Search

This paper describes the conceptual design of a highly reliable 108‐Bit Bubble Domain Memory for the Space Program. The Memory has random access to blocks of closed‐loop shift registers, and utilizes self‐contained bubble domain chips with on‐chip decoding. Tradeoff studies show that the highest reliability and lowest power dissipation is obtained when the memory is organized on a bit‐per‐chip basis. The final design, has 800 bits/register, 128 registers/chip, 16 chips/plane, and 112 planes, of which only seven are activated at a time. A word has 64 data bits +32 checkbits, used in a “16‐adjacent” code to provide correction of any combination of errors in one plane. 100 KHz maximum rotational frequency keeps power low (≤ 25 watts) and also allows asynchronous operation. Data rate is 6.4 megabits/sec, access time is 200 μsec to an 800‐word block and an additional 4 msec (average) to a word.

The storage of information in present day computer systems is shared between high‐speed, high‐cost memories and low‐speed, low‐cost mechanical storage devices. With the advent of cheap bulk memory technology one can envision a storage hierarchy with three levels of storage: buffer, memory, and a mechanical backing store. The individual cost and performance characteristics of the three technologies involved affect the overall computer system cost‐performance. In fact, it is the judicious selection of technologies and system configurations which leads to a well designed storage system. This paper first describes the automatic management of information in a three level storage system where the objective is to keep current information in the fastest device and the rest of the information in the slower devices. Then the evaluation of the storage systems is described for a multiprogramming environment where total system dollars per instruction executed is the criterion for system cost‐performance. Finally, two ...

Numerical modeling of anisotropic media is a computationally intensive task since it brings additional complexity to the field problem in such a way that the physical properties are different in different directions. Largely used in the aerospace industry because of their lightweight nature, composite materials are a very good example of thermally anisotropic media. With advancements in video gaming technology, parallel processors are much cheaper today and accessibility to higher-end graphical processing devices has increased dramatically over the past couple of years. Since these massively parallel GPUs are very good in handling floating point arithmetic, they provide a new platform for engineers and scientists to accelerate their numerical models using commodity hardware. In this paper we implement a parallel finite difference model of thermal diffusion through anisotropic media using the NVIDIA CUDA (Compute Unified device Architecture). We use the NVIDIA GeForce GTX 560 Ti as our primary computing device which consists of 384 CUDA cores clocked at 1645 MHz with a standard desktop pc as the host platform. We compare the results from standard CPU implementation for its accuracy and speed and draw implications for simulation using the GPU paradigm.

This paper reviews the status of quantitative models for field‐access bubble domain devices. At present, quasistatic single‐bubble margins can be predicted with reasonable accuracy. The essential effects which such a model must include are discussed first, and then used as a basis for comparing the approaches published so far. Inclusion of bubble‐bubble interactions and long‐term statistical effects is discussed briefly. The paper concludes with some comments on device optimization and scaling.

This paper reviews some recent measuremtns on high‐performance Josephson latching logic, and discusses a compatible ac power scheme which provides good voltage regulation and low clock skew performance for a large computing machine.

This paper describes an analytical study to determine the stability requirements for a holographic system indended for the NASA space shuttle. The primary emphasis is on optimization of the holographic geometry to minimize the effect of film variations between hologram exposure in space, and wavefront reconstuction on the ground. It is shown that it is extremely important to use a balanced geometry (equal reference and object beam angles) and if possible, a specific oject beam angle.

This paper shows some examples of time‐dependent picture processing by means of a hybrid optical processing system. It consists of electronical and optical parts. The advantages of both are combined in our TV‐optical feedback system.

This paper reviews theoretical and expermental backgrounds for autostereoscopic 3‐D televisions, and describes, in some detail, the volume scanning method and the parallax panoramagram method, which are applied to autostereoscopic 3‐D televisions. It also briefly describes some of recent experiments related to recording and reconstruction of 3‐D images of various kinds.

This paper will present data on the latest ISR results on charm and beauty production, in an effort towards a more global synthesis of previous ISR data. The total charmed hadron cross‐section of 1 mb implied by these new results will be discussed. Expectations for new heavy flavor production at the CERN pp Collider will finally be discussed, based on what we have learned thus far at the ISR.

In the past 15 years digital computers have been used to greatly improve from images. In this paper we review and discuss a number of techniques for the enhancement of imagery by digital computer. The implications of this technology to verification of arms control treaties is left to the reader.

