Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Parallel Supercomputers”

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 577 records · Page 32

A Blocked Linear Method for Optimizing Large Parameter Sets in Variational Monte Carlo

We present a modification to variational Monte Carlo’s linear method optimization scheme that addresses a critical memory bottleneck while maintaining compatibility with both the traditional ground state variational principle and our recently-introduced variational principle for excited states. For wave function ansatzes with tens of thousands of variables, our modification reduces the required memory per parallel process from tens of gigabytes to hundreds of megabytes, making the methodology a much better fit for modern supercomputer architectures in which data communication and per-process memory consumption are primary concerns. We verify the efficacy of the new optimization scheme in small molecule tests involving both the Hilbert space Jastrow antisymmetric geminal power ansatz and real space multi-Slater Jastrow expansions. Satisfied with its performance, we have added the optimizer to the QMCPACK software package, with which we demonstrate on a hydrogen ring a prototype approach for making systematically convergent, non-perturbative predictions of Mott-insulators’ optical band gaps.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

A Sparse Distributed Gigascale Resolution Material Point Method

In this paper, we present a four-layer distributed simulation system and its adaptation to the Material Point Method (MPM). The system is built upon a performance portable C++ programming model targeting major High-Performance-Computing (HPC) platforms. A key ingredient of our system is a hierarchical block-tile-cell sparse grid data structure that is distributable to an arbitrary number of Message Passing Interface (MPI) ranks. We additionally propose strategies for efficient dynamic load balance optimization to maximize the efficiency of MPI tasks. Our simulation pipeline can easily switch among backend programming models, including OpenMP and CUDA, and can be effortlessly dispatched onto supercomputers and the cloud. Finally, we construct benchmark experiments and ablation studies on supercomputers and consumer workstations in a local network to evaluate the scalability and load balancing criteria. We demonstrate massively parallel, highly scalable, and gigascale resolution MPM simulations of up to 1.01 billion particles for less than 323.25 seconds per frame with 8 OpenSSH-connected workstations.

97 MATHEMATICS AND COMPUTING↗

Extreme-Scale Bayesian Inference for Uncertainty Quantification of Complex Simulations

Uncertainty quantification (UQ)—that is, quantifying uncertainties in complex mathematical models and their large-scale computational implementations—is widely viewed as one of the outstanding challenges facing the field of CS&E over the coming decade. The EUREKA project set to address the most difficult class of UQ problems: those for which both the underlying PDE model as well as the uncertain parameters are of extreme scale. In the project we worked on these extreme-scale challenges in the following four areas: 1. Scalable parallel algorithms for sampling and characterizing the posterior distribution that exploit the structure of the underlying PDEs and parameter-to-observable map. These include structure-exploiting versions of the randomized maximum likelihood method, which aims to overcome the intractability of employing conventional MCMC methods for solving extreme-scale Bayesian inversion problems by appealing to and adapting ideas from large-scale PDE-constrained optimization, which have been very successful at exploring high-dimensional spaces. 2. Scalable parallel algorithms for construction of prior and likelihood functions based on learning methods and non-parametric density estimation. Constructing problem-specific priors remains a critical challenge in Bayesian inference, and more so in high dimensions. Another challenge is construction of likelihood functions that capture unmodeled couplings between observations and parameters. We will create parallel algorithms for non-parametric density estimation using high dimensional N-body methods and combine them with supervised learning techniques for the construction of priors and likelihood functions. 3. Bayesian inadequacy models, which augment physics models with stochastic models that represent their imperfections. The success of the Bayesian inference framework depends on the ability to represent the uncertainty due to imperfections of the mathematical model of the phenomena of interest. This is a central challenge in UQ, especially for large-scale models. We propose to develop the mathematical tools to address these challenges in the context of extreme-scale problems. 4. Parallel scalable algorithms for Bayesian optimal experimental design (OED). Bayesian inversion yields quantified uncertainties in the model parameters, which can be propagated forward through the model to yield uncertainty in outputs of interest. This opens the way for designing new experiments to reduce the uncertainties in the model parameters and model predictions. Such experimental design problems have been intractable for large-scale problems using conventional methods; we will create OED algorithms that exploit the structure of the PDE model and the parameter-to-output map to overcome these challenges. Parallel algorithms for these four problems were created, analyzed, prototyped, implemented, tuned, and scaled up for leading-edge supercomputers, including UT-Austin’s own 10 petaflops Stampede system, ANL’s Mira system, and ORNL’s Titan system. While our focus is on fundamental mathematical/computational methods and algorithms, we will assess our methods on model problems derived from several DOE mission applications, including multiscale mechanics and ice sheet dynamics.

