Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “parallel, scaling”

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 1,189 records · Page 66

Multi-operator Production Cost Modeling

As contiguous power systems are often operated by several regional system operators, optimization models have traditionally been designed to generate optimal system-wide scheduling decisions. This report introduces an approach to reflect the existence of multiple system operators making scheduling decisions within an interconnected power system. The approach is explicitly designed to reflect each operator's limited knowledge about other regions when making scheduling decisions. Detailed analysis using a modified IEEE Reliability Test System reveals that the approach mimics real system operations by providing imperfect information to each operator about external operations and external constraints. The approach scales to a 78,000-bus model of the Eastern Interconnection, where a regression analysis provides quantitative insight into which constraints cause inefficiencies when improperly managed in a multi-operator simulation. Because the approach generates distinct regional scheduling problems, unit commitment and economic dispatch simulations can be applied to individual regions and solved in parallel, enabling production cost simulations at a previously intractable scale.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Distributed Halide

Many image processing tasks are naturally expressed as a pipeline of small computational kernels known as stencils. Halide is a popular domain-specific language and compiler designed to implement image processing algorithms. Halide uses simple language constructs to express what to compute and a separate scheduling co-language for expressing when and where to perform the computation. This approach has demonstrated performance comparable to or better than hand-optimized code. Until now, however, Halide has been restricted to parallel shared memory execution, limiting its performance for memory-bandwidth-bound pipelines or large-scale image processing tasks. We present an extension to Halide to support distributed-memory parallel execution of complex stencil pipelines. These extensions compose with the existing scheduling constructs in Halide, allowing expression of complex computation and communication strategies. Existing Halide applications can be distributed with minimal changes, allowing programmers to explore the tradeoff between recomputation and communication with little effort. Approximately 10 new of lines code are needed even for a 200 line, 99 stage application. On nine image processing benchmarks, our extensions give up to a 1.4× speedup on a single node over regular multithreaded execution with the same number of cores, by mitigating the effects of non-uniform memory access. The distributed benchmarks achieve up to 18× speedup on a 16 node testing machine and up to 57× speedup on 64 nodes of the NERSC Cori supercomputer.

97 MATHEMATICS AND COMPUTING↗

Gyrokinetic simulation of global and local Alfvén eigenmodes driven by energetic particles in a DIII-D discharge

The unstable spectrum of Alfvén eigenmodes (AEs) driven by neutral beam-sourced energetic particles (EPs) in a benchmark DIII-D discharge (142111) is calculated in a fully gyrokinetic model using the GYRO code's massively parallel linear eigenvalue solver. One cycle of the slow (equilibrium scale) frequency sweep of the reverse shear Alfvén eigenmode (RSAE) at toroidal mode number n=3 is mapped. The RSAE second harmonic and an unstable beta-induced Alfvén eigenmode (BAE) are simultaneously tracked alongside the primary RSAE. An observed twist in the eigenmode pattern, caused mostly by shear in the driving EP profile, is shown through artificially varying the E×B rotational velocity shear to depend generally on shear in the local wave phase velocity. Coupling to the BAE and to the toroidal Alfvén eigenmode limit the RSAE frequency sweeps at the lower and upper end, respectively. While the present fully gyrokinetic model (including thermal ions and electrons) constitutes the best treatment of compressibility physics available, the BAE frequency is overpredicted by about 20% against experiment here and is found to be sensitive to energetic beam ion pressure. The RSAE frequency is more accurately matched except when it is limited by the BAE. Simulations suggest that the experiment is very close to marginal AE stability at points of RSAE-BAE coupling. A recipe for comparing the radial profile of quasilinear transport flux from local modes to that from global modes paves the way for the development of a stiff (critical gradient) local AE transport model based on local mode stability thresholds.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

T-FSM: A Scalable Distributed Task-Based System for Frequent Subgraph Pattern Mining from a Big Graph

