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

A signature of anisotropic cosmic-ray transport in the gamma-ray sky

A crucial process in Galactic cosmic-ray (CR) transport is the spatial diffusion due to the interaction with the interstellar turbulent magnetic field. Usually, CR diffusion is assumed to be uniform and isotropic all across the Galaxy. However, this picture is clearly inaccurate: several data-driven and theoretical arguments, as well as dedicated numerical simulations, show that diffusion exhibits highly anisotropic properties with respect to the direction of a background (ordered) magnetic field (i.e., parallel or perpendicular to it). In this paper we focus on a recently discovered anomaly in the hadronic CR spectrum inferred by the Fermi-LAT gamma-ray data at different positions in the Galaxy, i.e. the progressive hardening of the proton slope at low Galactocentric radii. We propose the idea that this feature can be interpreted as a signature of anisotropic diffusion in the complex Galactic magnetic field: in particular, the harder slope in the inner Galaxy is due, in our scenario, to the parallel diffusive escape along the poloidal component of the large-scale, regular, magnetic field. We implement this idea in a numerical framework, based on the DRAGON code, and perform detailed numerical tests on the accuracy of our setup. We discuss how the effect proposed depends on the relevant free parameters involved. Based on low-energy extrapolation of the few focused numerical simulations aimed at determining the scalings of the anisotropic diffusion coefficients, we finally present a set of plausible models that reproduce the behavior of the CR proton slopes inferred by gamma-ray data.

79 ASTRONOMY AND ASTROPHYSICS↗

Partitioning and Communication Strategies for Sparse Non-negative Matrix Factorization

Non-negative matrix factorization (NMF), the problem of finding two non-negative low-rank factors whose product approximates an input matrix, is a useful tool for many data mining and scientific applications such as topic modeling in text mining and unmixing in microscopy. In this paper, we focus on scaling algorithms for NMF to very large sparse datasets and massively parallel machines by employing effective algorithms, communication patterns, and partitioning schemes that leverage the sparsity of the input matrix. We consider two previous works developed for related problems, one that uses a fine-grained partitioning strategy using a point-to-point communication pattern and one that uses a Cartesian, or checkerboard, partitioning strategy using a collective-based communication pattern. We show that a combination of the previous approaches balances the demands of the various computations within NMF algorithms and achieves high efficiency and scalability. From the experiments, we see that our proposed strategy runs up to 10x faster than the state of the art on real-world datasets.

Kaya, Oguz↗

Julia as a unifying end-to-end workflow language on the Frontier exascale system

We evaluate Julia as a single language and ecosystem paradigm powered by LLVM to develop workflow components for high-performance computing. We run a Gray-Scott, 2-variable diffusion-reaction application using a memory-bound, 7-point stencil kernel on Frontier, the US Department of Energy’s first exascale supercomputer. We evaluate the performance, scaling, and trade-offs of (i) the computational kernel on AMD’s MI250x GPUs, (ii) weak scaling up to 4,096 MPI processes/GPUs or 512 nodes, (iii) parallel I/O writes using the ADIOS2 library bindings, and (iv) Jupyter Notebooks for interactive analysis. Results suggest that although Julia generates a reasonable LLVM-IR, a nearly 50% performance difference exists vs. native AMD HIP stencil codes when running on the GPUs. As expected, we observed near-zero overhead when using MPI and parallel I/O bindings for system-wide installed implementations. Consequently, Julia emerges as a compelling high-performance and high-productivity workflow composition language, as measured on the fastest supercomputer in the world.

Godoy, William↗

Axolotl: a scalable genomics library based on Apache Spark (Axolotl) v1.0.0

Axolotl is a Python library for scalable distributed genome and metagenome data analysis. Existing tools and systems that we rely on are struggling to keep up with the rapid explosion of genomic data. Compounding this issue, developing scalable solutions require a steep learning curve in parallel programming, which presents a barrier to academic researchers. While we do have scalable solutions for specific tasks, we lack comprehensive, end-to-end solutions. It's this gap in our toolkit that we aim to address with Axolotl. The Axolotl library is built for easy parallel processing, efficiently handling multiple tasks or large datasets simultaneously, and scaling up to meet the demands of extensive genomic data analysis.