Massively parallel systems are presently the focus of intense interest for a variety of reasons. A key problem is how to control, or ‘‘program’’ these systems. In previous work by the authors, the ‘‘optimum finding’’ properties of Hopfield neural nets were applied to the nets themselves to create a ‘‘neural compiler.’’ This was done in such a way that the problem of programming the attractors of one neural net (called the Slave net) was expressed as an optimization problem that was in turn solved by a second neural net (the Master net). The procedure is effective and efficient. In this series of papers we extend that approach to programming nets that contain interneurons (sometimes called ‘‘hidden neurons’’), and thus we deal with nets capable of universal computation. Our work is closely related to recent work of Rummelhart et al. (also Parker, and LeChun), which may be viewed as a special case of this formalism and therefore of ‘‘computing with attractors.’’ In later papers in this series, we present the theory for programming time dependent behavior, and consider practical implementations. One may expect numerous applications in view of the computation universality of these networks.

This paper summarises recent results of theoretical analysis and numerical simulation, in fully connected networks of the Little‐Hopfield class. The theoretical analysis is based on methods of statistical mechanics as applied to spin‐glass problems, and the numerical work involves massively parallel simulations on the ICL Distributed Array Processor (DAP). Specific applications include: (i) exact results for the fraction of nominal vectors which are perfectly stored by the usual Hebbian rule; (ii) a numerical estimate of the position of the second phase transition in the Hopfield model, at which there is effectively total loss of memory capacity; (iii) a numerical study of the nature of the spurious states in the model; (iv) an exploration of the performance of a learning algorithm, including the exact storage of up to 512 (random) nominal vectors in a 512 node model; (v) a theoretical study of the phase transitions in generalizations where the energy function is a monomial in the state vectors.

This paper outlines an approach to the automated generation of structured networks of threshold elements or ‘‘neurons’’ for computational purposes. The structured networks are obtained by recursive operations on primitive elements, constrained to produce networks whose recursive description is relatively simple. A preliminary analysis of the performance of structured networks on a specific continuous coding problem is presented, and compared to the performance of unstructured networks on the same problem. This comparison shows that recursively structured networks can exhibit efficient learning and high performance, while permitting generalization to larger instances of the same problem.

This paper develops a particular error correction code which can be effectively decoded by a relatively simple neural network. In high noise situations, this code is comparable to that used at present in deep space communications. The neural decoder has N! stable states with only N2 neurons, and can quickly extract information from analog noise. This example illustrates the effectiveness of neural networks in solving real problems when the problem can be cast in such a fashion that it fits gracefully on the network.

This paper shows how the internal representations necessary for memorization can be automatically formed in neural networks or artificial systems.

This paper presents a simple algorithm which converts any context‐free grammar into a connectionist network which parses strings (of arbitrary but fixed maximum length) in the language defined by that grammar. The network is fast, O(n), and deterministicd. It consists of binary units which compute a simple function of their input. When the grammar is put in Chomsky normal form, O(n3) units needed to parse inputs of length up to n.

If we think of our experiences as being recorded continuously on film, then human memory can be compared to a film library that is indexed by the contents of the film strips stored in it. Moreover, approximate retrieval cues suffice to retrieve information stored in this library: We recognize a familiar person in a fuzzy photograph or a familiar tune played on a strange instrument.This paper is about how to construct a computer memory that would allow a computer to recognize patterns and to recall sequences the way humans do. Such a memory is remarkably similar in structure to a conventional computer memory and also to the neural circuits in the cortex of the cerebellum of the human brain.The paper concludes that the frame problem of artificial intelligence could be solved by the use of such a memory if we were able to encode information about the world properly.

This paper compares the performance of matched filters with several versions of nonlinear associative memory for a specific pattern recognition problem, where we attempt to identify the presence and orientation of an angled line. Comparative simulations are presented on the above processors for 5–by–5 and 9–by–9 pixel data fields when the input line patterns are corruped by Gaussian noise. We conclused that considerable sophistication was required in the nonlinear associative memory processor design to achieve equivalent performance to the matched filter. In all cases that matched filter was shown to exhibit a reduced computational load.

Neural network architectures and algorithms provide an anthropomorphic framework for analysis and synthesis of learning networks for correlation, tracking and identification applications. Many researchers in neuroscience believe that through evolution nature has developed efficient structures for multi‐sensor integration and data fusion. Consequently, innovations in electronic surveillance and advanced computing may result from current interdisciplinary research in neural networks and natural intelligence (NI). In this paper we propose a network learning paradigm, called entropy learning, based upon the Principle of Maximum Entropy (PME). The close relationship between entropy learning and simulated annealing in network solutions to combinatorial optimization problems is discussed.

In this paper, we address the capabilities and cost‐effectiveness of the current models of neural networks for carrying out general computation. We show that neural networks are formally capable of performing any conventional computation. However, we show that for hard ’algorithmic’ problems, the speed with which the solution is achieved comes at the cost of an excessive size for the network.

