing applications in biochemistry, molecular biology, and cellular biology with access to supercomputing systems, large databases, and other resources through Web-based browsers. Thus, complex simulations related to receptor sites and other drug design R&D can be conducted over the Internet. NCRR will integrate 3-D graphics software, software tools for magnetic resonance spectroscopy data analysis and molecular structure determination, and other software tools to provide new capabilities for structure-based drug design. CIT will improve methods for predicting protein-drug binding energies, and will develop high performance computing methods for biomedical applications. NCI provides state of the art capabilities in a fully integrated high performance computing center, applies high performance parallel computing and communication methods to biomedical applications, and evaluates new scalable parallel architectures for biomedical applications. 
Coupling high performance computing systems with visualization tools
permits theoretical investigations of complex biomedical systems. The
figure shows a lipoprotein complex model with the lipids (green) in the
center surrounded by the proteins (blue and red). High-density
lipoprotein (HDL) circulates in the bloodstream, extracting cholesterol
from body tissues and transporting it to the liver for excretion or
recycling. Increased levels of HDL have been correlated with a decreased
risk of arteriosclerosis, a primary cause of cardiovascular disease.
NIH-supported researchers combined experimental evidence about the
structure of HDL particles with protein structure predictions to produce
a preliminary model of HDL. Predicting 3-D protein structure from
one-dimensional sequence data uses high performance parallel computing
techniques. Understanding complex biological functions requires
determining the structure and properties of large molecular complexes.
Pictured at the left is a conformation of prion protein as determined
through magnetic resonance spectroscopy. Prions are a novel class of
"infectious" pathogens distinct from viruses with respect to
both their structure and the neurodegenerative diseases that they cause.
Prion diseases are manifested as sporadic, inherited, and infectious
disorders including scrapie and bovine spongiform encephalopathy (mad
cow disease) of animals as well as kuru, Creutzfeldt-Jakob Disease
(CJD), and fatal familial insomnia of humans.
Prion protein (PrP) is the major, if not the only, component of prions. PrP exists in two isoforms: the normal cellular form (PrPC) shown here and the abnormal disease (scrapie)-related form (PrPSc). |
|
Reconstruction of 3-D positron emission tomography images |
In positron emission tomography (PET), images are reconstructed from a set of projected measurements in what is classically known as the inverse problem. While advances in detector technology have enabled researchers to use these techniques to study small animals, the data exhibit poor counting statistics. The quality of the reconstructed image can be improved by incorporating a statistical model of the scanning process into the reconstruction algorithm. In particular, reconstruction based on the maximum likelihood (ML) criterion show both reduced noise and improved resolution. These methods are being used to study the phenotypical consequences of genomic manipulation in mice and rats. The Nuclear Medicine Department at NIH is developing a high resolution small animal PET imaging system using new scintillation crystal technology. Image sensitivity is being improved both through acquiring 3-D data and using ML-based reconstruction. CIT has developed 3-D algorithms for these scanners. These algorithms exploit the inherent sparsity and symmetries in the matrix that models the scanner. The reconstruction software also decomposes the problem in parallel and takes advantage of high bandwidth I/O in CIT's IBM SP2 parallel computing system. Early visual stimulus studies revealed unprecedented cortical and subcortical activity in mouse and rat brains. CIT plans to complete this work in FY 1998. The Nuclear Medicine Department will attempt to achieve the once unattainable goal of 0.5-mm resolution with less than 10 percent coefficient of variation for this new-generation small animal scanning system. In FY 1999, CIT will continue developing methods and algorithms for biomedical applications that can benefit from computational speedup, such as image processing of electron micrographs, radiation treatment planning, medical imaging, and protein and nucleic acid sequence analysis. 