Wang, Zhong↗

Compressed basis GMRES on high-performance graphics processing units

Krylov methods provide a fast and highly parallel numerical tool for the iterative solution of many large-scale sparse linear systems. To a large extent, the performance of practical realizations of these methods is constrained by the communication bandwidth in current computer architectures, motivating the investigation of sophisticated techniques to avoid, reduce, and/or hide the message-passing costs (in distributed platforms) and the memory accesses (in all architectures). This article leverages Ginkgo’s memory accessor in order to integrate a communication-reduction strategy into the (Krylov) GMRES solver that decouples the storage format (i.e., the data representation in memory) of the orthogonal basis from the arithmetic precision that is employed during the operations with that basis. Given that the execution time of the GMRES solver is largely determined by the memory accesses, the cost of the datatype transforms can be mostly hidden, resulting in the acceleration of the iterative step via a decrease in the volume of bits being retrieved from memory. Together with the special properties of the orthonormal basis (whose elements are all bounded by 1), this paves the road toward the aggressive customization of the storage format, which includes some floating-point as well as fixed-point formats with mild impact on the convergence of the iterative process. We develop a high-performance implementation of the “compressed basis GMRES” solver in the Ginkgo sparse linear algebra library using a large set of test problems from the SuiteSparse Matrix Collection. We demonstrate robustness and performance advantages on a modern NVIDIA V100 graphics processing unit (GPU) of up to 50% over the standard GMRES solver that stores all data in IEEE double-precision.

97 MATHEMATICS AND COMPUTING↗

EPSI 2013-2014. Annual Progress Report

Initially a separate orthogonal expansion for the low-k components and FEM for high-k components was considered, or a multi-scale FEM method, but a more attractive hybrid FDM/FEM model for parallel computations has been devised with minimal impact to formerly developed code. This system involves using Lagrange polynomials along magnetic field lines to establish a FDM discretization of the toroidal contributions to the perpendicular Laplacian, which is the operator of importance in gyrokinetic solution of the vector potential. The method was found analytically to be stable, due to some dissipation from varying field-line lengths. The accuracy of this method has been evaluated using the method of manufactured solutions (with Bei Wang of Princeton University), whereby we find that for a Poisson problem the maximum error roughly diminishes linearly as a function of the grid resolution (uniformly refined), whereas the L2 norm as square root of the resolution improvement. Most of the error appears to be contributed near boundaries and where the FEM grid has anisotropy, so we are trying to improve this analysis using improved tessellations obtained from Mark Shepard, and publish the results as soon as the problems with the grid have been eliminated.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Visualization and Data Analysis for High-Performance Computing

This is a set of slides from a guest lecture for a class at the University of Texas, El Paso on visualization and data analysis for high-performance computing. The topics covered are the following: trends in high-performance computing; scientific visualization, such as OpenGL, ray tracing and volume rendering, VTK, and ParaView; data science at scale, such as in-situ visualization, image databases, distributed memory parallelism, shared memory parallelism, VTK-m, "big data", and then an analysis example.

97 MATHEMATICS AND COMPUTING↗

Program Verification for Extreme-Scale Applications (Final Scientific/Technical Report)

The project seeks to develop tools and techniques to help software developers verify the correctness and accuracy of their code. The target domain is scientific software of the kind widely used and developed in the Department of Energy research community, with a particular focus on "extreme scale" programs - those that are expected to involve possibly millions of parallel threads of execution. The report covers the University of Delaware contribution to the collaborative project. The project had a number of successful outcomes, especially regarding the development and extension of the CIVL software verification framework. CIVL is a verification tool for C or Fortran programs that use MPI, OpenMP, CUDA, and/or Pthreads for parallelization. CIVL went through vast improvements and extensions, and was successfully applied to a number of challenging codes. It found a subtle bug in the Devito PDE framework. It was able to verify the functional correctness of a conjugate gradient solver using a novel probabilistic technique with vanishingly small chance of error. CIVL was used very successfully in verification competitions, and the CIVL solutions to the competition challenges were presented and published. CIVL was also used successfully by other researchers on a computational chemistry kernel.

97 MATHEMATICS AND COMPUTING↗

MEMS RF accelerators for nuclear energy and advanced manufacturing

Energetic ions are widely used to develop radiation hard structural materials and fuels for advanced nuclear energy. Ion beams are also used to generate neutrons for nuclear materials testing, as well as in a series of high impact, high value adding manufacturing processes (ranging from doping of semiconductors to hardening of materials against wear and to increase bio-compatibility of materials). But to date, high power beams of high energy ions are simply too expensive because they are delivered from large single beam accelerators. We have recently demonstrated that MEMS technology enables massively parallel, low cost batch fabrication of ion beams. We propose to scale intense ion accelerators based on MEMS (micro-electro mechanical systems) to high beam power (>10 kW). With our approach, hundreds to thousands of ion beamlets will be densely packed on silicon wafers. Ions are injected and accelerated across gaps formed in stacks of wafers, leading to uniquely high current densities for intense ion accelerators with variable kinetic energy (0.1 to 10 MeV) and variable ion species. Our team brings together experts in high power ion accelerators from Lawrence Berkeley National Laboratory and leading experts in MEMS technology from Cornell University. We will deliver a disruptive technology of low cost, flexible and massively scalable ion accelerators that will enable the rapid development of radiation hard nuclear materials for advanced nuclear energy and enable new applications in manufacturing.

