Search NASA⌕ Search

SEARCH · Search NASA

Results for “Parallelization efficiency”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,045 records · Page 58

Full core reactor analysis: Running Denovo on Jaguar

Fully-consistent, full-core, 3D, deterministic neutron transport simulations using the orthogonal mesh code Denovo were run on the massively parallel computing architecture Jaguar XT5. Using energy and spatial parallelization schemes, Denovo was able to efficiently scale to more than 160 k processors. Cell-homogenized cross sections were used with step-characteristics, linear-discontinuous finite element, and trilinear-discontinuous finite element spatial methods. It was determined that using the finite element methods gave considerably more accurate eigenvalue solutions for large-aspect ratio meshes than using step-characteristics. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

DOMINO: A fast 3D cartesian discrete ordinates solver for reference PWR simulations and SPN validation

As part of its activity, EDF R and D is developing a new nuclear core simulation code named COCAGNE. This code relies on DIABOLO, a Simplified PN (SPN) method to compute the neutron flux inside the core for eigenvalue calculations. In order to assess the accuracy of SPN calculations, we have developed DOMINO, a new 3D Cartesian SN solver. The parallel implementation of DOMINO is very efficient and allows to complete an eigenvalue calculation involving around 300 x 10{sup 9} degrees of freedom within a few hours on a single shared-memory supercomputing node. This computation corresponds to a 26-group S{sub 8} 3D PWR core model used to assess the SPN accuracy. At the pin level, the maximal error for the SP{sub 5} DIABOLO fission production rate is lower than 0.2% compared to the S{sub 8} DOMINO reference for this 3D PWR core model. (authors)

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Dynamically tuned high-Q AC-dipole implementation

AC-dipole magnets are typically implemented as a parallel LC resonant circuit. To maximize efficiency, it's beneficial to operate at a high Q. This, however, limits the magnet to a narrow frequency range. Current designs therefore operate at a low Q to provide a wider bandwidth at the cost of efficiency. Dynamically tuning a high Q resonant circuit tries to maintain a high efficiency while providing a wide frequency range. The results of ongoing efforts at BNL to implement dynamically tuned high-Q AC dipoles will be presented.

43 PARTICLE ACCELERATORS↗

Scalable Graph Analytics and HPC Operational Enhancement: Parallel Computing and ML/DL Innovations

Parallel computing plays a pivotal role in the efficient processing of large-scale graphs. Complex network analysis stands as a capti- vating research frontier, holding promise across diverse scientific domains such as sociology, biology, online media, and recommenda- tion systems. In this era, Machine Learning (ML) and Deep Learning (DL) have emerged as indispensable tools, underpinning remarkable technological achievements. Within this dynamic landscape, my research revolves around advancing parallel algorithms tailored for large-scale graph operations. To achieve this, I harness the power of cutting-edge technologies including OpenMP, MPI, HIP, and CUDA, on the High-Performance Computing (HPC) platforms to unlock optimal performance. I also apply ML/DL techniques to HPC operational data, to streamline the monitoring and maintenance of supercomputers, alleviating the complexities associated with their upkeep and enhancing user support. My research echoes the syn- ergy between parallel computing, large-scale graph analysis, and ML/DL, improving computational efficiency and user experience.