97 MATHEMATICS AND COMPUTING↗

A parallel-vector Lanczos eigen-solver for structural vibration problems

The Lanczos algorithm for the solution of generalized eigen-problem has been receiving a lot of attention in recent years due to its computational efficiency. The focus of this paper is to develop a Lanczos algorithm which can exploit both the parallel and vector capabilities provided by modern high-performance computers. A partial restoring orthogonality scheme is also developed and incorporated into the basic Lanczos algorithm. The numerical performance in terms of accuracy and efficiency of the proposed parallel-vector Lanczos algorithm is demonstrated by solving for the frequencies and mode shapes of structural problems on multiprocessor supercomputers.

Qin, Jiangning↗

Code Optimization and Parallelization on the Origins: Looking from Users' Perspective

Parallel machines are becoming the main compute engines for high performance computing. Despite their increasing popularity, it is still a challenge for most users to learn the basic techniques to optimize/parallelize their codes on such platforms. In this paper, we present some experiences on learning these techniques for the Origin systems at the NASA Advanced Supercomputing Division. Emphasis of this paper will be on a few essential issues (with examples) that general users should master when they work with the Origins as well as other parallel systems.

Chang, Yan-Tyng Sherry↗

End-to-end experiment management in HPC

Experiment management in any domain is challenging. There is a perpetual feedback loop cycling through planning, execution, measurement, and analysis. The lifetime of a particular experiment can be limited to a single cycle although many require myriad more cycles before definite results can be obtained. Within each cycle, a large number of subexperiments may be executed in order to measure the effects of one or more independent variables. Experiment management in high performance computing (HPC) follows this general pattern but also has three unique characteristics. One, computational science applications running on large supercomputers must deal with frequent platform failures which can interrupt, perturb, or terminate running experiments. Two, these applications typically integrate in parallel using MPI as their communication medium. Three, there is typically a scheduling system (e.g. Condor, Moab, SGE, etc.) acting as a gate-keeper for the HPC resources. In this paper, we introduce LANL Experiment Management (LEM), an experimental management framework simplifying all four phases of experiment management. LEM simplifies experiment planning by allowing the user to describe their experimental goals without having to fully construct the individual parameters for each task. To simplify execution, LEM dispatches the subexperiments itself thereby freeing the user from remembering the often arcane methods for interacting with the various scheduling systems. LEM provides transducers for experiments that automatically measure and record important information about each subexperiment; these transducers can easily be extended to collect additional measurements specific to each experiment. Finally, experiment analysis is simplified by providing a general database visualization framework that allows users to quickly and easily interact with their measured data.