43 PARTICLE ACCELERATORS↗

Nonadiabatic electron response in the Hasegawa-Wakatani equations

Tokamak edge turbulence is strongly influenced by parallel electron physics, which relaxes density and potential fluctuations towards electron adiabatic response. Beginning with the paradigmatic Hasegawa-Wakatani equations (HWEs) for resistive tokamak edge turbulence, a unique decomposition of the electric potential (φ) into adiabatic (a) and nonadiabatic (b) portions is derived, based on the requirement that a neither drive nor respond to the parallel current j{sub ∥}. The form of the decomposition clarifies that, at perpendicular scales large relative to the sound radius, the electron adiabatic response controls the nonzonal φ, not the fluctuating density n. Simple energy balance arguments allow one to rigorously bound the ratio of rms nonzonal nonadiabatic fluctuations (b(tilde sign)) relative to adiabatic ones (ã). The role of the vorticity nonlinearity in transferring energy between adiabatic and nonadiabatic fluctuations aids intuitive understanding of self-sustained turbulence in the HWEs. When the normalized parallel resistivity is weak, b(tilde sign) becomes effectively slaved, allowing the reduction to an approximate one-field model that remains valid for strong turbulence. In addition to guiding physical intuition, the one-field reduction should greatly ease further analytical manipulations. Direct numerical simulation of the 2D HWEs confirms the convergence of the asymptotic formula for b(tilde sign)

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

High energy density picoliter-scale zinc-air microbatteries for colloidal robotics

The recent interest in microscopic autonomous systems, including microrobots, colloidal state machines, and smart dust, has created a need for microscale energy storage and harvesting. However, macroscopic materials for energy storage have noted incompatibilities with microfabrication techniques, creating substantial challenges to realizing microscale energy systems. Here, we photolithographically patterned a microscale zinc/platinum/SU-8 system to generate the highest energy density microbattery at the picoliter (10 −12 liter) scale. The device scavenges ambient or solution-dissolved oxygen for a zinc oxidation reaction, achieving an energy density ranging from 760 to 1070 watt-hours per liter at scales below 100 micrometers lateral and 2 micrometers thickness in size. The parallel nature of photolithography processes allows 10,000 devices per wafer to be released into solution as colloids with energy stored on board. Within a volume of only 2 picoliters each, these primary microbatteries can deliver open circuit voltages of 1.05 ± 0.12 volts, with total energies ranging from 5.5 ± 0.3 to 7.7 ± 1.0 microjoules and a maximum power near 2.7 nanowatts. We demonstrated that such systems can reliably power a micrometer-sized memristor circuit, providing access to nonvolatile memory. We also cycled power to drive the reversible bending of microscale bimorph actuators at 0.05 hertz for mechanical functions of colloidal robots. Additional capabilities, such as powering two distinct nanosensor types and a clock circuit, were also demonstrated. The high energy density, low volume, and simple configuration promise the mass fabrication and adoption of such picoliter zinc-air batteries for micrometer-scale, colloidal robotics with autonomous functions.

Robotics↗

Meta Parallel Coordinates For Visualizing Features in Large, High-Dimensional, Time-Varying Data

Managing computational complexity and designing effective visual representations are two important challenges for the visualization of large, complex, high-dimensional datasets. Parallel coordinates are an effective technique for visualizing high- dimensional data, but do not scale well to very large datasets. The addition of the temporal dimension leads to more uncertainty due to clutter on screen. Consequently, this poses a significant challenge for visually finding trends and patterns that maximize insight about the underlying time-varying properties of the data. To address these problems, we present meta parallel coordinates, a parallel coordinates display that is guided by perceptually mo- tivated visual metrics. These metrics describe the visual structures typically found in parallel coordinates and thus aid the user’s analysis by providing meaningful views of the data. Since they are computed in screen space, our metrics are computationally more efficient than data-based metrics. Our choice of metrics is driven by the different analytical tasks that a user typically wants to perform with time-varying multivariate data. In particular, we have worked with domain scientists who performed simulations of bioremediation experiments, and use their data and results to demonstrate the usefulness of our approach