A computational experiment for simulating building performance needs basically three important input data sets, namely the material properties, environmental conditions and initial conditions. As the result of any prediction depends strongly on a considered geographical location, a database of locations and climatic data is required to support any reliable numerical simulation. In this paper, we present an online application for gathering, sharing and processing of climatic data from stations all around the world. An illustrative example confirms the necessity of such a software tool.

This paper discussed the motivation of students and teachers and the production and testing of materials for teaching contemporary physics topics within the context of raising ideas for the working group on teaching modern physics. Some of the ideas arose as part of the author’s assocation with the Conference on the Teaching of Modern Physics held at Fermilab in 1986. The role for teachers at all levels in pursuing this goal is also discussed.

We report on the planning of a conference with the purpose of creating cooperative networks in physics education. In this introductory paper we describe the organization process, the main activities and results of the Oaxtepec Conference.

This paper has two objectives. First to make some comments on Tiberghien’s paper (in the proceedings); particularly on her view of the factors influencing the understanding of physics concepts. A stronger emphasis could be made for considering the types of preconceptions that students bring to the learning situation, which is seen as a crucial factor to better grasp physics concepts. The second is to present the types of preconceptions that students possess about three independent characteristics when two orthogonal velocities simultaneously influence a moving body. Data was collected through individual interviews before instruction and through a questionnarier in a class setting after instruction. Educational implications for making explicity what is defined as ‘‘implicit characteristic’’ are also discussed.

Computational Science is at the very beginning of centuries of growth, comparable to the four centuries of experimental advances since Galileo. The Grand Challenges to Computational Science are unsolved scientific problems of extraordinary breadth and importance which will demand continuing computational advances throughout the forthcoming computational era. Supercomputers can be used to see phenomena not directly accessible to experiment in key scientific and engineering areas such as atmospheric science, astronomy, materials science, molecular biology, aerodynamics, and elementary particle physics. However, the benefits of supercomputers will be greatly increased if some major difficulties are overcome. In this paper, I address some of the tougher requirements on current grand challenge research to ensure that it has enduring value. The problems of algorithm development, error control, software productivity, and the fostering of technological advances are especially important.

This paper is an overview of part of the beam optics code development effort in the Accelerator Technology Division at Los Alamos National Laboratory. The aim of this effort is to improve our capability to design advanced beam optics systems. The work reported is being carried out by a collaboration of permanent staff members, visiting consultants, and student research assistants. The main components of the effort are building a new framework of common supporting utilities and software tools to facilitate further development. research and development on basic computational techniques in classical mechanics and electrodynamics, and evaluation and comparison of existing beam optics codes, and support for their continuing development.

Autonomy is a key requirement of robots used in self‐contrained manufacturing systems. This is motivated by the high cost of access to self‐contained environments and the awkwardness of programming by teach pendant. We address this requirement in this paper and describe a simple and efficient algorithm for generating a 2n‐tree (generalized quad‐tree) representation of the free configuration space for a manipulator moving in a workspace with obstacles. This algorithm is based on the existence of uniform bounds on the Jacobians relating the differential motions of points on the manipulator to differential joint motions. We also describe how this representation can be searched to generate a collision‐free path and optimized with dynamic contraints to produce an executable trajectory.

The preliminary specifications for various new detectors throughout the world including those at the Superconducting Super Collider (SSC) already make it clear that existing event building techniques will be inadequate for the high trigger and data rates anticipated for these detectors. In the world of high‐energy physics many approaches have been taken to solving the problem of reading out data from a whole detector and presenting a complete event to the physicist, while simultaneously keeping deadtime to a minimum. This paper includes a review of multiprocessor and telecommunications interconnection networks and how these networks relate to event building in general, illustrating advantages and disadvantages of the various approaches. It presents a more detailed study of recent research into new event building techniques which incorporate much greater parallelism to better accommodate high data rates. The paper lists trends in data acquisition system needs giving specific requirements and a proposed arc...

Reason is a software package to allow one to do physics analysis with the look and feel of the Apple Macintosh. It was implemented on a NeXT computer which does not yet support the standard HEP packages for graphics and histogramming. This paper will review our experiences and the program.

CPS is a package of software tools for splitting of computational task, called a job, among a set of processes distributed over one or more processors. It is designed to function as a tool for solving computing problems which require many computing cycles per I/O byte, and is well suited for computing platforms consisting of ‘‘farms’’ of processors, operating in parallel. This paper describes three essential features of CPS: data transfers between cooperating processes, remote subroutine calls, and process queues.

