Search NASA⌕ Search

SEARCH · Search NASA

Results for “distributed parallelization”

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 955 records · Page 53

Evaluation of pre-training large language models on leadership-class supercomputers

Large language models (LLMs) have arisen rapidly to the center stage of artificial intelligence as the foundation models applicable to many downstream learning tasks. However, how to effectively build, train, and serve such models for many high-stake and first-principle-based scientific use cases are both of great interests and of great challenges. Moreover, pre-training LLMs with billions or even trillions of parameters can be prohibitively expensive not just for academic institutions, but also for well-funded industrial and government labs. Furthermore, the energy cost and the environmental impact of developing LLMs must be kept in mind. Here, in this work, we conduct a first-of-its-kind performance analysis to understand the time and energy cost of pre-training LLMs on the Department of Energy (DOE)’s leadership-class supercomputers. Employing state-of-the-art distributed training techniques, we evaluate the computational performance of various parallelization approaches at scale for a range of model sizes, and establish a projection model for the cost of full training. Our findings provide baseline results, best practices, and heuristics for pre-training such large models that should be valuable to HPC community at large. We also offer insights and optimization strategies for using the first exascale computing system, Frontier, to train models of the size of GPT-3 and beyond.

97 MATHEMATICS AND COMPUTING↗

Parallel derivative-free optimization for simulation-based design of behind-the-meter energy systems

In this work, the integrated design and dispatch of behind-the-meter or distributed resources (e.g. stationary battery storage and solar PV generation) is considered. A simulation-based framework is employed, generating high-fidelity results with closed-loop predictive control at a fine resolution, at the expense of high computational cost (several minutes to a few hours per design point). To address this challenge, parallel derivative-free design methods are considered. Four methods are compared, including state-of-the-art surrogate-based methods (Radial-Basis Functions and Gaussian processes) and sampling strategies, an evolutionary-based method, and a simple sequential grid refinement method. As a case study, two types of design problem with increasing complexity are considered, namely, the design of behind-the-meter resources (three design variables) and the inclusion of grid capacity (four design variables). The second yields a constrained design problem for which violations can only be determined after solving the computationally expensive simulation. For the three-dimensional case, all methods present a good performance, achieving a solution within 1% of the optimum after the first iteration, with the sequential grid refinement exhibiting the fastest convergence and achieving the best final objective value. This indicates that the parallel evaluation of multiple sampling points may be more important than the choice of method for small decision spaces. For the four-dimensional constrained case, the Genetic Algorithm presents the best tradeoff between performance and computational effort, while the rough objective function terrain generated by constraint violation penalties reduces the performance of surrogate-based methods. Contour plots with flat regions indicate flexibility in the optimal design and highlight the importance of characterizing the solution space.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Quantum Monte Carlo Endstation for Petascale Computing

The major achievements enabled by QMC Endstation grant include * Performance improvement on clusters of x86 multi-core systems, especially on Cray XT systems * New and improved methods for the wavefunction optimizations * New forms of trial wavefunctions * Implementation of the full application on NVIDIA GPUs using CUDA The scaling studies of QMCPACK on large-scale systems show excellent parallel efficiency up to 216K cores on Jaguarpf (Cray XT5). The GPU implementation shows speedups of 10-15x over the CPU implementation on older generation of x86. We have implemented hybrid OpenMP/MPI scheme in QMC to take advantage of multi-core shared memory processors of petascale systems. Our hybrid scheme has several advantages over the standard MPI-only scheme. * Memory optimized: large read-only data to store one-body orbitals and other shared properties to represent the trial wave function and many-body Hamiltonian can be shared among threads, which reduces the memory footprint of a large-scale problem. * Cache optimized: the data associated with an active Walker are in cache during the compute-intensive drift-diffusion process and the operations on an Walker are optimized for cache reuse. Thread-local objects are used to ensure the data affinity to a thread. * Load balanced: Walkers in an ensemble are evenly distributed among threads and MPI tasks. The two-level parallelism reduces the population imbalance among MPI tasks and reduces the number of point-to-point communications of large messages (serialized objects) for the Walker exchange. * Communication optimized: the communication overhead, especially for the collective operations necessary to determine ET and measure the properties of an ensemble, is significantly lowered by using less MPI tasks. The multiple forms of parallelism afforded by QMC algorithms make them ideal candidates for acceleration in the many-core paradigm. We presented the results of our effort to port the QMCPACK simulation code to the NVIDIA CUDA GPU platform. We restructured the CPU algorithms to express additional parallelism, minimize GPU-CPU communication, and efficiently utilize the GPU memory hierarchy. Using mixed precision on GT200 GPUs and MPI for intercommunication and load balancing, we observe typical full-application speedups of approximately 10x to 15x relative to quad-core Xeon CPUs alone, while reproducing the double-precision CPU results within statistical error. We developed an all-electron quantum Monte Carlo (QMC) method for solids that does not rely on pseudopotentials, and used it to construct a primary ultra-high-pressure calibration based on the equation of state of cubic boron nitride. We computed the static contribution to the free energy with the QMC method and obtained the phonon contribution from density functional theory, yielding a high-accuracy calibration up to 900 GPa usable directly in experiment. We computed the anharmonic Raman frequency shift with QMC simulations as a function of pressure and temperature, allowing optical pressure calibration. In contrast to present experimental approaches, small systematic errors in the theoretical EOS do not increase with pressure, and no extrapolation is needed. This all-electron method is applicable to first-row solids, providing a new reference for ab initio calculations of solids and benchmarks for pseudopotential accuracy. We compared experimental and theoretical results on the momentum distribution and the quasiparticle renormalization factor in sodium. From an x-ray Compton-profile measurement of the valence-electron momentum density, we derived its discontinuity at the Fermi wavevector finding an accurate measure of the renormalization factor that we compared with quantum-Monte-Carlo and G0W0 calculations performed both on crystalline sodium and on the homogeneous electron gas. Our calculated results are in good agreement with the experiment. We have been studying the heat of formation for various Kubas complexes of molecular hydrogen on Ti(1,2)ethylene-nH2 using Diffusion Monte Carlo. This work has been started and is ongoing. We are studying systems involving 1 and 2 Ti bonding sites with up to 10 hydrogen molecules in numerous configurations. This work will establish a benchmark that will test the accuracy of density functional calculations and establish the feasibility of our methods for similar systems.