Sattar, Naw Safrin↗

SiC Multi-Chip Power Modules as Power-System Building Blocks

The term "SiC MCPMs" (wherein "MCPM" signifies "multi-chip power module") denotes electronic power-supply modules containing multiple silicon carbide power devices and silicon-on-insulator (SOI) control integrated-circuit chips. SiC MCPMs are being developed as building blocks of advanced expandable, reconfigurable, fault-tolerant power-supply systems. Exploiting the ability of SiC semiconductor devices to operate at temperatures, breakdown voltages, and current densities significantly greater than those of conventional Si devices, the designs of SiC MCPMs and of systems comprising multiple SiC MCPMs are expected to afford a greater degree of miniaturization through stacking of modules with reduced requirements for heat sinking. Moreover, the higher-temperature capabilities of SiC MCPMs could enable operation in environments hotter than Si-based power systems can withstand. The stacked SiC MCPMs in a given system can be electrically connected in series, parallel, or a series/parallel combination to increase the overall power-handling capability of the system. In addition to power connections, the modules have communication connections. The SOI controllers in the modules communicate with each other as nodes of a decentralized control network, in which no single controller exerts overall command of the system. Control functions effected via the network include synchronization of switching of power devices and rapid reconfiguration of power connections to enable the power system to continue to supply power to a load in the event of failure of one of the modules. In addition to serving as building blocks of reliable power-supply systems, SiC MCPMs could be augmented with external control circuitry to make them perform additional power-handling functions as needed for specific applications: typical functions could include regulating voltages, storing energy, and driving motors. Because identical SiC MCPM building blocks could be utilized in a variety of ways, the cost and difficulty of designing new, highly reliable power systems would be reduced considerably. Several prototype DC-to-DC power-converter modules containing SiC power-switching devices were designed and built to demonstrate the feasibility of the SiC MCPM concept. In anticipation of a future need for operation at high temperature, the circuitry in the modules includes high-temperature inductors and capacitors. These modules were designed to be stacked to construct a system of four modules electrically connected in series and/or parallel. The packaging of the modules is designed to satisfy requirements for series and parallel interconnection among modules, high power density, high thermal efficiency, small size, and light weight. Each module includes four output power connectors two for serial and two for parallel output power connections among the modules. Each module also includes two signal connectors, electrically isolated from the power connectors, that afford four zones for signal interconnections among the SOI controllers. Finally, each module includes two input power connectors, through which it receives power from an in-line power bus. This design feature is included in anticipation of a custom-designed power bus incorporating sockets compatible with snap-on type connectors to enable rapid replacement of failed modules.

Lostetter, Alexander↗

Layer-Parallel Training of Deep Residual Neural Networks

Residual neural networks (ResNets) are a promising class of deep neural networks that have shown excellent performance for a number of learning tasks, e.g., image classification and recognition. Mathematically, ResNet architectures can be interpreted as forward Euler discretizations of a nonlinear initial value problem whose time-dependent control variables represent the weights of the neural network. Hence, training a ResNet can be cast as an optimal control problem of the associated dynamical system. For similar time-dependent optimal control problems arising in engineering applications, parallel-in-time methods have shown notable improvements in scalability. This paper demonstrates the use of those techniques for efficient and effective training of ResNets. The proposed algorithms replace the classical (sequential) forward and backward propagation through the network layers with a parallel nonlinear multigrid iteration applied to the layer domain. This adds a new dimension of parallelism across layers that is attractive when training very deep networks. From this basic idea, we derive multiple layer-parallel methods. The most efficient version employs a simultaneous optimization approach where updates to the network parameters are based on inexact gradient information in order to speed up the training process. Finally, using numerical examples from supervised classification, we demonstrate that the new approach achieves a training performance similar to that of traditional methods, but enables layer-parallelism and thus provides speedup over layer-serial methods through greater concurrency.

