Information Technology Frontiers for a New Millenium High End Computing and Computation

	Introduction High end architectures High end hardware components Foundational and algorithm research Software High end applications
Introduction	HECC research and development (R&D) provides the foundation for U.S. leadership in high end computing, promoting its use in government, academia, and industry. HECC researchers are developing computation-intensive algorithms and software for modeling and simulating complex physical, chemical, and biological systems; information-intensive science and engineering applications; and advanced concepts in quantum, biological, and optical computing.   The HECC Working Group (HECCWG) coordinates Federal R&D dedicated to maintaining and expanding U.S. leadership in high performance computing and computation, which includes systems architecture, hardware, foundational and algorithmic research, software, and high end mission-oriented applications. The HECCWG also promotes Federal cooperation in high end computing and computation R&D among Government laboratories, academia, and industry.   This section describes the FY 1999 accomplishments and FY 2000 plans in HECC R&D.
High end architectures	High end architectures FY 1999 accomplishments and FY 2000 plans
Hybrid Technology Multithreaded (HTMT) Architecture	Researchers funded by DARPA, NASA, and the National Security Agency (NSA) are evaluating the feasibility of constructing a computing system capable of a sustained rate of 1015 floating point operations per second (one petaflop). Preliminary evaluations of conventional architectures and mainstream technologies indicate that such a system will necessitate a radical departure from current research. HTMT architecture would blend modified semiconductor technology with leading-edge hybrid technologies including superconducting technology, optical interconnects, high speed Very Large Scale Integration (VLSI) semiconductors, and magnetic storage technology configured to satisfy the architecture requirements. Fundamental drivers are multi-gigahertz speeds, exceptional bandwidths, and very large and cost-efficient memory size. If results from FY 1999 R&D are encouraging, researchers are expected to build major sections of a prototype beginning in FY 2000.     Artist's 3-D depiction of a prototype computing system employing HTMT architecture. DARPA, NASA, and NSA researchers are evaluating the feasibility of constructing a hybrid HTMT system that would be capable of a sustained rate of 1015 floating point operations per second (one petaflop).
Quantum computing	Current trends in miniaturizing computer components suggest that they will reach the quantum scale by 2020. Today's computers count in binary digits - zeroes and ones. A theoretical quantum computer (QC) counts in "qubits," which are superpositions of zeroes and ones that might be represented, for example, by the direction of a spinning electron. Quantum computers will be able to perform nonclassical logic operations and use the phenomenology of quantum parallelism to enable the solution of problems that are intractable with even the most powerful conventional parallel supercomputers that can be envisioned today.   Supported by NSA, more than nine universities and companies are conducting QC research. Projected accomplishments include: Developing the quantum analogues of classical information and communication theory to help in future QC design and evaluation Applying the "hidden subgroup" problem to the design of new quantum algorithms Studying the computational power achievable in models positioned between classical and quantum computing Verifying a new pulsed-laser technique to implement quantum logic gates between photons Creating entanglements of three photons Achieving coherent quantum state manipulations of a few ions Uncovering clear evidence of quantum superposition, entanglement, and multi-qubit operations Characterizing the resulting quantum coherence Future QC research includes demonstrating 1-qubit operations and performing experiments to achieve 2-qubit operations using the optical-lattice method of trapping atoms and simulating the dynamics of a set of qubits in finer detail than has previously been achieved.   The initial five-year phase of NSA's quantum computing project ends in FY 1999. Full-scale development of a practical quantum computer may require 20 or more years of research.
Quantum teleportation	DOE's Oak Ridge National Laboratory (ORNL) has established a state-of-the-art laboratory for quantum teleportation, one research area in the field of quantum computing and communications. Quantum theory indicates that one can instantaneously teleport a quantum state over arbitrary distances, but one can use it only after a subsequent "classical" communication is carried out. To attempt to understand what is actually teleported, and what can and cannot be achieved supraluminally, ORNL is conducting experiments in which a signal transmitter and the transmitted signal are quantum mechanically entangled. High power femtosecond lasers are used for the experiments.
Commodity clusters: high performance computing on desktop systems	The high performance virtual machine (HPVM) software project, supported by NSF, allows off-the-shelf desktop computers to be used as high performance clusters. Researchers at the NSF-funded National Computational Science Alliance (NCSA) and National Partnership for Advanced Computational Infrastructure (NPACI) have built a Windows NT supercluster consisting of hundreds of workstations running on Intel Pentium II processors, and have run the complex astrophysical hydrodynamics code, ZEUS-MP, at the rate of several gigaflops. HPVM allows users to easily port their Message Passing Interface (MPI) codes to the NT cluster environment, providing a low-cost alternative to conventional supercomputing.
Evaluation of high end architectures	ORNL is evaluating new high end architectures to assess their applicability for DOE and other Government applications. One of these architectures is the SRC-6, which combines commodity microprocessor technology with a new high performance Memory Algorithm Processor (MAP) subsystem, a programmable hardware component that allows programmers new flexibility to customize the hardware for specific applications. ORNL is evaluating the MAP prototype subsystem and plans to take delivery of an SRC-6.   NPACI, along with DOE's National Energy Research Supercomputer Center (NERSC), Cal Tech, and Boeing Computer Services are also evaluating the Tera Multi-Threaded Architecture (MTA), located at the San Diego Supercomputer Center (SDSC). This system seeks to achieve high degrees of parallelism by splitting a program into hundreds of individual execution threads and having the system hardware execute whichever thread is ready to progress.
High end hardware components	High end hardware components FY 1999 accomplishments and FY 2000 plans
Automating component design	NSF supports basic research in Electronic Design Automation (EDA) and in applicable VLSI design technologies, such as systems-on-a-chip, embedded systems, and multi-technology optical and micro-electromechanical design methods. Research areas include scientific methods, intellectual processes, abstractions, search paradigms, and information models used in VLSI design, covering all phases of theEDA design cycle for integrated circuits and systems from conception through manufacturing and testing. Other areas of research include: Theoretical foundations models (algorithms, tools, analysis, synthesis, simulation, validation, and verification) System design methods (systems-on-a-chip, multi-chip, and multi-technology systems) Manufacturing (fault models and algorithms for diagnosis and test in digital, analog, and mixed signal designs) Design and system prototyping methods, tools, and environments, including information infrastructure
Diamond-based Multi-Chip Modules (MCMs)	NSA's MARQUISE/SOLITAIRE program repackages a high performance computer using diamond-based MCMs and thin film spray cooling. The SOLITAIRE prototype will demonstrate the maturity of high density packaging technologies incorporating die-last MCMs, diamond substrates, and spray cooling. Environmental testing of the integrated system will demonstrate its reliability.