IEEE P1596, the Scalable Coherent Interface (formerly known as SuperBus) is based on experience gained while developing Fastbus (ANSI/IEEE 960‐1986, IEC 935), Futurebus (IEEE P896.x) and other modern high‐performance buses. SCI goals include a minimum and bandwidth of 1 GByte/sec per processor in multiprocessor systems with thousands of processors; efficient support of a coherent distributed‐cache image of distributed shared memory; support for bridges which interface to existing or future buses; and support for inexpensive small rings as well as for general switched interconnections like Banyan, Omega, or crossbar networks.This paper reports to the status of the work in progress and suggests some applications in data acquisition and physics.

This paper describes a system for performing histogramming of n‐tuple data at interactive rates using a commercial SIMD processor array connected to a work‐station running the well‐known Physics Analysis Workstation software (PAW). Results indicate that an order of magnitude performance improvement over current RISC technology is easily achievable.

‘‘This work is a result of a collaboration between the LEPP laboratory and the ARCHIPEL S.A. company, and is part of the Archipel Parallel Computing Concept presented below. We have developed a Transputer based board for VMEbus systems with a dual‐ported memory, and we have written a Unix driver for that board. Several existing tools have been ported. In order to assess the system we used to GEXAM1 benchmark program. The results can be found in this paper.’’ (AIP)

Fado Tagging System has been developed in the context of the 4th level trigger of the DELPHI Data Acquisition Project at CERN. We shall present in this paper the differences between this tagging system approach and the traditional one.

This paper is essentially a verbatim record of the talk given at the end of the conference summarizing the topics discussed in the Architectures and Algorithms sessions.

We review the sessions related to Implementation and Management. The papers presented at the Conference cover a wide range of important issues in software engineering and management. They indicate a trend toward more use of commercial systems and standards. This trend will likely have a significant influence on plans for future systems.

This paper introduces the application of fractal dimension to the analysis of HEP events. A top‐down technique for analyzing High Energy Physics data based on fractal dimension is also discussed. Fractal methods and powerful imaging tools combine to offer an imaging application that enables the physicists to manipulate large data sets in a way in which all the data can be seen and interpreted at one time.

Architecture is an overused and ill‐defined term. This paper discusses the architecture of future data acquisition systems in terms of a Science of building such systems. We examine some of the projects which are in progress and which may enable us to build data acquisition systems more scientifically. In particular we review work on modelling and simulation and examine some ideas for Level 2 Triggering and Event building, which are both novel and particularly suitable for comparative studies between simulation and reality. We discuss some of the lessons learned from current data acquisition systems, which can hopefully become architectural principles for future data acquisition systems.

In this paper we will review recent activities in the area of code development as it affects the accelerator community. We will first discuss the changing computing environment. We will review how the computing environment has changed in the last 10 years, with emphasis on computing power, operating systems, computer languages, graphics standards, and massively parallel processing. Then we will discuss recent code development activities in the areas of electromagnetics codes and beam dynamics codes.

A total of 36 workshop participants attended at least one session of the Long‐Term Stability working group. We avoided turning these sessions into a specialized seminar series by meeting in two subgroups, loosely labeled Analysis and Diffusion and Tracking, so that working discussions among a resonably small number of people were possible. Nonetheless, no attempt is made to categorize the 13 group B papers according to original subgroup. (AIP)

Various performance measures have been developed for performance evaluation of the stratified Thermal Energy Storage (TES) tank. There are many methods which have been developed including numerical, analytical to determine the TES tank efficiency using Figure of Merit (FOM) and Thermocline Thickness (WTC) parameters as the performance indicators. However, these methods are more complicated and need more technical data. Therefore, a simple and easy method is required to predict the performance of TES tank. In this paper, a comparative study using results from analytically solved sigmoid dose response function and artificial neural network (ANN) is conducted to determine the parameters.

This paper presents the further development of a program complex for solving CFD problems, oriented on heterogeneous GPU-based computer systems. Basing on finite volumes method the difference scheme is constructed for Quasi Gas Dynamic equations system in 3D formulation on arbitrary non-orthogonal structured index grid. The use of multi block grids is provided. To improve the stability condition the flux relaxation approach was used. The algorithm efficiency was verified on a set of test problems. A detailed investigation of speed-up and scaling were made. Good parallelization efficiency was achieved up to very large number of parallel GPUs in use.