97 MATHEMATICS AND COMPUTING↗

Mode Multiplexing for Scalable Cavity-Enhanced Operations in Neutral-Atom Arrays

Neutral-atom arrays provide a versatile platform for quantum information processing. However, in large-scale arrays, efficient photon collection remains a bottleneck for key tasks such as fast, nondestructive qubit readout and remote entanglement distribution. We propose a cavity-based approach that enables fast, parallel operations over many atoms using multiple modes of a single optical cavity. By selectively shifting the relevant atomic transitions, each atom can be coupled to a distinct cavity mode, allowing independent simultaneous processing. We present practical system designs that support cavity-mode multiplexing with up to 50 modes, enabling rapid mid-circuit syndrome extraction and significantly enhancing entanglement distribution rates between remote atom arrays. This approach offers a scalable solution to core challenges in neutral-atom arrays, advancing the development of practical quantum technologies.

Aqua, Ziv [Massachusetts Institute of Technology (↗

Simulation of Grazing-Incidence Synchrotron White Beam X-ray Topographic Images of Micropipes in 4H-SiC and Determination of Their Dislocation Senses

Detailed comparison has been drawn between synchrotron white beam X-ray topographic images of micropipes in 4H-SiC, recorded using pyramidal plane reflections in grazing-incidence geometry, and images simulated using the ray-tracing method. The simulations were carried out with and without the influence of surface relaxation effects. The images simulated in the absence of surface relaxation effects appear as white elliptical shaped features, canted to one side or other of the g-vector (depending on the dislocation sense), surrounded by a dark contrast perimeter which thickens at both ends of the major axis and which exhibits two fold symmetry axes parallel to the major and minor axes. On the other hand, on the images simulated taking into account the effects of surface relaxation, the features are again canted to one side or other of the g-vector (depending on the dislocation sense) but do not exhibit the same symmetry in that one of the sides of the oval shape is flattened (along the major axis) and the distribution of intensity around the perimeter no longer possesses the two fold symmetry axes parallel to the major and minor axes. While the details of the distribution of the dark perimeter contrast observed on the recorded images are not so easy to correlate with the simulated distributions, the sense of cant of the roughly elliptical white features on the observed images can be easily correlated with the simulations. Since the sense of cant has the same behavior as a function of dislocation sense for simulations carried out with and without surface relaxation, this provides a high level of confidence that the sense of cant of these features can be readily used to determine the senses of the micropipes.