Micro spray cooling	NSA's spray cooled power converter research program is developing a high density, high efficiency power converter using spray cooling and planar laminated transformers. Current research focuses on applying isothermal spray cooling to overcome the fundamental coefficient of thermal expansion mismatch encountered during hybrid integration of silicon or gallium arsenide (GaAs) devices with diamond substrates.     Pictured is the diamond Multi-Chip Module (MCM) aerosol spray cooled Cray J90 supercomputer mounted vertically in the black "clamshell enclosure" on the left. One redundant aerosol spray pump/heat exchanger is visible as a silver cylinder on the lower right. The pump controller-fail safe electronics are visible in the upper right. Mounted on the far left wall of the integration platform are the internal power supplies for the pump control electronics.
Superconductive crossbar switch	Research suggests that superconducting supercomputers can deliver very high performance with very low power requirements. NSA supports research in superconducting electronics to provide high performance computing alternatives to current silicon and GaAs technologies, which have speed and power limitations. NSA is constructing a 128 x 128 superconductive crossbar switch that operates at 2.5 gigabits per second (Gbps) per port for use in supercomputing and network applications. It is self-routing, is scalable in size from 32 x 32 to more than 1024 x 1024, and has a latency of less than 4 nanoseconds.   Although the crossbar electronics operate at 4 degrees Kelvin, its input and output ports are at room temperature, and the cryogenic elements are cooled by a refrigerator, allowing use in a standard room temperature environment. Extended to higher speed and size, this switch is a candidate element for use in HTMT architectures.
Optoelectronic research	NSA supports research on optical techniques that can theoretically support data rates of several hundred Gbps for future communications and computing systems by providing logical functions that expedite the data routing process. Transmission at high data rates demands techniques that can restore data signals so that low bit error rates can be maintained without compromising signal quality. NSA is also conducting research on a high performance spectrometer on a chip and a semiconductor optical amplifier.
Smart memories	NSA supports research to produce a flexible computing architecture more power-efficient than the current evolutionary path of reduced instruction set chip (RISC) architectures while remaining programmable in a high level language. This architecture will be able to reconfigure itself to optimize for currently executing computations in order to accommodate a wide class of coarse- to fine-grained algorithms.
High Performance Storage System (HPSS)	A consortium of DOE national laboratories, in collaboration with International Business Machines (IBM), has developed the High Performance Storage System (HPSS). With more than two years of operation and deployment at about twenty sites, HPSS has become a standard for storage systems in the high performance computing community. HPSS 4.1 contains new features such as support for the Distributed File System (DFS), file families, Message Passing Interface-Input/Output (MPI-IO), scalable accounting, and performance improvements. In FY 1999, Sun Microsystems and Storage Technology Corporation joined the HPSS community, where research continues on large database storage and transfer.
Advanced scalable computing for weather forecasting	Building on HPCC advances, NOAA will lease a new IBM high performance scalable computer that will significantly improve weather, flood, and climate forecasting for the country. Expanded NOAA forecast models are computationally demanding and require high reliability. This new high performance computing system will use a highly parallel computer architecture to run models with improved physics and greater resolution, producing forecasts with better accuracy and allowing NOAA to operate more sophisticated models of the atmosphere and oceans.
Foundational and algorithm research	Foundational and algorithm research FY 1999 accomplishments and FY 2000 plans
Theory of computing	NSF supports fundamental research in computing theory in three areas: Application-specific theory supports developing models and techniques for solving problems that arise in areas of science and engineering such as molecular biology, communications networks, and computational linguistics. Of particular interest are theoretical developments having potential impact on experimental or applied areas of computer science research, and strategies that mix theory with experimentation. Core theory covers computational complexity, cryptography, interactive computation, computational learning, parallel and distributed computation, computation on random data, online computation, and reasoning about knowledge. Fundamental algorithms include developing combinatorial, approximation, parallel, online, numerical, geometric, and graph algorithms that span application domains.
Numeric, symbolic, and geometric computation	NSF supports fundamental research in computation and graphics, including computer algebra, numerical computation and modeling of physical processes, mathematical optimization, computational geometry, imaging, deductive methods for reasoning in computational logics, and automated deduction. These areas combine mathematical analysis with advanced algorithms. NSF also supports integrating research results into problem-solving environments to support computational science and engineering. Supported research addresses generic as well as scientific and engineering discipline-specific computation, including innovative applications of advanced computational and graphic techniques in scientific and engineering systems such as manufacturing and design, proof support systems, prototyping, and design verification.
Whisker Weaving (WW)	Research in mathematical and computational sciences at DOE's Sandia National Laboratories (SNL) is currently focused on WW -- an algorithm for generating 3-D unstructured hexahedral meshes -- and mesh optimization. Most hexahedral meshing algorithms work on only a small class of geometries, restrict the surface mesh, and require extensive user input. WW technology is designed to eliminate these shortcomings and could significantly reduce the time needed to generate hexahedral meshes for complex geometries. As time to mesh is currently the most serious impediment to developing the automated design capability of DOE's Accelerated Strategic Computing Initiative (ASCI) and other major programs, the potential impact of both WW and mesh optimization is substantial.
NSF research in parallel and distributed computing	NSF's FY 1999 HPCC research includes parallel computation models, parallel algorithms and software for scientific computing, dynamic compilation and optimizing parallel compilation, distributed operating systems, superscalar architectures, high performance memory systems, and signal processing and communications systems research, especially in wireless information technology and networks.   NSF FY 2000 research plans include a wireless center, hardware/software co-design, and high performance scientific and commercial applications.
Global optimization	The aim of SNL's global optimization project is to develop new, innovative methods representing better algorithmic trade-offs between the ability to find a global optimum and the ability to generate near-optimal solutions quickly. The algorithmic development will focus on domain-independent methods applicable to a wide range of applications. SNL will develop a global optimization method on applications with continuous and combinatorial search domains to provide a practical evaluation of the analyses. This project has recently begun to focus on practical folding methods, including methods for off-lattice models.
Large-scale scientific and engineering design optimization	A result of years of progress in modeling and simulating large structural mechanics, thermal analysis, and fluid dynamics problems is that these codes can now be used to design systems and develop predictive computational models for these systems. This in turn has created a need for new optimization algorithms to overcome difficulties such as noisy and expensive function evaluations and lack of derivative information. To address this need, SNL has developed an object-oriented library, OPT++, for nonlinear optimization. The library includes a wide variety of methods available for both unconstrained and bound constrained optimization, large scale constrained optimization algorithms for computational chemistry problems, and new methods for rapid, robust parallel optimization.
Data mining	Data mining is the process of automatically extracting large amounts of valid, coherent data from large and sometimes confusing databases. Data mining is becoming a vital process to large corporations that often store many years of data that has been compiled but never analyzed. There are many different techniques used for data mining, and often the technique that is chosen depends on the type of data and information that is to be extracted from the database. Some of the most popular techniques are association, sequence-based analysis, clustering, classification, estimation, fuzzy logic, genetic algorithms, fractal-based transforms, and neural networks.   Among the best of NCSA's data mining tools are machine learning algorithms, so-called because they are primarily adaptive systems that can learn on the job. These algorithms rely on an iterative process through which the programs learn from examples. Machine learning algorithms can be used for tactical and strategic decision making.   The ability to analyze very large data collections -- defined as billions of objects -- is needed to support research in such areas as astronomical digital sky surveys and high energy physics. The challenge is to identify a accurate subset of data that is represented by a large number of attributes. Scalable statistical algorithms for clustering data are being developed within NPACI at the University of California-Davis.