Finding frequent subgraph patterns in a big graph is an important problem with many applications such as classifying chemical compounds and building indexes to speed up graph queries. Since this problem is NP-hard, some recent parallel and distributed systems have been developed to accelerate the mining. However, they often have a huge memory cost, very long running time, suboptimal load balancing, poor scale-out capability, and possibly inaccurate results. In this article, we propose an efficient system called T-FSM for parallel mining of frequent subgraph patterns in a big graph. T-FSM supports a new anti-monotonic frequentness measure called Fraction-Score, which is more accurate than the widely used MNI measure. The execution engine of T-FSM supports both intra-machine parallelism and inter-machine parallelism. For intra-machine parallelism, T-FSM adopts a novel task-based execution model to ensure high multithreading concurrency, bounded memory consumption, and effective load balancing. For inter-machine parallelism, T-FSM ensures good scale-out performance with a lightweight pattern rebalancing approach that reduces workload skewness of pattern evaluations among machines. To avoid recomputing the contexts for migrated patterns, we design a novel context cache table to support concurrent and asynchronous requesting and caching of remote context data, which can timely evict and garbage collect used pattern contexts that are no longer needed to keep memory consumption bounded. Extensive experiments show that T-FSM is orders of magnitude faster than existing state-of-the-art parallel systems (more than 10×, 51×, 131×, 55× speedup over ScaleMine, DistGraph, Pangolin and Peregrine, respectively) and distributed systems (more than 42× and 88× over ScaleMine and DistGraph, respectively) for frequent subgraph pattern mining, and it scales out satisfactorily to 512 CPU cores on the Polaris supercomputer at Argonne National Laboratory.

97 MATHEMATICS AND COMPUTING↗

Using parallel banded linear system solvers in generalized eigenvalue problems

Subspace iteration is a reliable and cost effective method for solving positive definite banded symmetric generalized eigenproblems, especially in the case of large scale problems. This paper discusses an algorithm that makes use of two parallel banded solvers in subspace iteration. A shift is introduced to decompose the banded linear systems into relatively independent subsystems and to accelerate the iterations. With this shift, an eigenproblem is mapped efficiently into the memories of a multiprocessor and a high speed-up is obtained for parallel implementations. An optimal shift is a shift that balances total computation and communication costs. Under certain conditions, we show how to estimate an optimal shift analytically using the decay rate for the inverse of a banded matrix, and how to improve this estimate. Computational results on iPSC/2 and iPSC/860 multiprocessors are presented.

Zhang, Hong↗

Using parallel banded linear system solvers in generalized eigenvalue problems

Subspace iteration is a reliable and cost effective method for solving positive definite banded symmetric generalized eigenproblems, especially in the case of large scale problems. This paper discusses an algorithm that makes use of two parallel banded solvers in subspace iteration. A shift is introduced to decompose the banded linear systems into relatively independent subsystems and to accelerate the iterations. With this shift, an eigenproblem is mapped efficiently into the memories of a multiprocessor and a high speedup is obtained for parallel implementations. An optimal shift is a shift that balances total computation and communication costs. Under certain conditions, we show how to estimate an optimal shift analytically using the decay rate for the inverse of a banded matrix, and how to improve this estimate. Computational results on iPSC/2 and iPSC/860 multiprocessors are presented.