43 PARTICLE ACCELERATORS↗

Parallel Runtime Interface for Fortran (PRIF): A Multi-Image Solution for LLVM Flang

Fortran compilers that provide support for Fortran’s native parallel features often do so with a runtime library that depends on details of both the compiler implementation and the communication library, while others provide limited or no support at all. This paper introduces a new generalized interface that is both compiler- and runtime-library-agnostic, providing flexibility while fully supporting all of Fortran’s parallel features. The Parallel Runtime Interface for Fortran (PRIF) was developed to be portable across shared- and distributed-memory systems, with varying operating systems, toolchains and architectures. It achieves this by defining a set of Fortran procedures corresponding to each of the parallel features defined in the Fortran standard that may be invoked by a Fortran compiler and implemented by a runtime library. PRIF aims to be used as the solution for LLVM Flang to provide parallel Fortran support. This paper also briefly describes our PRIF prototype implementation: Caffeine.

Bonachea, Dan↗

Parallel computing for power system climate resiliency: Solving a large-scale stochastic capacity expansion problem with mpi-sppy

Here we propose a nodal stochastic generation and transmission expansion planning model that incorporates the output from high-resolution global climate models through load and generation availability scenarios. We implement our model in Pyomo and perform computational studies on a realistically-sized test case of the California electric grid in a high performance computing environment. We propose model reformulations and algorithm tuning to efficiently solve this large problem using a variant of the Progressive Hedging Algorithm. We utilize the parallelization capabilities and overall versatility of mpi-sppy, exploiting its hub-and-spoke architecture to concurrently obtain inner and outer bounds on an optimal expansion plan. Initial results show that instances with 360 representative days on a system with over 8,000 buses can be solved to within 5% of optimality in under 4 h of wall clock time, a first step towards solving a large-scale power system expansion planning problem across a wide range of climate-informed operational scenarios.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Temporal Decompostion of a Distribution System Quasi-Static Time-Series Simulation

This paper documents the first phase of an investigation into reducing runtimes of complex OpenDSS models through parallelization. As the method seems promising, future work will quantify - and further mitigate - errors arising from this process. In this initial report, we demonstrate how, through the use of temporal decomposition, the run times of a complex distribution-system-level quasi-static time series simulation can be reduced roughly proportional to the level of parallelization. Using this method, the monolithic model runtime of 51 hours was reduced to a minimum of about 90 minutes. As expected, this comes at the expense of control- and voltage-errors at the time-slice boundaries. All evaluations were performed using a real distribution circuit model with the addition of 50 PV systems - representing a mock complex PV impact study. We are able to reduce induced transition errors through the addition of controls initialization, though small errors persist. The time savings with parallelization are so significant that we feel additional investigation to reduce control errors is warranted.

14 SOLAR ENERGY↗

Transport of transient solar wind particles in Earth's cusps

An important problem in space physics still not understood well is how the solar wind enters the Earth's magnetosphere. Evidence is presented that transient solar wind particles produced by solar disturbances can appear in the Earth's mid-altitude ({approx}5 R{sub E} geocentric) cusps with densities nearly equal to those in the magnetosheath. That these are magnetosheath particles is established by showing they have the same ''flattop'' electron distributions as magnetosheath electrons behind the bow shock. The transient ions are moving parallel to the magnetic field (B) toward Earth and often coexist with ionospheric particles that are flowing out. The accompanying waves include electromagnetic and broadband electrostatic noise emissions and Bernstein mode waves. Phase-space distributions show a mixture of hot and cold electrons and multiple ion species including field-aligned ionospheric O{sup +} beams.

58 GEOSCIENCES↗

Dust kinetic Alfven and acoustic waves in a Lorentzian plasma

Dust kinetic Alfven waves (DKAWs) with finite Larmor radius effects have been examined rigorously in a uniform dusty plasma in the presence of an external magnetic field. A dispersion relation of low-frequency DKAW on the dust acoustic velocity branch is obtained in a low-{beta} Lorentzian plasma. It is found that the influence of the Lorentzian distribution function is more effective for perpendicular component of group velocity as compared with parallel one. Lorentzian-type charging currents are obtained with the aid of Vlasov theory. Damping/instability due to dust charge fluctuation is found to be insensitive with the form of distribution function for DKAW. The possible applications to dusty space plasmas are pointed out.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Interfacial water on crystalline silica: a comparative molecular dynamics simulation study