97 MATHEMATICS AND COMPUTING↗

Algorithms for parallel flow solvers on message passing architectures

The purpose of this project has been to identify and test suitable technologies for implementation of fluid flow solvers -- possibly coupled with structures and heat equation solvers -- on MIMD parallel computers. In the course of this investigation much attention has been paid to efficient domain decomposition strategies for ADI-type algorithms. Multi-partitioning derives its efficiency from the assignment of several blocks of grid points to each processor in the parallel computer. A coarse-grain parallelism is obtained, and a near-perfect load balance results. In uni-partitioning every processor receives responsibility for exactly one block of grid points instead of several. This necessitates fine-grain pipelined program execution in order to obtain a reasonable load balance. Although fine-grain parallelism is less desirable on many systems, especially high-latency networks of workstations, uni-partition methods are still in wide use in production codes for flow problems. Consequently, it remains important to achieve good efficiency with this technique that has essentially been superseded by multi-partitioning for parallel ADI-type algorithms. Another reason for the concentration on improving the performance of pipeline methods is their applicability in other types of flow solver kernels with stronger implied data dependence. Analytical expressions can be derived for the size of the dynamic load imbalance incurred in traditional pipelines. From these it can be determined what is the optimal first-processor retardation that leads to the shortest total completion time for the pipeline process. Theoretical predictions of pipeline performance with and without optimization match experimental observations on the iPSC/860 very well. Analysis of pipeline performance also highlights the effect of uncareful grid partitioning in flow solvers that employ pipeline algorithms. If grid blocks at boundaries are not at least as large in the wall-normal direction as those immediately adjacent to them, then the first processor in the pipeline will receive a computational load that is less than that of subsequent processors, magnifying the pipeline slowdown effect. Extra compensation is needed for grid boundary effects, even if all grid blocks are equally sized.

Vanderwijngaart, Rob F.↗

Interval Management with Spacing to Parallel Dependent Runways (IMSPIDR) Experiment and Results

An area in aviation operations that may offer an increase in efficiency is the use of continuous descent arrivals (CDA), especially during dependent parallel runway operations. However, variations in aircraft descent angle and speed can cause inaccuracies in estimated time of arrival calculations, requiring an increase in the size of the buffer between aircraft. This in turn reduces airport throughput and limits the use of CDAs during high-density operations, particularly to dependent parallel runways. The Interval Management with Spacing to Parallel Dependent Runways (IMSPiDR) concept uses a trajectory-based spacing tool onboard the aircraft to achieve by the runway an air traffic control assigned spacing interval behind the previous aircraft. This paper describes the first ever experiment and results of this concept at NASA Langley. Pilots flew CDAs to the Dallas Fort-Worth airport using airspeed calculations from the spacing tool to achieve either a Required Time of Arrival (RTA) or Interval Management (IM) spacing interval at the runway threshold. Results indicate flight crews were able to land aircraft on the runway with a mean of 2 seconds and less than 4 seconds standard deviation of the air traffic control assigned time, even in the presence of forecast wind error and large time delay. Statistically significant differences in delivery precision and number of speed changes as a function of stream position were observed, however, there was no trend to the difference and the error did not increase during the operation. Two areas the flight crew indicated as not acceptable included the additional number of speed changes required during the wind shear event, and issuing an IM clearance via data link while at low altitude. A number of refinements and future spacing algorithm capabilities were also identified.

Baxley, Brian T.↗

Extending OpenSHMEM with Aggregation Support for Improved Message Rate Performance

OpenSHMEM is a highly efficient one-sided communication API that implements the PGAS parallel programming model, and is known for its low latency communication operations that can be mapped efficiently to RDMA capabilities of network interconnects. However, applications that use OpenSHMEM can be sensitive to point-to-point message rates, as many-to-many communication patterns can generate large amounts of small messages which tend to overwhelm network hardware that has predominantly been optimised for bandwidth over message rate. Additionally, many important emerging classes of problems such as data analytics are similarly troublesome for the irregular access patterns they employ. Message aggregation strategies have been proven to significantly enhance network performance, but their implementation often involves complex restructuring of user code, making them unwieldy. This paper shows how to combine the best qualities of message aggregation within the communication model of OpenSHMEM such that applications with small and irregular access patterns can improve network performance while maintaining their algorithmic simplicity. We do this by providing a path to a message aggregation framework called conveyors through a minimally intrusive OpenSHMEM extension introducing aggregation contexts that fit more naturally to the OpenSHMEM atomics, gets, and puts model. We test these extensions using four of the bale 3.0 applications which contain essential many-to-many access patterns to show how they can produce performance improvements of up to 65×.