Software	Software FY 1999 accomplishments and FY 2000 plans     This 3-D shaded relief representation of a portion of Pennsylvania uses color to show maximum daily temperature. Displaying multiple data sets at once, and interactively changing the display, helps users quickly and intuitively explore their data, either to formulate or confirm hypotheses.   NPACI researchers have developed software to extract the grey-white matter cortical surface from high resolution, 3-D human brain data obtained from UCLA. Dynamic programming methods were used to define the maximal contour of the grey-white surface from a quarter-resolution version of the UCLA data. Processing of the full-resolution data will require NPACI infrastructure.
Next Generation Software (NGS) program	NSF's NGS program fosters research to develop performance engineering technologies for the design, management, and control of computing systems; and to create a new system software architecture to provide runtime support for complex applications executing on complex computing platforms, such as computational grids and future petaflop platforms.   NGS views the computing system as having several layers: the application layer, grid platform architecture, the system software, processing nodes, and interconnect layers. NGS research will lead to software systems that understand the interrelation among these layers as it affects the behavior of a computing system, and will guide the design, management, operation, and control of the system.
Operating systems and compilers	NSF research in operating systems and distributed systems includes developing mechanisms and applications programming interfaces (APIs) for uniform access and management of resources in local area networks (LANs) and wide area networks (WANs); middleware infrastructure for building scalable services; resource management for new applications and quality of service requirements; and security and electronic commerce. Research in compilers and runtime systems includes dynamic compilation; models of storage consistency and storage-hierarchy performance; and compiler support for programming on the World Wide Web.
Programming methods and languages	NSA supports research in computational methods and languages for massively parallel, distributed heterogeneous computing platforms and special-purpose processors. Research includes developing the compiler AC -- that targets the Silicon Graphics/Cray Research (SGI/Cray) T3D and T3E architectures, among others -- and implementing the "futures" model of message-passing programming.
Software engineering and languages	NSF supports fundamental research to develop quality software-based systems, including domain-specific languages for specification and design; constructive approaches to software design and evolution; issues of software modularity and composition; enhancement of confidence and quality; automating stages of software development; distributed and network environment issues, including distributed development and software security; and formal foundations for all aspects of software engineering and programming languages.
Java numerics	The rapid and widespread adoption of the Java language and environment for network-based computing has created a demand for reliable and reusable numerical software components to support the development of scientific software. The National Institute of Standards and Technology (NIST) works with the Java Grande Forum to enable the use of Java in high performance numerical computing.
Aztec Iterative Solver	SNL is conducting R&D on Aztec, a library of state-of-the-art iterative methods for solving linear systems. Its goal is to perform well on parallel computers while being easy for application engineers to use. Aztec's simplicity comes from employing a global distributed matrix that allows a user to specify pieces (different rows for different processors) of an application matrix exactly as would be done in the serial setting (that is, using a global numbering scheme). Speed comes from using standard distributed memory techniques, including having a transformation function compute local addresses, ghost variables, and message information, increasing the speed of calculations and communication of data dependencies. Aztec takes advantage of advanced partitioning techniques and uses efficient dense matrix algorithms when solving block sparse matrices.   In applications, Aztec performs a critical task requiring a significant portion of the total simulation time. Aztec has facilitated the work of application engineers by providing them with leading-edge iterative methods and by freeing them from cumbersome programming tasks associated with parallel iterative methods. In addition to internal use, including DOE's Accelerated Strategic Computing Initiative (ASCI) applications, there are more than 100 licensed Aztec external users worldwide.
Distributed memory software for ocean modeling	NOAA's new Modular Ocean Model, version 3 (MOM 3) improves model physics and for the first time can be run on distributed-memory supercomputers. Tests using message passing have been done on the SGI/Cray T3E-900 and T90 supercomputers at NOAA's Geophysical Fluid Dynamics Laboratory (GFDL). The baroclinic and tracer portions of the model achieve near linear scaling. Continuing areas of research include better scaling of the barotropic equation and developing parallel techniques for outputting data efficiently during model execution. The latter issue, being addressed jointly with the NERSC, will also lead to high resolution modeling of the southern ocean.
New fill-reducing ordering algorithms	DARPA-funded researchers at ORNL are working with Xerox Palo Alto Research Center (PARC) and the University of Tennessee to develop a family of modular parallel sparse matrix solvers based on preconditioning iterative methods by combining direct sparse solvers with inherent iterative parallelism and scalability. This will enable defense and industry engineers to use parallel machines to solve large problems effectively, without writing complicated domain-specific parallel linear system solvers.
Scalable Unstructured Mesh Algorithms and Applications (SUMAA3d)	SUMAA3d is a suite of software tools for unstructured mesh computations on distributed-memory computers that emphasizes scalability on widely varying architectures ranging from tightly coupled massively parallel processors (MPPs) to networks of workstations, techniques to achieve accurate solutions efficiently, and usability and interoperability. Applications include modeling piezoelectric crystals, high temperature superconductors, and cracked pressure vessels. The SUMAA3d project is a collaboration among researchers at DOE's Argonne National Laboratory (ANL), the Pennsylvania State University, the University of British Columbia, and Virginia Tech.   New SUMAA3d algorithms and software include: Automatic mesh generation tools for unstructured meshes with mixed element types Mesh optimization procedures to improve mesh quality Adaptive mesh refinement/de-refinement using edge bisection techniques to model rapidly changing solutions and multiscale geometries accurately Dynamic mesh partitioning/repartitioning after mesh modification Recent efforts have focused on interoperability with the Advanced Large-scale Integrated Computational Environment (ALICE) (page 21) for solving large scale partial differentiatial equation (PDE) based applications. Current efforts focus on interfacing SUMAA3d with solver and visualization components and using the knowledge gained to develop design specifications for a general mesh component. Efficient interfaces between SUMAA3d and both the Portable Extensible Toolkit for Scientific computation (PETSc) (page 18) and BlockSolve95 have been completed.