The image on the left consists of transverse sections of a 3-D positron
emission tomography (PET) image reconstructed using a maximum likelihood
(ML) algorithm. In this NIH study, a rat was injected with the
fluorine-18 analogue of glucose (18F-2-deoxy-D-glucose) and subjected to
visual stimuli. High-resolution cortical and subcortical activations are
visible. The images on the right are transverse, coronal, and sagittal
cross sections of a 3-D PET image reconstructed using an ML algorithm.
|
|
Image management and communications systems |
CIT, NCI, and the NIH Clinical center are implementing a prototype high-speed telemedicine network to support multimedia communication for medical research and education. The prototype will include two Asynchronous Transfer Mode (ATM) switches, with 16 ports each, to support multimedia communication at 155 Mb/s. Washington University in St. Louis has developed an ATM-based multimedia consultation environment that includes prototype medical workstations. This network environment will initially support high performance radiation therapy planning. |
|
Radiology Consultation Workstation (RCWS) |
RCWS is a specialized imaging workstation for real-time telemedicine consultation and for medical education. Designed to provide high-resolution display of medical images, the RCWS allows remote consultations between radiologists and other medical specialists over the ATM network described above. A shared cursor permits collaborators to use outlining to identify organs and lesions. The identified regions of interest are transmitted in real time to all other RCWS sites participating in the consultation to plan a radiotherapy treatment. The CIT IBM SP2 parallel computing system will be used for the computationally intensive radiation therapy planning calculations. RCWS goals include connecting the NIH prototype ATM telemedicine network to Washington, DC-area military medical centers and to Washington University in St. Louis. To this end, NIH is connected to the Advanced Technology Development Network (ATDNet) -- begun by DARPA and now funded by several Federal agencies -- that circles the Washington, DC metropolitan area. NIH will use NASA's Advanced Communication Technology Satellite (ACTS), which connects to ATDNet at the NASA Goddard Space Flight center (GSFC), and provides connectivity to Kansas City, MO. In FY 1998, additional RCWS nodes will be installed at Walter Reed Army Medical center and the National Naval Medical center, to allow multimedia telemedicine consultations between those institutions and NCI. Image communication software will be developed to allow the interchange of computerized tomography (CT) and magnetic resonance images (MRI) between the RCWS and select commercial radiotherapy treatment planning systems. In FY 1998, cardiology applications will be the second medical specialty for which the RCWS is adapted. |
|
Environmental protection |
The goal of EPA's program is to accelerate the evolution of HECC technologies to meet environmental protection mission objectives. EPA R&D in problem solving environments for multi-discipline environmental modeling builds on earlier achievements in technology development and integration for an air quality modeling framework. EPA HECC R&D develops scalable computational algorithms and integrates HECC technologies developed by other agencies to explore the feasibility of new paradigms for community problem solving environments and for multi-discipline, multi-pollutant, multi-scale environmental exposure assessment. EPA fosters interagency collaboration on development and testing of HECC technologies to provide a foundation upon which the scientific and HPC communities can build, component by component, complex high performance environmental management tools. A collaborative approach can leverage scientific and technology advancements of other Federal agencies, academia, and research institutions to more rapidly evolve a unified, comprehensive approach to multi-discipline environmental modeling. |
|
EPA's scalable software libraries |
EPA-supported projects develop scalable software libraries, cross-platform distributed data management and computing methods, multivariate visualization techniques including high performance geospatial analysis, and modular software for an integrated problem solving framework to simplify community development and use of complex multidiscipline ecosystem and human exposure models. The framework technology facilitates interdisciplinary coordination, data and code sharing, and rapid infusion of new technology and science into integrated environmental problem solving tools. In FY 1998, EPA will:
|
| Environmental modeling |
EPA's Models-3 is a flexible software system for developing and using
environmental assessment and decision support tools. The Community
Multiscale Air Quality model has been integrated into Models-3 for urban
to regional scale air quality simulation of ground level ozone, acid
deposition, visibility, and fine particulates. Components of Models-3
assist in design and preparation of source emission inventories
compatible with a variety of air quality modeling capabilities. Models-3
is a community framework for continual advancement and use of
environmental assessment tools.
The Models-3 framework provides an interface between the user and operational models, between the scientist and developing models, and between the hardware and software, thus enhancing the user's ability to perform a wide range of environmental management tasks, from regulatory and policy analysis to understanding the interactions of atmospheric chemistry and physics. The framework uses specialized object libraries and a layered design that isolates critical system components, thus minimizing the impact of hardware and software upgrades. A client-server architecture in conjunction with a standardized data interface and object-oriented database containing metadata enables transparent use of scalable computing platforms and access to data across the network. The object data base also contains application-specific globally shared data, such as model domain, map projections, grid resolution, and chemical species, that enable the interchange of science codes while maintaining user control of model specifics. A library-based graphical user interface facilitates easy model executions and access to a variety of visualization and analysis packages. In FY 1999, EPA will:
|
| Aquifer modeling |
In FY 1998, new developments in analytical modeling are extending
regional scale representations of porous media aquifers. This EPA-funded
work is based on the superposition of many closed form analytic
solutions representing hydrologic features, such as point sinks for
wells, line-sinks for rivers, and areal elements for variable recharge.