DISTRIBUTED MEMORY MULTIPROCES↗

CO 2 Capture by Cold Membrane Operation with Actual Power Plant Flue Gas

The main objective of the project was to develop a post-combustion CO 2 capture process based on the hybrid cold temperature membrane operation. The CO 2 in the flue gas from coal fired power plant is pre-concentrated to >60% CO 2 in the first stage membrane operation followed by further liquefaction of permeate stream to achieve >99% CO 2 purity. The aim of the project was based on DOE program goal of 90% CO 2 capture with >95% CO 2 purity from Pulverized Coal (PC) fired power plants with $40/tonne of carbon capture cost by 2025. The project moves the technology from TRL 4 to TRL 5. The project involved optimization of Air Liquide commercial 12” PI-1 bundle to improve the bundle productivity by >30% compared to the previous baseline (DE-FE0004278) using computational fluid dynamics (CFD) modeling and bundle testing with synthetic flue gas at 0.1 MWe bench scale skid located at Delaware Research and Technology Center (DRTC). In parallel, the next generation polyimide based novel PI-2 membrane was developed with 10 times CO 2 permeance compared to the commercial PI-1 membrane. The novel PI-2 membrane was scaled from mini-permeator to 1” permeator and 1” bundle for testing. Bundle development was conducted with a Development Spin Unit (DSU) installed at MEDAL. Air Liquide’s cold membrane technology was demonstrated with real coal fired flue gas at the National Carbon Capture Center (NCCC) with a 0.3 MWe field-test unit (FTU). The FTU was designed to incorporate testing of two PI-1 commercial membrane bundles (12” or 6” diameter) in parallel or series. A slip stream was sent to the next generation PI-2 membrane for testing with real flue gas. The system exceeded performance targets with stable PI-1 membrane operation for over 500 hours of single bundle, steady state testing. The 12” PI-1 bundle exceeded the productivity target by achieving ~600 Nm3/hr, where the target was set at ~455 Nm3/hr at 90% capture rate. The cost of 90% CO 2 capture from a 550 MWe net coal power plant was estimated between 40 and $45/tonne. A 6” PI-1 bundle exhibited superior bundle performance compared to the 12” PI-1 bundle. However, the carbon capture cost was not lower with the 6” PI-1 bundle due to the higher bundle installed cost. A 1” PI-1 bundle was tested to compare bundles with different length / diameter ratios. This bundle exhibited the lowest performance due to the different fiber winding pattern and increased bundle non-ideality. Several long-term and parametric tests were conducted with 3,200 hours of total run-time at NCCC. Finally, the new PI-2 membrane fiber was tested at a small scale (1” modules) in real flue gas and exhibited up to 10 times the CO 2 permeance and slightly lower CO 2 /N 2 selectivity as the commercial PI-1 fiber. This corresponded to a projected 4 - 5 times increase in the productivity per bundle and a potential cost reduction of $3/tonne for CO2 capture, as compared with PI-1. An analytical campaign was conducted to trace different impurities such as NOx, mercury, Arsenic, Selenium in gas and liquid samples through the carbon capture system. An Environmental, Health and Safety (EH&S) analysis was completed to estimate emissions from a 550 MWe net power plant with carbon capture using cold membrane. A preliminary design and cost analysis was completed for 550 tpd (~25 MWe) plant to assess the capital investment and carbon capture cost for PI-1 and PI-2 membrane solutions from coal fired flue gas. A comparison was made with an amine based solution with significant cost advantage for the membrane at this scale. Additional preliminary design and cost analysis was completed between coal, natural gas and SMR flue gas for carbon capture at 550 tpd (~25 MWe) plant.

01 COAL, LIGNITE, AND PEAT↗

Towards Generic Parallel Programming in Computer Science Education with Kokkos

Parallel patterns, views, and spaces are promising abstractions to capture the programmer's intent as well as the contextual information that can be used by an underlying runtime to efficiently map software to parallel hardware. These abstractions can be valuable in cases where an algorithm must accommodate requirements of code and performance portability across hardware architectures and vendor programming models. Kokkos is a parallel programming model for host- and accelerator architectures that relies on these abstractions and targets these requirements. It consists of a pure C++ interface, a specification, and a programming library. The programming library exposes patterns and types and maps them to an underlying abstract machine model. The abstract machine model offers a generic view of parallel hardware. While Kokkos is gaining popularity in large-scale HPC applications at some DOE laboratories, we believe that the implemented concepts are of interest to a broader audience including academia as they may contribute to a generic, vendor, and architecture-independent education of parallel programming. In this work, we give an insight into the design considerations of this programming model and list important abstractions. Further, we document best practices obtained from giving virtual classes on Kokkos and give pointers to resources that the reader may consider valuable for a lecture on generic parallel programming for students with preexisting knowledge on this matter.

Ciesko, Jan↗

A Parallel Alternative for Energy-Efficient Neural Network Training and Inferencing

Energy efficiency of training and inferencing with large neural network models is a critical challenge facing the future of sustainable large-scale machine learning workloads. This paper introduces an alternative strategy, called phantom parallelism, to minimize the net energy consumption of traditional tensor (model) parallelism, the most energy-inefficient component of large neural network training. The approach is presented in the context of feed-forward network architectures as a preliminary, but comprehensive, proof-of-principle study of the proposed methodology. We derive new forward and backward propagation operators for phantom parallelism, implement them as custom autograd operations within an end-to-end phantom parallel training pipeline and compare its parallel performance and energy-efficiency against those of conventional tensor parallel training pipelines. Formal analyses that predict lower bandwidth and FLOP counts are presented with supporting empirical results on up to 256 GPUs that corroborate these gains. Experiments are shown to deliver ∼50% reduction in the energy consumed to train FFNs using the proposed phantom parallel approach when compared with conventional tensor parallel methods. Additionally, the proposed approach is shown to train smaller phantom models to the same model loss on smaller GPU counts as larger tensor parallel models on larger GPU counts offering the possibility for even greater energy savings.