Dynamic load balancing and data migration for adaptive numerical methods	Adaptive numerical methods that automatically refine or coarsen meshes and/or vary the order of accuracy of a method offer greater reliability, robustness, and computational efficiency than traditional approaches for solving partial differential equations. On parallel computing systems, however, adaptive algorithms introduce complications because they decompose and redistribute dynamic rather than static data structures. SNL is investing in hierarchical adaptive mesh refinement of unstructured grids for the application codes MPSalsa, ALEGRA, and SIERRA. Each of these projects uses quadtree/octree data structures to represent the refined meshes. That is, during refinement, elements are split into smaller elements that are stored as "children" of the initial "parent" elements in a hierarchical data structure. Quadtrees and octrees are widely used in both mesh generation and adaptive refinement on serial computers. Using local load-balancing techniques, quadtrees for 2-D adaptive, structured-mesh methods with local time stepping were successfully balanced. Issues that must be resolved for more general 3-D unstructured grids, steady-state problems, and method-of-lines time-stepping include: Designing efficient dynamic load-balancing methods that account for data's current location to reduce migration costs Developing load-balancing heuristics for hierarchical meshes Efficiently distributing octree data structures on parallel computers Incorporating accurate models and measurements of processor work loads and migration costs into load-balancing algorithms To address these and related issues, a dynamic load balancing tool called Zoltan is being developed for use in a range of applications. Adaptive PDE solvers are the most significant target; but the tool will also be used to support particle calculations, contact detection in crash simulations, and other dynamic and adaptive codes. In FY 2000, additional load-balancing algorithms, support for heterogeneous computers, tools for data migration, and other enhancements will be added.
PETSc, The Portable, Extensible Toolkit for Scientific Computation	PETSc is a suite of interoperable software components for the large scale simulation of physical phenomena modeled by partial differential equations. PETSc includes an expanding suite of data structures and numerical kernels for use in application codes written in C, C++, and Fortran. PETSc is relatively easy for beginners to use, yet allows advanced users detailed control over solutions.   PETSc 2.0 uses an MPI for all message-passing communication, enabling portability from networks of workstations through large scale parallel processors. The PETSc programming model is also the most appropriate for non-uniform memory access (NUMA) shared-memory machines, which require the same attention to memory hierarchies as distributed-memory machines.   ANL-led multisite computational science projects built around the PETSc software include research on multi-model multi-domain computational methods in aerodynamics and acoustics (an NSF Multidisciplinary Challenges Program) conducted jointly with Old Dominion University, Courant Institute, University of Colorado at Boulder, University of Notre Dame, and Boeing Computer Services.
Scalable tools for weather modeling	NOAA's Forecast Systems Laboratory (FSL) has published advanced components of its Scalable Modeling System (SMS) of tools for developing and porting geophysical fluid dynamics models on parallel computing architectures. SMS is available as experimental public domain software.     Los Alamos National Laboratory's (LANL's) POPTEX tool provides interactive visualization capabilities for the DOE Global Climate Modeling Grand Challenge. Historically, global climate data has been viewed using video technology. While useful for viewing the progress of the simulation, however, videos cannot be modified without creating a new video. LANL's goal was to provide the benefit of video visualizations (putting the results of the simulation into motion) while adding capabilities that enable dynamic, flexible, and interactive exploration of scientific data. The figure shows a POPTEX result, the temperature of the ocean with low values at the blue end of the spectrum and high values at the red end.
Parallel I/O for ocean and weather data	Improving single-processor performance is more difficult for the full-featured, 3-D fluid dynamics applications that are typical of the simulation and prediction models of climate and weather research. Better compilers and tools and more on-site technical expertise are needed to address this system limitation for NOAA's data-intensive applications -- a challenge as important as good scalability in achieving full potential in these systems. Scientists at NOAA's GFDL are collaborating with DOE scientists at the Lawrence Berkeley National Laboratory (LBNL) to develop a parallel I/O implementation of the netCDF Common Data Format, which is widely used in the oceanographic and meteorological research communities.
Information based computing	A growing need within scientific disciplines is support for information discovery and data handling. Supercomputer data management technologies that automate discovery and retrieval are being deployed in the NPACI Data Intensive Computing Environment. The infrastructure is composed of archival storage systems (HPSS), data movement systems (SDSC Storage Resource Broker), meta-data catalogs (SDSC MCAT), digital libraries for data discovery services, and information mediators for data presentation. The resulting infrastructure supports the publication of scientific data sets, making it feasible to integrate scientific data collections with traditional text and image digital libraries.
APOALA	The Environmental Protection Agency (EPA) -supported APOALA project at Penn State is developing the ability to integrate a temporal geographic information/visualization environment in order to analyze complex, large scale environmental processes. The prototype will integrate a new database model for space-time data, a visual space-time query-building language, parallelism for efficient retrieval and manipulation of data, visualization that facilitates traditional and exploratory analysis of time sequences of 3-D data, and data exploration and knowledge discovery capabilities that are closely coupled with space-time visualization methods for Earth data.
The Globus project	Supported by DARPA, DOE, and NSF, the Globus project is developing fundamental technologies needed to build computational grids and execution environments that enable applications to integrate tens or hundreds of geographically-distributed computational and information resources, instruments, and displays. Globus has evolved out of the I-Soft software environment developed for the I-WAY networking experiment demonstrated at SC95 in San Jose, California, and is currently running on machines on every continent in the world except Antarctica.   The Globus team, from the Information Sciences Institute at the University of Southern California (USC) and ANL, won the High Performance Computing (HPC) Challenge Award at SC98 for their demonstration of wide area applications. The team showcased the Globus metacomputing toolkit and the associated Globus Ubiquitous Supercomputing Testbed Organization (GUSTO) testbed, the first large scale realization of a high performance distributed computing infrastructure, by performing three unique computations: Collaborative online analysis of data from a microtomographic beamline at ANL's Advanced Photon Source (APS) Record-setting distributed interactive simulation using multiple supercomputers High-throughput computing for crystallographic phase problems     Volume rendered view (lower left) of the 3-D structure of a meteorite determined by computed microtomography. This image was produced as part of the DOE Grand Challenge project "Supercomputer Solution of Massive Crystallographic and Microtomographic Structural Problems," which is using components of the Globus toolkit, including the Nexus communication library, to enable quasi-realtime reconstruction of microtomography data from the Advanced Photon Source. Examples of imaging diesel valves using DOE's ACTS toolkit appear above. The image at the right shows pressure and velocity around a moving valve in a diesel engine. The flow here was found as part of a computational fluid dynamics (CFD) effort to simulate the flow within the complex 3-D geometry of a diesel engine. The computation was carried out using the Overture Framework and the PADRE library for parallel data distribution.
ACTS Toolkit and Globus: Controlling remote instruments	The coupling of a scientific instrument with remote supercomputers and visualization devices can transform instrument capabilities by allowing quasi-realtime (taking minutes rather than hours) and remote analysis, imaging, and control. DOE's Advanced Computational Testing and Simulation (ACTS) Toolkit combines with the Globus toolkit to achieve such a coupling. Brilliant Xrays, now available at facilities such as ANL's Advanced Photon Source (APS), make it feasible to record high resolution 3-D tomographic data on time scales of less than a second per image. This technology is used to image and simulate biological and archaeological data, providing high performance parallel implementations of reconstruction algorithms for microtomographic datasets by means of filtered back projection techniques. Globus software moves data efficiently among detector, secondary storage, supercomputer, and remote users. An interactive analysis system operates on the ImmersaDesk and Cave Automatic Virtual Environment (CAVE) virtual reality systems to integrate data as it is acquired from the APS and reconstruction system, allowing users to check intermediate results. Together, these components reduce the time required for reconstruction and analysis from hours to minutes. reconstruction and analysis from hours to minutes.