All-atom molecular dynamics simulations were conducted to study the dynamics of aqueous electrolyte solutions confined in slit-shaped silica nanopores of various degrees of protonation. Five degrees of protonation were prepared by randomly removing surface hydrogen atoms from fully protonated crystalline silica surfaces. Aqueous electrolyte solutions containing NaCl or CsCl salt were simulated at ambient conditions. In all cases, the ionic concentration was 1 M. The results were quantified in terms of atomic density distributions within the pores, and the self-diffusion coefficient along the direction parallel to the pore surface. We found evidence for ion-specific properties that depend on ion surface, water ion, and only in some cases ion ion correlations. The degree of protonation strongly affects the structure, distribution, and the dynamic behavior of confined water and electrolytes. Cl ions adsorb on the surface at large degrees of protonation, and their behavior does not depend significantly on the cation type (either Na+ or Cs+ ions are present in the systems considered). The cations show significant ion-specific behavior. Na+ ions occupy different positions within the pore as the degree of protonation changes, while Cs+ ions mainly remain near the pore center at all conditions considered. For a given degree of protonation, the planar self-diffusion coefficient of Cs+ is always greater than that of Na+ ions. The results are useful for better understanding transport under confinement, including brine behavior in the subsurface, with important applications such as environmental remediation.

74 ATOMIC AND MOLECULAR PHYSICS↗

NEW CONSTRAINTS ON THE GALACTIC HALO MAGNETIC FIELD USING ROTATION MEASURES OF EXTRAGALACTIC SOURCES TOWARD THE OUTER GALAXY

We present a study of the Milky Way disk and halo magnetic field, determined from observations of Faraday rotation measure (RM) toward 641 polarized extragalactic radio sources in the Galactic longitude range 100 Degree-Sign -117 Degree-Sign , within 30 Degree-Sign of the Galactic plane. For |b| < 15 Degree-Sign , we observe a symmetric RM distribution about the Galactic plane. This is consistent with a disk field in the Perseus arm of even parity across the Galactic mid-plane. In the range 15 Degree-Sign < |b| < 30 Degree-Sign , we find median RMs of -15 {+-} 4 rad m{sup -2} and -62 {+-} 5 rad m{sup -2} in the northern and southern Galactic hemispheres, respectively. If the RM distribution is a signature of the large-scale field parallel to the Galactic plane, then this suggests that the halo magnetic field toward the outer Galaxy does not reverse direction across the mid-plane. The variation of RM as a function of Galactic latitude in this longitude range is such that RMs become more negative at larger |b|. This is consistent with an azimuthal magnetic field of strength 2 {mu}G (7 {mu}G) at a height 0.8-2 kpc above (below) the Galactic plane between the local and the Perseus spiral arm. We propose that the Milky Way could possess spiral-like halo magnetic fields similar to those observed in M51.

79 ASTRONOMY AND ASTROPHYSICS↗

Center for Programming Models for Scalable Parallel Computing

Rice University's achievements as part of the Center for Programming Models for Scalable Parallel Computing include: (1) design and implemention of cafc, the first multi-platform CAF compiler for distributed and shared-memory machines, (2) performance studies of the efficiency of programs written using the CAF and UPC programming models, (3) a novel technique to analyze explicitly-parallel SPMD programs that facilitates optimization, (4) design, implementation, and evaluation of new language features for CAF, including communication topologies, multi-version variables, and distributed multithreading to simplify development of high-performance codes in CAF, and (5) a synchronization strength reduction transformation for automatically replacing barrier-based synchronization with more efficient point-to-point synchronization. The prototype Co-array Fortran compiler cafc developed in this project is available as open source software from http://www.hipersoft.rice.edu/caf.

97 MATHEMATICS AND COMPUTING↗

ON THE SPATIAL DISTRIBUTION AND THE ORIGIN OF HYPERVELOCITY STARS

Hypervelocity stars (HVSs) escaping away from the Galactic halo are dynamical products of interactions of stars with the massive black hole(s) (MBH) in the Galactic Center (GC). They are mainly B-type stars with their progenitors unknown. OB stars are also populated in the GC, with many being hosted in a clockwise-rotating young stellar (CWS) disk within half a parsec from the MBH and their formation remaining puzzles. In this paper, we demonstrate that HVSs can well memorize the injecting directions of their progenitors using both analytical arguments and numerical simulations, i.e., the ejecting direction of an HVS is almost anti-parallel to the injecting direction of its progenitor. Therefore, the spatial distribution of HVSs maps the spatial distribution of the parent population of their progenitors directly. We also find that almost all the discovered HVSs are spatially consistent with being located on two thin disk planes. The orientation of one plane is consistent with that of the (inner) CWS disk, which suggests that most of the HVSs originate from the CWS disk or a previously existed disk-like stellar structure with an orientation similar to it. The rest of HVSs may be correlated with the plane of the northern arm of the mini-spiral in the GC or the plane defined by the outer warped part of the CWS disk. Our results not only support the GC origin of HVSs but also imply that the central disk (or the disk structure with a similar orientation) should persist or be frequently rejuvenated over the past 200 Myr, which adds a new challenge to the stellar disk formation and provides insights to the longstanding problem of gas fueling into MBHs.