We have designed and implemented the Data Management Facility (DMF) as the first generation data archive for the Hubble Space Telescope. The DMF is based on optical disk, jukebox, and relational database technologies. The archive consists of three major subsystems: the archive server, which builds and manages the optical disk data archive, the catalog server, which is a relational database (Sybase) describing the contents of the disk archive, and the User Interface, which manages the user’s session displays and interactions with the disk archive and catalog. The DMF is designed as a distributed system using a client‐server model, wherein multiple copies of the User Interface communicate with the single copies of the two servers.This paper is a case study of the DMF project. We show how the requirements for long life (30 years) and the graceful evolution to new technology for databases, archiving media, and network protocols are accommodated and discuss the lessons learned and their implications for future...

This paper presents work in process. It is intended to elicit discussion and debate. Our premise is that there is a great need for a unifying theory and morphology of science data. Present practice and theory does not address the semantic nature of data. Furthermore, the heuristic nature of present efforts may be counter productive. Encapsulating semantic knowledge with data promotes productivity, longevity, and communication.The final portion of the paper presents an initial morphology of data types. It is not yet blown into a full semantic model. The technology used in this conceptual model is an object‐oriented paradigm. We do not have an implementation proposal. Such efforts must await the completion of the theory.

Software, both source and executable versions, designed using LR parser generator technology offers superior data compression characteristics both with respect to the compression ratio, which is essentially optimal, and computational resources efficiency. Ten‐fold compression should make possible highly sophisticated command and control systems for use during a space mission, on route and at the target. Carrying source in compressed form on manned missions will mean that significant modifications and even redesign of large systems can be accomplished. Basic aspects of this compression scheme are described in the paper.

ARGUS is a three-dimensional, electromagnetic, particle-in-cell (PIC) simulation code that is being distributed to U.S. accelerator laboratories in collaboration between SAIC and the Los Alamos Accelerator Code Group. It uses a modular architecture that allows multiple physics modules to share common utilities for grid and structure input., memory management, disk I/O, and diagnostics, Physics modules are in place for electrostatic and electromagnetic field solutions., frequency-domain (eigenvalue) solutions, time- dependent PIC, and steady-state PIC simulations. All of the modules are implemented with a domain-decomposition architecture that allows large problems to be broken up into pieces that fit in core and that facilitates the adaptation of ARGUS for parallel processing ARGUS operates on either Cray or workstation platforms, and MOTIF-based user interface is available for X-windows terminals. Applications of ARGUS in accelerator physics and design are described in this paper.

There are many well developed codes available for beamline design and analysis. A significant fraction of each of these codes is devoted to processing its own unique input language for describing the problem. None of these large, complex, and powerful codes does everything. Adding a new bit of specialized physics can be a difficult task whose successful completion makes the code even larger and more complex. This paper describes an attempt to move in the opposite direction, toward a family of small, simple, single purpose physics and utility modules, linked by an open, portable, public domain database framework. These small specialized physics codes begin with the beamline parameters already loaded in the database, and accessible via the handful of subroutines that constitute ADLIB. Such codes are easier to write, and inherently organized in a manner suitable for incorporation in model based control system algorithms. Examples include programs for analyzing beamline misalignment sensitivities, for simulating and fitting beam steering data, and for translating among MARYLIE, TRANSPORT, and TRACE3D formats.

This paper presents work on a scalable software architecture designed for high‐performance local and wide‐area networks of parallel computers. The goal of the work is to support operational image understanding systems with processing requirements greater than that available from any single computer. The top‐level environment supports multiple threads in a shared address space. Each task can be a sequential function or a fully data parallel function. Data parallel functions are executed by physical SIMD computers or by MIMD computers supporting data parallel libraries. Scheduling is handled opportunistically under low‐load conditions and using cost functions and priorities under high‐load conditions. By providing a range of performance options for all data parallel functions and making scheduling decisions dynamically, we can achieve higher overall system performance than possible with more rigid application structures.

The need for large scale parallel computer systems has been intensified by the number of applications which require manipulation of large amount of data. Many scientific and engineering applications require high performance parallel computers to accomplish the computational tasks. This paper presents a multiple bus architecture capable of being scaled to massive dimension, to handle large amounts of data. The memory bandwidth of this architecture is analyzed. This architecture has low latency and high bus bandwidth. Also the architecture can enforce cache consistency through bus snooping cache coherence protocols.

The Infrared Astronomical Satellite (IRAS) was operated in 1983 to make an all‐sky catalog of point‐like celestial sources. The performance of the satellite, which was much better than specified, combined with 10 years of progress in computer technology and data analysis techniques has made it possible to derive ever improving products from the original data set. Publishing of, or providing access to standard products cannot keep up with the newest developments. This paper describes a service that is able to provide tailored analysis results from the original, unprocessed data, allowing the scientific community to profit immediately from any progress.