Dasgupta, Aritra↗

Eight-fold signal amplification of a superconducting nanowire single-photon detector using a multiple-avalanche architecture

Superconducting nanowire avalanche single-photon detectors (SNAPs) with n parallel nanowires are advantageous over single-nanowire detectors because their output signal amplitude scales linearly with n. However, the SNAP architecture has not been viably demonstrated for n > 4. To increase n for larger signal amplification, we designed a multi-stage, successive-avalanche architecture which used nanowires, connected via choke inductors in a binary-tree layout. We demonstrated an avalanche detector with n = 8 parallel nanowires and achieved eight-fold signal amplification, with a timing jitter of 54 ps.

47 OTHER INSTRUMENTATION↗

Towards Scalable Deep Learning via I/O Analysis and Optimization

Deep learning systems have been growing in prominence as a way to automatically characterize objects, trends, and anomalies. Given the importance of deep learning systems, researchers have been investigating techniques to optimize such systems. An area of particular interest has been using large supercomputing systems to quickly generate effective deep learning networks: a phase often referred to as “training” of the deep learning neural network. As we scale existing deep learning frameworks—such as Caffe—on these large supercomputing systems, we notice that the parallelism can help improve the computation tremendously, leaving data I/O as the major bottleneck limiting the overall system scalability. In this paper, we first present a detailed analysis of the performance bottlenecks of Caffe on large supercomputing systems. Our analysis shows that the I/O subsystem of Caffe—LMDB—relies on memory-mapped I/O to access its database, which can be highly inefficient on large-scale systems because of its interaction with the process scheduling system and the network-based parallel filesystem. Based on this analysis, we then present LMDBIO, our optimized I/O plugin for Caffe that takes into account the data access pattern of Caffe in order to vastly improve I/O performance. Our experimental results show that LMDBIO can improve the overall execution time of Caffe by nearly 20-fold in some cases.

Pumma, Sarunya↗

Data-driven Fault Tolerance for Work Stealing Computations

Checkpoint-restart approaches to fault tolerance typically roll back all the processes to the previous checkpoint in the event of a failure. Work stealing is a promising technique to dynamically tolerate variations in the execution environment, including faults, system noise, and energy constraints. In this paper, we present fault tolerance mechanisms for task parallel computations, a popular computation idiom, employing work stealing. The computation is organized as a collection of tasks with data in a global address space. The completion of data operations, rather than the actual messages, is tracked to derive an idempotent data store. This information is used to accurately identify the tasks to be re-executed, therefore to recompute only the lost data, in the presence of random work stealing. We consider three recovery schemes that present distinct trade-offs -- lazy recovery with potentially increased re-execution cost, immediate collective recovery with associated synchronization overheads, and noncollective recovery enabled by additional communication. We employ distributed work stealing to dynamically rebalance the tasks on the live processes and evaluate the three schemes using candidate application benchmarks. We demonstrate that the overheads (space and time) of the fault tolerance mechanism are low, the cost incurred due to failures are small, and the overheads decrease with per-process work at scale.

fault tolerance, task parallelism, load balancing,↗

Template Interfaces for Agile Parallel Data-Intensive Science

Tigres provides a programming library to compose and execute large-scale data-intensive scientific workflows from desktops to supercomputers. DOE User Facilities and large science collaborations are increasingly generating large enough data sets that it is no longer practical to download them to a desktop to operate on them. They are instead stored at centralized compute and storage resources such as high performance computing (HPC) centers. Analysis of this data requires an ability to run on these facilities, but with current technologies, scaling an analysis to an HPC center and to a large data set is difficult even for experts. Tigres is addressing the challenge of enabling collaborative analysis of DOE Science data through a new concept of reusable "templates" that enable scientists to easily compose, run and manage collaborative computational tasks. These templates define common computation patterns used in analyzing a data set.

Ramakrishnan, Lavanya↗

Scaling Up Coordinate Descent Algorithms for Large ℓ1 Regularization Problems

We present a generic framework for parallel coordinate descent (CD) algorithms that has as special cases the original sequential algorithms of Cyclic CD and Stochastic CD, as well as the recent parallel Shotgun algorithm of Bradley et al. We introduce two novel parallel algorithms that are also special cases---Thread-Greedy CD and Coloring-Based CD---and give performance measurements for an OpenMP implementation of these.

Scherrer, Chad↗