79 ASTRONOMY AND ASTROPHYSICS↗

An assessment of full wave effects on the propagation and absorption of lower hybrid waves

Lower hybrid (LH) waves ({omega}{sub ci}<<{omega}<<{omega}{sub ce}, where {omega}{sub i,e}{identical_to}Z{sub i,e}eB/m{sub i,e}c) have the attractive property of damping strongly via electron Landau resonance on relatively fast tail electrons and consequently are well-suited to driving current. Established modeling techniques use Wentzel-Kramers-Brillouin (WKB) expansions with self-consistent non-Maxwellian distributions. Higher order WKB expansions have shown some effects on the parallel wave number evolution and consequently on the damping due to diffraction [G. Pereverzev, Nucl. Fusion 32, 1091 (1991)]. A massively parallel version of the TORIC full wave electromagnetic field solver valid in the LH range of frequencies has been developed [J. C. Wright et al., Comm. Comp. Phys. 4, 545 (2008)] and coupled to an electron Fokker-Planck solver CQL3D[R. W. Harvey and M. G. McCoy, in Proceedings of the IAEA Technical Committee Meeting, Montreal, 1992 (IAEA Institute of Physics Publishing, Vienna, 1993), USDOC/NTIS Document No. DE93002962, pp. 489-526] in order to self-consistently evolve nonthermal electron distributions characteristic of LH current drive (LHCD) experiments in devices such as Alcator C-Mod and ITER (B{sub 0}{approx_equal}5 T, n{sub e0}{approx_equal}1x10{sup 20} m{sup -3}). These simulations represent the first ever self-consistent simulations of LHCD utilizing both a full wave and Fokker-Planck calculation in toroidal geometry.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Dispersive dielectric and conductive effects in 2D resistor--capacitor networks

How to predict and better understand the effective properties of disordered material mixtures has been a long-standing problem in different research fields, especially in condensed matter physics. In order to address this subject and achieve a better understanding of the frequency-dependent properties of these systems, a large 2D L x L square structure of resistors and capacitors was used to calculate the immittance response of a network formed by random filling of binary conductor/insulator phases with 1000 O resistors and 10 nF capacitors. The effects of percolating clusters on the immittance response were studied statistically through the generation of 10 000 different random network samples at the percolation threshold. The scattering of the imaginary part of the immittance near the dc limit shows a clear separation between the responses of percolating and non-percolating samples, with the gap between their distributions dependent on both network size and applied frequency. These results could be used to monitor connectivity in composite materials. The effects of the content and structure of the percolating path on the nature of the observed dispersion were investigated, with special attention paid to the geometrical fractal concept of the backbone and its influence on the behavior of relaxation-time distributions. For three different resistor-capacitor proportions, the appropriateness of many fitting models was investigated for modeling and analyzing individual resistor-capacitor network dispersed frequency responses using complex-nonlinear-least-squares fitting. Several remarkable new features were identified, including a useful duality relationship and the need for composite fitting models rather than either a simple power law or a single Davidson-Cole one. Good fits of data for fully percolating random networks required two dispersive fitting models in parallel or series, with a cutoff at short times of the distribution of relaxation times of one of them. In addition, such fits surprisingly led to cutoff parameters, including a primitive relaxation or crossover time, with estimated values comparable to those found for real dispersive materials.

36 MATERIALS SCIENCE↗

Model-based Dynamic Control of Speculative Forays in Parallel Computation

In simulations running in parallel, the processors would have to synchronize with other processors to maintain correct global order of computations. This can be done either by blocking computation until correct order is guaranteed, or by \textit{speculatively} proceeding with the best guess (based on local information) and later correcting errors if/as necessary. Since the gainful lengths of speculative forays depend on the dynamics of the application software and hardware at runtime, an online control system is necessary to dynamically choose and/or switch between the blocking and speculative strategies. In this paper, we formulate the reversible speculative computing in large-scale parallel computing as a dynamic linear feedback control (optimization) system model and evaluate its performance in terms of time and cost savings as compared to the traditional (forward) computing. We illustrate with an exact analogy in the form of vehicular travel under dynamic, delayed route information. The objective is to assist in making the optimal decision on what computational approach is to be chosen, by predicting the amount of time and cost savings (or losing) under different environments represented by different parameters and probability distribution functions. We consider the cases of Gaussian, exponential and log-normal distribution functions. The control system is intended for incorporating into speculative parallel applications such as optimistic parallel discrete event simulations to decide at runtime when and to what extent speculative execution can be performed gainfully.

Perumalla, Kalyan↗