Welch, Aaron↗

Massively Parallel Algorithms for Real-Time Wavefront Control of a Dense Adaptive Optics System

In this paper massively parallel algorithms and architectures for real-time wavefront control of a dense adaptive optic system (SELENE) are presented. We have already shown that the computation of a near optimal control algorithm for SELENE can be reduced to the solution of a discrete Poisson equation on a regular domain. Although this represents an optimal computation, due the large size of the system and the high sampling rate requirement, the implementation of this control algorithm poses a computationally challenging problem since it demands a sustained computational throughput of the order of 10 GFlops. We develop a novel algorithm, designated as Fast Invariant Imbedding algorithm, which offers a massive degree of parallelism with simple communication and synchronization requirements. Due to these features, our algorithm is significantly more efficient than other Fast Poisson Solvers for implementation on massively parallel architectures.

massively↗

Spin-orbit torques for current parallel and perpendicular to a domain wall

We report field- and current-induced domain wall (DW) depinning experiments in Ta\Co{sub 20}Fe{sub 60}B{sub 20}\MgO nanowires through a Hall cross geometry. While purely field-induced depinning shows no angular dependence on in-plane fields, the effect of the current depends crucially on the internal DW structure, which we manipulate by an external magnetic in-plane field. We show depinning measurements for a current sent parallel to the DW and compare its depinning efficiency with the conventional case of current flowing perpendicularly to the DW. We find that the maximum efficiency is similar for both current directions within the error bars, which is in line with a dominating damping-like spin-orbit torque (SOT) and indicates that no large additional torques arise for currents perpendicular to the DW. Finally, we find a varying dependence of the maximum depinning efficiency angle for different DWs and pinning levels. This emphasizes the importance of our full angular scans compared with previously used measurements for just two field directions (parallel and perpendicular to the DW) to determine the real torque strength and shows the sensitivity of the SOT to the precise DW structure and pinning sites.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Adaptive multiple super fast simulated annealing for stochastic microstructure reconstruction

Fast image reconstruction from statistical information is critical in image fusion from multimodality chemical imaging instrumentation to create high resolution image with large domain. Stochastic methods have been used widely in image reconstruction from two point correlation function. The main challenge is to increase the efficiency of reconstruction. A novel simulated annealing method is proposed for fast solution of image reconstruction. Combining the advantage of very fast cooling schedules, dynamic adaption and parallelization, the new simulation annealing algorithm increases the efficiencies by several orders of magnitude, making the large domain image fusion feasible.

adaptive multiple super fast simulated annealing↗

Parallel linear equation solvers for finite element computations

The overall objective of this research is to develop efficient methods for the solution of linear and nonlinear systems of equations on parallel and supercomputers, and to apply these methods to the solution of problems in structural analysis. Attention has been given so far only to linear equations. The methods considered for the solution of the stiffness equation Kx=f have been Choleski factorization and the conjugate gradient iteration with SSOR and Incomplete Choleski preconditioning. More detail on these methods will be given on subsequent slides. These methods have been used to solve for the static displacements for the mast and panel focus problems in conjunction with the CSM testbed system based on NICE/SPAR.

Ortega, James M.↗

Parallel Programming in MCNP6

Monte Carlo N-Particle (MCNP)1 is a general-purpose Monte Carlo particle transport code developed by Los Alamos National Laboratory (LANL). To efficiently handle long simulations, MCNP version 6 (MCNP6) supports parallel execution using two primary programming models: • Shared-memory task-based threading using OpenMP (Open Multi-Processing), and • Distributed-memory calculations using MPI (Message Passing Interface). The OpenMP and MPI programming models enable MCNP6 to scale from desktop systems to high-performance computing (HPC) clusters, allowing users to run MCNP in one of three parallel modes: • OpenMP-only, • MPI-only, and • Hybrid (MPI + OpenMP). The choice of parallelization mode depends on the underlying computer architecture and the characteristics of the simulation problem.