Seal, Sudip [ORNL] (ORCID:0000000332330656)↗

IOPro: a parallel I/O profiling and visualization framework for high-performance storage systems

Efficient execution of large-scale scientific applications requires high-performance computing systems designed to meet the I/O requirements. To achieve high-performance, such data-intensive parallel applications use a multi-layer layer I/O software stack, which consists of high-level I/O libraries such as PnetCDF and HDF5, the MPI library, and parallel file systems. To design efficient parallel scientific applications, understanding the complicated flow of I/O operations and the involved interactions among the libraries is quintessential. Such comprehension helps identify I/O bottlenecks and thus exploits the potential performance in different layers of the storage hierarchy. To profile the performance of individual components in the I/O stack and to understand complex interactions among them, we have implemented a GUI-based integrated profiling and analysis framework, IOPro. IOPro automatically generates an instrumented I/O stack, runs applications on it, and visualizes detailed statistics based on the user-specified metrics of interest. We present experimental results from two different real-life applications and show how our framework can be used in practice. By generating an end-to-end trace of the whole I/O stack and pinpointing I/O interference, IOPro aids in understanding I/O behavior and improving the I/O performance significantly.

97 MATHEMATICS AND COMPUTING↗

Mesoscale and large-scale variability of the Antarctic circumpolar current

This investigation of the physical oceanography of the Southern Ocean will carry out two parallel efforts during the years preceding the launch of TOPEX/POSEIDON. First, the Geosat data will be used to develop a preliminary descriptive picture of the mesoscale and large-scale, low-frequency surface circulation of the Southern Ocean. Some of this analysis of Geosat data has already begun. For example, as a measure of the geographical distribution of mesoscale variability, a color-coded map of the standard deviation of sea level from two years of Geosat data is shown. Efforts are presently under way to investigate the seasonal and year-to-year variability of this mesoscale energy. The data are also being used to generate low-pass filtered fields of sea level from which the temporal evolution of large-scale variability in the Southern Ocean may be investigated. The second parallel effort is the development and test of modeling and data assimilation techniques that will later be applied to TOPEX/POSEIDON data during the postlaunch phase. One objective of the modeling and data assimilation is to investigate the relation between mesoscale sea level variations and eddy flux in the Southern Ocean. Uncertainties in present estimates of the various components of meridional oceanic heat transport are large. The evidence presented indicates that the very energetic mesoscale variability in the ACC apparently accounts for much of the estimated 0.45x10(exp 15) watts of poleward heat transport across the ACC required to balance the heat budget. Eddy variability is strongly coherent vertically in the ACC, at least in the vicinity of Drake Passage where nearly all of the historical in situ data have been collected.

Chelton, Dudley B.↗

Dynamic model-based sensor network design algorithm for system efficiency maximization

Here, a dynamic model-based sensor network design (DMSND) algorithm has been developed for maximizing system efficiency for an estimator-based control system. The algorithm synthesizes the optimal sensor network in the face of disturbances or set point changes. Computational expense of the large-scale combinatorial optimization problem is significantly reduced by parallel computing and by using combination of three novel strategies: multi-rate sampling frequency, model order reduction, and use of an incumbent solution that enables early termination of evaluation of infeasible sensor sets. The developed algorithm is applied to an acid gas removal unit as part of an integrated gasification combined cycle power plant with carbon capture. Even though there are more than thousand process states and more than hundred candidate sensor locations, the optimal sensor network design problem for maximizing process efficiency could be solved within couple of hours for a given budget.

47 OTHER INSTRUMENTATION↗

Accelerating nuclear configuration interaction calculations through a preconditioned block iterative eigensolver

We describe a number of recently developed techniques for improving the performance of large-scalenuclear configuration interaction calculations on high performance parallel computers. We show thebenefit of using a preconditioned block iterative method to replace the Lanczos algorithm that hastraditionally been used to perform this type of computation. The rapid convergence of the block iterativemethod is achieved by a proper choice of starting guesses of the eigenvectors and the construction of aneffective preconditioner. These acceleration techniques take advantage of special structure of the nuclearconfiguration interaction problem which we discuss in detail. The use of a block method also allows usto improve the concurrency of the computation, and take advantage of the memory hierarchy of modernmicroprocessors to increase the arithmetic intensity of the computation relative to data movement. Wealso discuss the implementation details that are critical to achieving high performance on massivelyparallel multi-core supercomputers, and demonstrate that the new block iterative solver is two to threetimes faster than the Lanczos based algorithm for problems of moderate sizes on a Cray XC30 system.