Advanced Large-Scale Integrated Computational Environment (ALICE)	The goal of the ALICE project at ANL is to eliminate barriers in using independently developed software to construct high performance numerical applications, laying the groundwork for widespread exploitation of teraflop-scale computational resources and the resulting new scientific insights. ALICE addresses the data management and interoperability problems when combining multiple software packages for mesh manipulation, numerical solution of partial differential equations, optimization, sensitivity analysis, and visualization.   ALICE research focuses on low-overhead integration of extensible software for scientific problem solving and component-based toolkits that encapsulate expert knowledge in numerical algorithms and parallel computing. ALICE development will ensure the relevance and practicality of large scale scientific applications design, supporting both new and legacy applications, and enabling scientists to reuse legacy kernels and program more comfortably in, for example, traditional Fortran.
ALICE Memory "Snooper" (AMS)	The ALICE Memory "Snooper" (AMS) is an application programming interface (API) that aids in writing computational steering, monitoring, and debugging tools. The AMS is a client/server, multithreaded API that also supports parallel applications using MPI.
Distributed problem-solving systems	At ORNL, NetSolve is transforming disparate computers and software libraries into a unified, easy-to-access computational service. By aggregating the hardware and software resources of any number of computers loosely connected across a network, NetSolve can tap their combined power through a familiar client interface that hides the underlying complexity of the system, making supercomputing transparently available to a broad range of users on ordinary network platforms.
Collaborative User Migration, User Library for Visualization and Steering (CUMULVS)	Computational steering could revolutionize computational experiments by allowing scientists to interactively explore (steer) a simulation in time and/or space. Instead of the typical simulation mode -- manually setting input parameters, computing results, storing data to disk, visualizing the results in a separate visualization package, then starting again from the beginning -- computational steering allows the scientist or engineer to interactively manipulate algorithmic and model parameters beyond their initial values. The CUMULVS collaborative computational steering tool provides remote visualization, steering, and heterogeneous checkpoint/restart to large distributed simulations. CUMULVS is being used at NPACI, DOE, and DoD sites across the country.
ALICE Differencing Engine (ADE) visualization toolkit	ANL researchers are developing the ADE visualization toolkit so that a scientist or engineer can visualize pairwise differences in up to ten data sets containing vector and scalar data in an immersive, interactive, virtual reality environment. Standard techniques such as vector field glyphs, animated streamlines and flow fields, and scalar field color remappings provide insight into individual data sets. Multiple data sets can be displayed simultaneously along an interactively defined cutting plane. The user may cull away portions of the data set to investigate regions of maximum difference, magnify the differences, and remap the coloring to correspond to alternative scalar fields.   The toolkit has been used to design a new aluminum furnace in a joint project between DOE and Air Products and Chemicals, Inc. Air fuel, an air/oxygen mix, and pure oxygen fuel data sets were displayed and analyzed in order to understand the effect of different fuel types on furnace efficiency.   By incorporating the communication mechanisms of the ALICE Memory Snooper, the ADE toolkit will soon be able to dynamically retrieve data from multiple timesteps of an ongoing simulation, allowing an application scientist to monitor the progress of the application's solution and investigate the differences between timesteps.
Computer simulation demonstrates "breathing" enzyme action	Using powerful supercomputers, NSF- and NIH-supported researchers have unlocked the mystery surrounding one of the fastest, most efficient enzymes in the human body. Acetylcholinesterase (AChE) works to instantly stop transmissions from one nerve cell to the next nerve or muscle cell by catalyzing the chemical reaction that breaks up the neurotransmitter acetylcholine (ACh), serving as the off-switch for the transmission. The active site where the reaction takes place was found deep inside a groove on AChE -- a groove too narrow to admit ACh quickly. Using large scale computer simulations run at the NSF-supported SDSC, researchers have shown that AChE does its job by "breathing." That is, this flexing motion causes AChE to inhale and exhale ACh molecules almost as fast as if the groove were always open.   The supercomputer simulation showed that this motion keeps larger molecules out of the space intended for ACh. This finding shows that enzymes can use movements to select particular substrates in crowded environments like the inside of a living organism.     Current generation where nerve meets muscle. An MCell simulation of ionic current generation at a neuromuscular junction in a rat diaphragm muscle shows the neurotransmitter ACh (cyan spheres) 300 microseconds after release of 6,200 ACh molecules. The highly convoluted membrane of the junction is covered with acetylcholine receptors (the cup-shaped objects on the membrane). After release, ACh molecules diffuse away from their release point and bind to receptors and AChE. The image shows bound AChE (black spheres), unbound AChE (gray spheres), and receptors -- unbound (purple), singly bound (red), doubly bound closed (green), and doubly bound open (yellow). Doubly bound open receptors conduct an ionic current that initiates a cascade of events leading to muscle fiber contraction. AChE breaks down ACh and prevents prolongation of the current. MCell was developed by NPACI-supported researchers at the Salk Institute for Biological Studies and at Cornell University. In the NPACI Neuroscience thrust area, they are collaborating with NetSolve researchers and developers at the University of Tennessee and UCSD.     NPACI researchers are unlocking the mystery surrounding one of the fastest, most efficient enzymes in the human body. Some of the largest acetylcholinesterase (AChE) enzyme simulations to date have been conducted by researchers at the University of Houston and UCSD. An open "side door" in one of the AChE subunits was recorded at 152 picoseconds of one such molecular dynamics simulation. A frame from this animation shows the "breathing" motions of the gorge or channel that leads from the region outside the AChE enzyme to the active site. These fluctuations in the width of the channel allow the substrate acetylcholine (ACh) to move from the outside into the active site. They also contribute to the selectivity of the enzyme by slowing the entrance of substrates that are larger than ACh. Full animation can be seen at http://chemcca10.ucsd.edu/ache_animated.html
Parallel volume-rendering system for scientific visualization	The demand for parallel supercomputing in interactive scientific visualization is increasing as the ability of the machines to produce large output datasets has dramatically increased. NASA's distributed volume rendering system, ParVox, will provide a solution for distributive visualization of large time-varying datasets on a scientist's desktop even when using low-speed networks and low-end workstations.   NASA is currently working on the functional pipelining of ParVox. It will be separated into three modules, an input module that reads data from files or from the network, a rendering module that performs the rendering and classification functions, and a compression/output module that compresses images and outputs them. The input module will support data-input on demand, allowing out-of-core rendering for very large datasets. The major milestone for FY 1999 is to add rendering functions for unstructured grid 4-D datasets.
ParAgent	Developed at Iowa State University with EPA support, ParAgent is an interactive tool for automatic parallelization of specific classes of programs that follows a knowledge-based approach, using program characteristics to guide the automatic parallelization process. ParAgent includes tools for Web-based performance monitoring, performance analysis, and visualization. The current version is designed for programs based on the Finite Difference Method (FDM), and has been implmented for parallel computing on 64 node cluster of 200 MHz Pentium II workstations at NASA's Ames Laboratory.