The solutions contain degrees of freedom that allow them to be chosen
and combined so that boundary conditions are met along certain internal
boundaries. Cooperative research with Indiana University exploring the
implementation of 3-D well elements in parallel processing computers has
resulted in public domain analytic element research code (ModAEM)
available for use on scalable parallel computers and personal computers.
The University of Minnesota is implementing variable density flow into
analytic element representations of coastal aquifers. A case study is
simulating the fresh water to salt water transition in the Upper
Chesapeake aquifer.
New developments in EPA-supported analytical modeling are extending
regional scale representations of porous media aquifers. The model to
the left illustrates axisymmetric flow to a partially penetrating well
in a stratified aquifer.
|
|
Large scale environmental flow and transport |
EPA-funded University of Kentucky researchers have simulated turbulent flows by combining the least-squares finite element method (LSFEM) with large eddy simulation (LES). This has been demonstrated in 3-D shear flow problems and natural convection problems, and has reduced the single-processor computational burden by 75 percent for large shear flow problems containing up to one million finite elements with over 7 million unknowns. This robust method enables more efficient solution of larger fluid flow and transport problems in turbulence prediction and control, meteorology, climatic changes, air pollution, water pollution, and soil contamination. |
|
Online reference data for computational science |
In FY 1998, NIST will evolve mathematical software repositories into problem-solving environments, complete object-oriented libraries for basic linear algebra and related capabilities, and demonstrate capabilities on distributed systems. NIST continues to develop network-based resources that will bring more of the process of doing science online. For example, the Matrix Market (http://math.nist.gov/MatrixMarket/) provides researchers and software developers with a collection of large sparse matrices from industrial applications that can be used in comparative studies of linear algebra algorithms and software. Such problems form the core of many large scale applications. NIST is revising and updating the classic NBS (National Bureau of Standards, NIST's predecessor) Handbook of Mathematical Functions to be the basis for an interactive online library of mathematical reference data. It will include extensive formulas, graphics, and tables suitable for downloading into computer algebra systems and word processors. |
|
High performance linear algebra software |
NIST continues work in software design strategies that promote usability and maintainability, as well as high performance, for core numerical linear algebra operations. For example, NIST is working with the Basic Linear Algebra Subprograms (BLAS) Technical Forum, an industrial/government/academic consortium, to specify a standard interface to sparse matrix kernels. Such standards permit manufacturers and software developers to develop specially tuned high performance kernels with a portable interface. NIST is also demonstrating the use of object-oriented software design strategies in developing numerical toolkits in C++ and Java. |
|
Algorithm development at NOAA |
In FY 1998 NOAA plans to continue algorithm development on a scalable system to achieve five- to ten-kilometer resolution in mesoscale atmospheric models. The agency will explore the design of next-generation environmental observing systems to test data assimilation needs for optimizing future forecast systems and will develop software tools to facilitate conversion from traditional shared-memory machines to scalable systems. NOAA will enhance scientific experiments running on high performance computing systems at its Princeton, NJ, Geophysical Fluid Dynamics Laboratory (GFDL), and will evaluate the performance of the ETA model at various grid resolutions and assess its potential for operational forecasting. |
|
Advancing the science of weather forecasting |
The Nation's meteorological observations are a billion-dollar resource. Because it has not been possible to design the observing system from a whole-system view, the availability of new technology has driven observing systems development. The goal of NOAA's North American Atmospheric Observing System (NAOS) Program is to exploit existing observational resources more fully and guide the development of future observing systems more purposefully. The tests and evaluations of meteorological observing systems that NAOS advocates require high performance computing hardware and techniques because both incorporating observations into prediction models and running the prediction models themselves are computationally demanding. In order to enable more rapid development of the models used in the NAOS program, NOAA's Forecast Systems Laboratory (FSL) in Boulder, CO, has developed the Scalable Modeling System (SMS) software toolset. Built on the industry-standard Message Passing Interface (MPI), SMS is designed to ease the transfer of existing numerical weather prediction models to highly parallel computing systems. A key feature is that models parallelized using SMS are portable to a wide variety of high performance computing systems. In FY 1999, FSL plans to acquire a high performance computing system, continue enhancing its national weather prediction (NWP) models to take advantage of the new hardware, and use the new hardware and higher resolution models to conduct NAOS experiments. |