97 MATHEMATICS AND COMPUTING↗

Kinetic study of shock formation and particle acceleration in laser-driven quasi-parallel magnetized collisionless shocks

Quasi-parallel magnetized collisionless shocks are believed to be one of the most efficient accelerators in the universe. Compared to quasi-perpendicular shocks, quasi-parallel shocks are more difficult to form in the laboratory and to simulate because of their large spatial scales and long formation times. Our two-dimensional particle-in-cell simulations show that the early stages of quasi-parallel shock formation are achievable in experiments planned for the National Ignition Facility and that particles accelerated by diffusive shock acceleration (DSA) are expected to be observable in the experiment. Repetitive ion acceleration by crossings of the shock front, a key feature of DSA, is seen in the simulations. Other characteristic features of quasi-parallel shocks such as upstream wave excitation by energetic ions are also observed, and energy partition between the ions and the electrons in the downstream of the shock is briefly discussed.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Segmentation of remotely sensed data using parallel region growing

The improved spatial resolution of the new earth resources satellites will increase the need for effective utilization of spatial information in machine processing of remotely sensed data. One promising technique is scene segmentation by region growing. Region growing can use spatial information in two ways: only spatially adjacent regions merge together, and merging criteria can be based on region-wide spatial features. A simple region growing approach is described in which the similarity criterion is based on region mean and variance (a simple spatial feature). An effective way to implement region growing for remote sensing is as an iterative parallel process on a large parallel processor. A straightforward parallel pixel-based implementation of the algorithm is explored and its efficiency is compared with sequential pixel-based, sequential region-based, and parallel region-based implementations. Experimental results from on aircraft scanner data set are presented, as is a discussioon of proposed improvements to the segmentation algorithm.

Tilton, J. C.↗

Automatic identification and classification of Palomar Transient Factory astrophysical objects in GLADE

Palomar Transient Factory (PTF) is a comprehensive detection system for the identification and classification of transient astrophysical objects. In this paper, we make two significant contributions to the PTF pipeline. First, we present an experimental study that evaluates a novel implementation of the real-time classifier in GLADE - a parallel data processing system that combines the efficiency of a database with the extensibility of map-reduce. We show how each stage in the classifier maps optimally into GLADE tasks by taking advantage of the unique features of the system - range-based data partitioning, columnar storage, multi-query execution, and in-database support for complex aggregate computation. Second, we introduce a novel parallel similarity join algorithm for advanced transient classification. In conclusion, we implement this algorithm in GLADE and execute it on a massive supercomputer with more than 3,000 threads, achieving more than three orders of magnitude improvement over the PostgreSQL solution.

79 ASTRONOMY AND ASTROPHYSICS↗

Parallel methods for dynamic simulation of multiple manipulator systems

In this paper, efficient dynamic simulation algorithms for a system of m manipulators, cooperating to manipulate a large load, are developed; their performance, using two possible forms of parallelism on a general-purpose parallel computer, is investigated. One form, temporal parallelism, is obtained with the use of parallel numerical integration methods. A speedup of 3.78 on four processors of CRAY Y-MP8 was achieved with a parallel four-point block predictor-corrector method for the simulation of a four manipulator system. These multi-point methods suffer from reduced accuracy, and when comparing these runs with a serial integration method, the speedup can be as low as 1.83 for simulations with the same accuracy. To regain the performance lost due to accuracy problems, a second form of parallelism is employed. Spatial parallelism allows most of the dynamics of each manipulator chain to be computed simultaneously. Used exclusively in the four processor case, this form of parallelism in conjunction with a serial integration method results in a speedup of 3.1 on four processors over the best serial method. In cases where there are either more processors available or fewer chains in the system, the multi-point parallel integration methods are still advantageous despite the reduced accuracy because both forms of parallelism can then combine to generate more parallel tasks and achieve greater effective speedups. This paper also includes results for these cases.

Mcmillan, Scott↗