VisAnalysis Systems Technology/Space-Time Toolkit (VAST/STT)	The EPA-supported VAST team at the University of Alabama in Huntsville (UAH) is developing Java-based software for visual integration of multi-source data. This VAST/STT accepts data in native space-time domains and allows "on-the-fly" data transformations into a user-selected interactive display domain that can integrate 3-D model grids, 2-D maps, Geographic Information System (GIS) vector data, vertical and horizontal profiles, dynamic particles, episodic events, and satellite and aircraft sensor data. In particular, the STT can be used to explore relationships among dynamic spatial data with temporal resolutions from microseconds to centuries.   The STT is being tested by interrelating two months of 1995 data -- obtained from aircraft and satellite-based sensors, meteorological and environmental model output, vertical wind profilers, point source measurements, and Doppler radar volumes -- for the Nashville Southern Oxidant Study (SOS).
Advanced data visualization and exploration	Today's limited capabilities for extracting and assimilating the information buried in data means that much scientific data is either never examined or examined only superficially. SNL research focuses on developing advanced capabilities for visualizing and exploring data in order to improve productivity and/or create new technologies in particular fields.
Molecular dynamics	Molecular dynamics: Unfolding of Ig and Fn-III protein domains by steered molecular dynamics simulation. The architecture of immunoglobulin-like (Ig) and fibronectin type III-like (Fn-III) domains constitutes possibly the most prevalent structural motif of proteins that have a role in cell-cell signaling, cell-cell aggregation, and embryogenesis, as well as in mechanically coordinating and strengthening cells and tissues. The proteins are implicated in the etiology of many diseases, ranging from heart insufficiency to cancer. Recently, atomic force microscopists have investigated the mechanical properties of connected Ig and Fn-III domains, by simulating experiments to explain the specific mechanical properties of these proteins using steered molecular dynamics, a new simulation technology developed at the NIH Resource for Macromolecular Modeling and Bioinformatics (Beckman Institute, University of Illinois), that permits researchers to mechanically manipulate models of proteins. The technology combines its high end graphics tool with its parallel simulation software.
High end applications	High end applications FY 1999 accomplishments and FY 2000 plans
Code that keeps blood flowing (NekTar)	New NekTar software for modeling blood and other complex fluid flows may lead to changes in accepted surgical practices that will extend the life expectancy of those suffering from arterial diseases like atherosclerosis. These codes can produce animated simulations of flows that once could only be captured accurately in "still" formats like magnetic resonance imaging (MRI). Research at Brown University and simulations at NCSA will allow scientists to more accurately model many types of fluid flows and to do computational steering -- that is, to zero in on specific areas of a calculation while the computation is running. This is especially useful in studying the complex human circulatory system which, with its miles of arteries, veins, and vessels, not only carries oxygen but also transports wastes and aids in heating and cooling the body.
Making and testing neuron models	Computational models describe the behavior of neurons and the communications among neurons in networks and are essential for investigating the functions of brain regions. Models assist in studying how interactions between regions underlie the functions of the entire nervous system, including behavior and thought. Models can be used to perform virtual experiments that are too difficult or impossible to conduct using biological tissue or living subjects. Computationally intensive simulations can incorporate greater numbers of neurons to model increasingly complex and realistic properties, both electrical and chemical. NSF-funded NPACI neuroscience researchers using high performance computing and infrastructure resources are conducting this work.
Road maps for understanding the human brain	To understand how the brain allows humans to perform multiple tasks (such as simultaneously observing traffic, listening to the radio, and stopping at a traffic signal) and to investigate changes in the brain related to pathological conditions and disease, NSF-funded NPACI neuroscientists are creating tools to produce road maps of the structures and connections in the brain. The researchers are producing software to analyze data collected from MRI scans, digitized images of cryosectioned brains -- brains frozen and cut into thin slices -- and high powered microscopes. The software can re-create a 3-D model of the brain and generate a road map for comparing a particular brain with others.     By combining the high temporal resolution of electroencephalography (EEG) and the high spatial resolution of functional magnetic resonance imaging (fMRI), psychologists are able to study the series of events that underlie the formation of memories in the brain. Here we see the cortical regions that are activated as a subject remembers a sequence of letters presented at 3-second intervals. Researchers at the Pittsburgh Supercomputing Center are focusing on improving the rapid, online analysis and visualization capability of fMRI. As microprocessor performance increases, realtime fMRI technology that was pioneered on supercomputers may soon be incorporated into dedicated hardware bundled with MRI scanners. This will allow immediate assessment of scan quality and facilitate the use of fMRI in psychiatry, drug evaluation, and neurosurgical planning.
Catalytic antibodies	When viruses, bacteria, or other harmful invaders enter the bloodstream, the immune system fends them off by generating antibodies -- proteins shaped to latch onto the intruder (analogous to a lock fitting a key) so that its ability to cause harm is immobilized. What if it were possible to harness this remarkable ability to manufacture proteins matched to the 3-D features of these microscopic intruders? For the pharmaceutical industry, these "catalytic antibodies" offer promise for rational drug design -- the creation of molecules with 3-D features sculpted to interact with other molecules.     Close-up view of antibody 13G5, with the cyclopentadienyl ferrocene molecule deeply buried in the antibody's binding-site. Scripps Research Institute scientists determined the structure of this antibody, opening the way for computational studies of its catalytic effect by UCLA scientists using the SGI/Cray Origin2000 at NCSA.
Taking the pulse of a red giant	NSF-funded University of Minnesota and NCSA astrophysicists have generated a 3-D simulation of a red giant with such detail that they could watch it pulsate. In a region that encompassed nearly the entire star and was the size of the orbital radius of Jupiter around the Sun, super hot gases were in lava-like turmoil. This global convective pattern flowed asymmetrically, with gas flowing outward from the center of the generally hotter side of the star to the cool side, emitting heat along the way. Once on the cool side, the gas sank, forming a funnel that reheated upon passing the hot, stellar core. This may help explain differences in illumination within these large, pulsating stars -- upon whose "standard candles" astronomers rely for mapping distances in the universe. It will also help scientists know what to expect from our own Sun that eventually will also become a red giant.     Model of a red giant star. At left, the relatively diffuse envelope has contracted to about its minimum size. At right, the same stage rendered transparently so that its hot stellar core is visible. Relatively warm temperatures are red and yellow, while relatively cool temperatures are blue and aqua. Researchers from the University of Minnesota generated the largest simulation to date of highly convective red giant stars -- 134 million computational cells -- by adapting their numerical approach to the distributed shared-memory capabilities of the SGI Cray Origin2000 at NCSA.