|
Geophysical Fluid Dynamics Laboratory (GFDL) |
The strongest hurricanes experienced on Earth may be upstaged by even more intense hurricanes over the next century as the Earth's climate is warmed by increasing levels of greenhouse gases in the atmosphere. Most hurricanes do not reach their maximum potential intensity before weakening over land or cooler ocean regions. However, those storms that do approach their upper-limit intensity are expected to be stronger in the warmer climate due to the higher sea surface temperatures. According to a new simulation by GFDL scientists, a 5 to 12 percent increase in wind speeds for the strongest hurricanes is projected if tropical sea surfaces warm by a little over 2°C. Although such an increase in the upper-limit intensity of hurricanes due to global warming was suggested on theoretical grounds a decade ago, this investigation is the first to examine the question using a hurricane prediction model that is being used operationally to simulate realistic hurricane structures. GFDL simulated samples of hurricanes from both the present-day climate and from a greenhouse-gas warmed climate by linking information from its global climate model into its high-resolution hurricane prediction model. The latter has been successfully used by NOAA's National centers for Environmental Prediction to predict tropical storm tracks for the last several hurricane seasons. The climate model is a leading model used by climate researchers to project possible effects of greenhouse gases on future climate. These accomplishments illustrate how research in two distinct areas that require high performance computing -- climate change and hurricane prediction -- can be combined to provide new information about the potential impact of global climate change upon future weather systems. 
NOAA simulated samples of hurricanes from both the present-day climate
and from a greenhouse-gas warmed climate.
|
|
|
|
|
Thrust 4 |
Infrastructure for research in HECC -- FY 1998 accomplishments and FY 1999 plans |
|
Partnerships for Advanced Computational Infrastructure (PACI) |
In FY 1998, NSF initiated the Partnerships for Advanced Computational Infrastructure (PACI) to provide access to high performance computing for the academic research community at a performance level two orders of magnitude greater than that available at a typical major research university. (See special section on PACI.) The National Energy Research Scientific Computing center (NERSC) supports the computational needs of DOE's energy science investigators. DOE supports High Performance Computing Resource Providers (HPCRPs) at Argonne, Los Alamos and Oak Ridge National Laboratories. Systems at these sites include Intel Paragon, IBM SP2, Cray T3E and C90, and SGI Origin 2000. DOE's Energy Sciences Network (ESnet) is part of the initial national computational grid. In FY 1998, DOE will evaluate the success of the HPCRPs, including productivity and usability. |
|
DOE High Performance Computing Resource Providers (HPCRPs) |
In FY 1998, NSF initiated the Partnerships for Advanced Computational Infrastructure (PACI) to provide access to high performance computing for the academic research community at a performance level two orders of magnitude greater than that available at a typical major research university. (See special section on PACI.) The National Energy Research Scientific Computing center (NERSC) supports the computational needs of DOE's energy science investigators. DOE supports High Performance Computing Resource Providers (HPCRPs) at Argonne, Los Alamos and Oak Ridge National Laboratories. Systems at these sites include Intel Paragon, IBM SP2, Cray T3E and C90, and SGI Origin 2000. DOE's Energy Sciences Network (ESnet) is part of the initial national computational grid. In FY 1998, DOE will evaluate the success of the HPCRPs, including productivity and usability. |
|
DOE Advanced Computational Testing and Research (ACTS) Toolkit |
FY 1998 ACTS research includes Sandia National Laboratory's computational biology work that focuses on developing and implementing algorithms for genome sequencing and protein folding. FY 1999 plans include prototype deployment of software tools designed around ACTS interface definitions. |
|
DOE's Academic Strategic Alliances Program (ASAP) |
DOE's Academic Strategic Alliances Program (ASAP), part of DOE's ASCI program, works with the CIC R&D program sin areas of common interest. Currently, ASAP works with U.S. universities to advance its high performance simulation capabilities:
|
|
|
|
|
|
|