Shao, Meiyue↗

Billion-atom synchronous parallel kinetic Monte Carlo simulations of critical 3D Ising systems

An extension of the synchronous parallel kinetic Monte Carlo (spkMC) algorithm developed by Martinez et al. [J. Comp. Phys. 227 (2008) 3804] to discrete lattices is presented. The method solves the master equation synchronously by recourse to null events that keep all processors' time clocks current in a global sense. Boundary conflicts are resolved by adopting a chessboard decomposition into non-interacting sublattices. We find that the bias introduced by the spatial correlations attendant to the sublattice decomposition is within the standard deviation of serial calculations, which confirms the statistical validity of our algorithm. We have analyzed the parallel efficiency of spkMC and find that it scales consistently with problem size and sublattice partition. We apply the method to the calculation of scale-dependent critical exponents in billion-atom 3D Ising systems, with very good agreement with state-of-the-art multispin simulations.

97 MATHEMATICS AND COMPUTING↗

A generalized parallel replica dynamics

Metastability is a common obstacle to performing long molecular dynamics simulations. Many numerical methods have been proposed to overcome it. One method is parallel replica dynamics, which relies on the rapid convergence of the underlying stochastic process to a quasi-stationary distribution. Two requirements for applying parallel replica dynamics are knowledge of the time scale on which the process converges to the quasi-stationary distribution and a mechanism for generating samples from this distribution. By combining a Fleming–Viot particle system with convergence diagnostics to simultaneously identify when the process converges while also generating samples, we can address both points. This variation on the algorithm is illustrated with various numerical examples, including those with entropic barriers and the 2D Lennard-Jones cluster of seven atoms.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Optical neural engine for solving scientific partial differential equations

Abstract Solving partial differential equations (PDEs) is the cornerstone of scientific research and development. Data-driven machine learning (ML) approaches are emerging to accelerate time-consuming and computation-intensive numerical simulations of PDEs. Although optical systems offer high-throughput and energy-efficient ML hardware, their demonstration for solving PDEs is limited. Here, we present an optical neural engine (ONE) architecture combining diffractive optical neural networks for Fourier space processing and optical crossbar structures for real space processing to solve time-dependent and time-independent PDEs in diverse disciplines, including Darcy flow equation, the magnetostatic Poisson’s equation in demagnetization, the Navier-Stokes equation in incompressible fluid, Maxwell’s equations in nanophotonic metasurfaces, and coupled PDEs in a multiphysics system. We numerically and experimentally demonstrate the capability of the ONE architecture, which not only leverages the advantages of high-performance dual-space processing for outperforming traditional PDE solvers and being comparable with state-of-the-art ML models but also can be implemented using optical computing hardware with unique features of low-energy and highly parallel constant-time processing irrespective of model scales and real-time reconfigurability for tackling multiple tasks with the same architecture. The demonstrated architecture offers a versatile and powerful platform for large-scale scientific and engineering computations.

Tang, Yingheng (ORCID:0009000153622546)↗

Solar wind compressible turbulence near proton scales: Cluster observations

The solar wind compressible turbulence attracts more attention recently due to its possible role in plasma heating. Using the plasma density and magnetic field measured simultaneously by the Cluster C1 probe, we statistically investigate the wave characteristics of the proton-scale compressible turbulence. In most cases, the density and magnetic field strength fluctuations almost show no positive correlation, suggesting the absence of fast mode waves. Moreover, the magnetic compressibility C{sub b} as measured by the level of magnetic fluctuations parallel to the background field, is enhanced continuously around proton scales and is consistent with kinetic Alfvén wave (KAW) prediction. Interestingly, kinetic slow mode waves (KSW) are identified in a few cases through large values of C{sub b} and plasma compressibility C{sub p}, together with the the anti-correlation between n and B. We suggest that both KAW and KSW can account for the compressible fluctuations energy, while their relative proportions under certain plasma condition require further studies.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