Assisted Model Building with Energy Refinement (AMBER)	The NSF- and NIH-supported AMBER computational chemistry project has developed software to study protein folding and the relative free energies of binding of two ligands to a given host (or two hosts to a given ligand), to investigate the sequence-dependent stability of proteins and nucleic acids, and to find the relative solvation-free energies of different molecules in various liquids.   Researchers recently used AMBER to make the longest molecular dynamics simulation of a molecule, a 36-residue protein called the villin headpiece subdomain. The simulation -- carried out over 100 days at the Pittsburgh Supercomputing Center on an SGI/Cray T3D and a T3E at SGI/Cray Research -- followed the molecule for a full microsecond, which was possible due to improvement of the code's efficiency for parallel computing.     The protein-folding problem -- how a protein's amino-acid sequence relates to its folded shape -- is the most important unsolved problem facing computational biochemistry. NSF-supported researchers from the University of California, San Francisco, using PSC's T3D and resources at Cray Research, performed a microsecond simulation of the folding of a small protein, the villin headpiece. This image shows a snapshot along the computed folding pathway. The computation is two orders of magnitude longer than the longest previous calculation of its kind and has identified the initial collapse of the protein to a "molten globular state" from which it searches for the ultimate fold.
NASA Grand Challenges	SGI is collaborating with NASA and Grand Challenge investigator teams to implement highly scalable applications on the Cray T3E series of scalable systems, using advanced systems and operating software to implement a wide range of models based on standard languages and libraries to achieve new levels of performance. By achieving high performance levels, teams will be able to perform more accurate modeling.   Currently, the testbed at NASA's Goddard Space Flight Center (GSFC) includes 1,024 processors supporting the Grand Challenge teams as well as NASA-directed programs. Applications are implemented using SGI-developed scalable operating systems, compilers, and development tools to achieve scaling and performance. The challenge is to achieve effective distribution of the task and associated interprocessor communication. Long latencies for communication among processors impede the ability of science models to scale in performance. NASA's Earth and Space Sciences (ESS) Round-2 teams, in concert with SGI analysts, are implementing Earth and space science models on the Cray T3E. Most 50 gigaflops milestone efforts have been completed and several 100 gigaflops milestones have been conducted. Work on the remaining 50 and 100 gigaflops milestones will continue through FY 1999.     Hundreds of thousands of asteroids cross Earth's orbit around the Sun. NASA-supported researchers modeled Castalia as a peanut-shaped solid rock 1.6 kilometers long and 0.8 kilometers wide, shown here in a cutaway view being hit by a house-sized rock traveling at 5 kilometers per second. The scattering white dots are fragments from the smaller rock. Fractures appear throughout the asteroid, with the greatest damage shown in red. Lasting merely a second, the collision approximates the force of an early atomic bomb. Nuclear weapons have been proposed for breaking up, or at least diverting, asteroids headed towards Earth. These simulations show that such an impact will fracture a solid asteroid, but, later, gravity will reassemble the pieces.     Sheared Active Region Fields: Pictured above is the evolution of an initially potential (current-free) dipole magnetic field typical of northern-hemisphere active regions on the Sun. The field was subjected to differential rotation and supergranular diffusion, modeling the effects of the Sun's surface convection, which stretch and spread the flux threading the solar surface. The objective is to demonstrate the development of properties of the magnetic field that show distinct preferences by hemisphere, independent of the sunspot cycle. This calculation used FCTMHD3-D, NASA's parallel, finite-volume 3-D code.   Upper left: Seen from overhead, white lines of magnetic force connect the regions of positive (red) and negative (blue) polarity in the initial state. There is no twist in the field. The force lines cross the magnetic neutral line at the surface -- the yellow-green boundary between the red and blue polarities -- at a 90 degree angle.   Upper right: After the shear and diffusion at the base have been applied for a time, the lines of force cross the magnetic neutral line at the surface at an acute angle. A component of the field now points to the right along the neutral line, when observed from the positive polarity (red) side.   Lower left: A sequence of lines of force drawn at increasing heights above the surface shows an apparent counterclockwise rotation of the loops as seen from above.   Lower right: Several lines of electric current are drawn here, showing that the current flows antiparallel to the lines of magnetic force - that is, the current points from the negative polarity (blue) side to the positive polarity (red) side. This imparts a left-handed twist, or negative helicity, to the magnetic field.
National Center for Environmental Prediction (NCEP)	Numerical weather models guide weather forecasters, incorporating observations of temperature, wind, precipitation, pressure, and other meteorological information from sources on the ground, in the air, and in space. These observations are processed by powerful computers that generate predictions for forecasters, allowing them to anticipate weather conditions from hours to weeks in advance -- or, as in the case of the recent El Niño, many months in advance.   NOAA's NCEP has worked with the Naval Research Laboratory (NRL) and NASA/GSFC to develop techniques to implement weather prediction models on parallel supercomputers. NCEP provided benchmarks for its most recent procurement that were independent of the architecture of the target machine, leading to the first-ever selection of a scalable architecture computer to run the U. S. weather forecasting suite of models. At the time of the announcement, Department of Commerce Secretary William M. Daley said, "Accurate weather forecasting is one of the great scientific achievements of the 20th century. We have reached unprecedented levels of accuracy in recent years as a result of much research, modernization and improvements such as supercomputers, radars, satellites and other technologies....We eagerly await the next generation of computational power because we know we can do even better in the future."
Predicting weather that affects aircraft	With support from the Federal Aviation Agency (FAA), NOAA's FSL has improved the Rapid Update Cycle (RUC-2) model to help meteorologists better predict weather that affects aircraft including icing conditions, winds, clouds, and clear air turbulence. While the initial version ran only once every three hours, RUC-2 can run hourly, has higher resolution, captures greater detail, and produces more accurate information.
Hurricane prediction	The 1998 season included several major hurricanes that caused considerable property damage to the southeastern United States and an enormous death toll in Central America. The GFDL Hurricane Prediction System was used extensively by NOAA/NCEP as part of its modeling suite to predict the track of these storms.
Realtime weather forecasting	Advances in time-urgent weather forecasting were demonstrated at the January American Meteorological Society (AMS) meeting in Dallas, TX. The input data were the detailed data stream from NOAA's new Doppler radar system. Each day, realtime, high resolution numerical weather forecasts were generated remotely using an entire 128-processor SGI/Cray Origin2000 supercomputer running almost non-stop at the NSF-funded NCSA.
NASA Intelligent Synthesis Environment demonstration	NASA's proposed Intelligent Synthesis Environment (ISE) program focuses on how future missions will be developed from initial concept through operations and conclusion. ISE will conduct research and develop tools and processes to allow scientists, technologists, and engineers with diverse expertise and interests to function as a team. By functioning as a networked collaboration among geographically dispersed and professionally diverse personnel involved in defining, designing, executing, and operating NASA's missions, development time will be shortened and life-cycle costs reduced. When fully developed, collaboration will take place in a full-sensory, immersive, virtual environment in which humans and computers interact through 3-D sight, sound, and touch in a computationally rich mission life-cycle simulation. Methods for trading cost, risk, and performance over total mission life cycles will be developed, enabling NASA to fully understand missions prior to committing to development.     In one ISE demonstration, exhibit staff and researchers at three NASA field centers will become virtual astronauts in a simulated spacewalk rehearsal of a future space station assembly flight. Subjects will wear virtual-reality headgear displaying mock-ups of space station parts derived from computer-aided design files. The "astronauts" will "maneuver" around the components while computers at each site exchange positional information over the Internet, continuously updating the scene from each user's point of view.
Efficient electromigration simulations using approximate Green's functions	In some applications of the boundary integral method, most notably when there is symmetry present, Green's function can lessen the computational effort, since it exactly satisfies the imposed boundary conditions on part of the surface. This piece of the boundary can then be eliminated from the calculations, reducing the time required to construct and solve the matrix system. While this is a powerful method, such exact Green's functions, which must also be simple enough to work with, are rare.   ORNL researchers have demonstrated that the benefits of this technique can be extended by means of "approximate Green's functions." They have employed this new algorithm in modeling the electromigration problem in materials science, in which a microelectronics wire has been deformed by a void. Here, an approximate Green's function can be constructed to reduce the computational domain to simply the void surface. Despite using an approximation to the true Green's function, comparisons with finite element results show that there is no loss of accuracy in the analysis. Moreover, by eliminating the corner formed by the intersection of the void and wire boundaries, the results in this area are much more accurate.   Electroforming simulation is another application where this technique will prove useful. A typical geometry consists of a tank and electrodes. With this new method, all tank surfaces can be eliminated from the calculation, saving time to discretize the model and in the analysis.
Biology WorkBench	The NSF-supported Biology WorkBench at NCSA is a revolutionary Web-based tool for biologists. The WorkBench allows biologists to search many popular protein and nucleic acid sequence databases. Database searching is integrated with access to a wide variety of analysis and modeling tools, all within a point and click interface that eliminates file format compatibility problems. This product was first released in 1996 and a new version 3.0 was released recently with many enhanced functionalities.
Neuroimaging Analysis Center (NAC)	NAC, supported by NIH's National Center for Research Resources (NCRR), develops image processing and analysis techniques for basic and clinical neurosciences. Core NAC research includes algorithms and techniques for segmenting brain structures, registration methods for associating image data to patient anatomy or one set of image data to another, visualization tools for displaying anatomical structures and quantitative information, software and hardware infrastructures for high performance computing, and digital anatomy atlases for interactive and algorithmic computational tools. NAC emphasizes the dissemination of concepts and techniques.   NAC technologies are employed in research on Alzheimer's disease and the aging brain, morphometric measures in schizophrenia and schizotypal disorder, quantitative analysis of multiple sclerosis, and interactive image-based planning and guidance in neurosurgery.
Laboratory of Neuro Imaging Resource (LONIR)	There is a rapidly growing need for brain models comprehensive enough to represent brain structure and function as they change across time, in large populations, in different disease states, across imaging modalities, across age and gender, and across species. The NIH-supported LONIR develops novel strategies to investigate brain structure and function in full multidimensional complexity. LONIR provides international networks of collaborators with a diverse array of tools to create, analyze, visualize, and interact with models of the brain. A major focus of these collaborations is to develop 4-D brain models that track and analyze complex patterns of dynamically changing brain structure in development and disease, expanding investigations of brain structure-function relationships to four dimensions.   Modeling research focuses on new strategies for surface and volume parameterization that provide an advanced analysis of surface and volumetric brain models, tracking their change across time. Additional research focuses on anatomical fundamentals, analyzing anatomical and cytoarchitectural attributes across multiple scales and across time; and visualization and animation, for the dissemination of brain models that visualize complex variations in brain structure and function across time. Ongoing national and international collaborations are analyzing normal and aberrant growth processes, brain development, tumor growth, Alzheimer's disease, and related degenerative disease processes, schizophrenia, and brain structure in normal and diseased twins. The range and sophistication of these strategies matches the broad scope of studies that focus on mapping and modeling the dynamically changing brain.
Virtual Cell	The NIH-supported National Resource for Cell Analysis and Modeling (NRCAM) is developing methods to model cell physiological processes in the context of the actual 3-D structure of individual cells. Approaches in computational cell biology are coupled with high resolution light microscopy to facilitate the interplay between experimental manipulation and computational simulation of specific cellular processes.   NRCAM is developing the Virtual Cell, a general computational framework for modeling cell biological processes. This new technology associates biochemical and electrophysiological data describing individual reactions with experimental microscopic image data describing their subcellular locations. Individual processes are integrated within a physical and computational infrastructure that will accommodate any molecular mechanism. Current development of the Virtual Cell will expand the generalized mathematical descriptions to include additional cell biological mechanisms, increasing accessibility to biologists studying different biological processes, and integrating the interface with a database of images and reaction mechanisms. Current applications of Virtual Cell include studies of calcium dynamics in neuroblastoma cells and cardiomyocytes, and studies of intracellular RNA trafficking in oligodendrocytes. Additional collaborative research projects include modeling diffusional processes in mitochondria, examining the role of [Ca2+]i in triggering Ca2+ sparks in the heart, and analyzing structural changes in the endoplasmic reticulum during egg activation.
Optical information processing	One of the major barriers to commercial application of optical technology to information processing is the high cost of system development and manufacture. This problem has been solved in other industries through the use of computer-aided design (CAD) and the integration of system design with manufacturing. The development of better system level metrology is needed to allow more computer-based methods to be used in the design and manufacture of optical information processing systems.   NIST is designing an optical pattern recognition system that will test an input image (at video rates) against a large reference set -- 1,000 human faces, for example -- with images of 640 by 480 pixels or larger. NIST has constructed realtime video system testbed versions of the two subsystems needed in a high speed optical pattern recognition application. NIST will evaluate the subsystems and components for future design work.
Micromagnetic modeling	NIST is developing computational tools for accurate and efficient micromagnetic calculations. Such calculations are essential to achieving higher densities and faster read-write times in magnetic disk drives. Given an earlier NIST study that showed how unreliable software for micromagnetic calculations is, NIST is now developing a reference code that will be thoroughly tested, compared to results from experimental measurements, and made publicly available.
Modeling the high speed machining process	A combined experimental and analytical program to study the dynamics of high speed machining is currently underway at NIST. Its goal is to provide accurate measurements and obtain a better scientific understanding of fundamental metal-cutting processes while evaluating the predictive capabilities of finite-element software for high speed machining operations. NIST has derived a mathematical model of the basic process of orthogonal machining, and finite-element software for its simulation has been developed and implemented on a NIST supercomputer.
Developing pollutant control strategies	EPA-funded researchers at North Carolina State University have been exploring how computer-based decision support systems (DSSs) can improve strategy design. In prototype DSSs focused on air quality management, alternative management strategies such as command-and-control and emissions trading programs are designed and tested with state-of-the-art air quality models. The prototype DSSs also have optimization components that can be used to identify cost-effective benchmark strategies, characterize the tradeoffs among design goals, and incorporate consideration of uncertainty in the design process. Techniques used for these analyses, such as genetic algorithms, are highly computationally intensive. Their use is made practical through high performance computing on a heterogeneous, distributed network of computers.