97 MATHEMATICS AND COMPUTING↗

Multi-node and Multi-core Performance Studies of a Monte Carlo code RMC

Over the past 15 years the architecture of high-performance computing systems has become increasingly heterogeneous with individual computing cores being subdivided into nodes and even auxiliary co-processing units such as graphics processing units or the Intel many integrated core co-processors. In fact, future computer platforms is moving toward a larger number of nodes and cores with lower memory available per node. Today, several production Monte Carlo codes use some types of hybrid parallelism strategy that makes use of shared-memory parallelism within a computing node and message-passing parallelism between nodes, which enables the codes to utilize available computing resources as much as possible, both among and within computing nodes on a cluster. The so-called hybrid parallelization is also possible with Reactor Monte Carlo Code RMC by activating both MPI and OpenMP parallel algorithms together. On HPC' which typically consists of several or several-ten thousand nodes equipping 16-32 computing intra-cores per node, simulation of RMC can be carried out in the hybrid parallelization with the MPI inter-node parallelization and the OpenMP intra-node parallelization. RMC has adopted a hybrid message-passing and shared-memory approach to parallelism, which enables the code to utilize available computing resources, both among and within computing nodes on a cluster. Shared-memory parallelism is especially valuable for large-model simulations because it enables multiple compute cores within a node to share common model information such as the geometry description, material, cross section and tally data, thus reducing the memory usage per processor. The calculation on the Tianhe-2 supercomputer took 6144 cores, resulting in a rate of 173,314,000 neutrons/minute for BEAVRS benchmark. While pin-wise full core H-M benchmark burnup calculations with millions of depletion regions was selected using MPI/OpenMP parallelism, the speed of 6,144 cores reach 1,843,200 burnable regions per minute. Keeping up with the continuing trend for increases in the number of computing resources within modern HPC systems is a significant challenge for teams that develop and maintain high-performance MC radiation transport solvers. order to address this challenge, RMC has adopted a hybrid message-passing and shared-memory approach to parallelism, which enables the code to utilize available computing resources, both among and within computing nodes on a cluster. Shared-memory parallelism is especially valuable for large-model simulations because it enables multiple compute cores within a node to share common model information such as the geometry description, material and composition definitions, and cross section data, thus reducing the memory usage per processor. The parallel performance has been studied by using two complex and representative benchmarks, H-M and BEAVRS benchmark. All simulations were run on Intel Xeon E5-2692 12 core, 2.2 GHz processors at MilkyWay-2 Super Computer with up to 12,288 available computing cores. While employing MPI parallelism, excellent parallel efficiency is achieved even above 12,000 processors and RMC reaches the parallel efficiency of 96% on 12,288 cores for the pin-by-pin computation of the BEAVRS benchmark. While BEAVRS criticality benchmark was selected using MPI/OpenMP parallelism, the speed of 6,144 cores reach 173,314,000 particles per minute. While pin-wise full core H-M benchmark burnup calculations with millions of depletion regions was selected using MPI/OpenMP parallelism, the speed of 6,144 cores reach 1,843,200 burnable regions per minute. Results from these studies demonstrate that RMC, using the hybrid parallel model described in this paper, is able to handle pin-wise full core burnup calculations with millions of depletion regions and scales well through thousands of processors. (authors)

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Advances in Supercomputing for the Modeling of Atomic Processes in Plasmas

An overview will be given of recent atomic and molecular collision methods developed to take advantage of modern massively parallel computers. The focus will be on direct solutions of the time-dependent Schroedinger equation for simple systems using large numerical lattices, as found in the time-dependent close-coupling method, and for configuration interaction solutions of the time-independent Schroedinger equation for more complex systems using large numbers of basis functions, as found in the R-matrix with pseudo-states method. Results from these large scale calculations are extremely useful in benchmarking less accurate theoretical methods and experimental data. To take full advantage of future petascale and exascale computing resources, it appears that even finer grain parallelism will be needed.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DeepHyper: A Python Package for Massively Parallel Hyperparameter Optimization in Machine Learning

Machine learning models are increasingly applied across scientific disciplines, yet their effectiveness often hinges on heuristic decisions—such as data transformations, training strategies, and model architectures—that are not learned by the models themselves. Automating the selection of these heuristics and analyzing their sensitivity is crucial for building robust and efficient learning workflows. DeepHyper addresses this challenge by democratizing hyperparameter optimization, providing accessible tools to streamline and enhance machine learning workflows from a laptop to the largest supercomputer in the world. Building on top of hyperparameter optimization, it unlocks new capabilities around ensembles of models for improved accuracy and uncertainty quantification. All of these organized around efficient parallel computing.

ensemble↗

OpenSn: A massively parallel, open-source simulation environment for discrete ordinates radiation transport

OpenSn is an open-source, massively parallel deterministic radiation transport code for solving the discrete-ordinates ( S N ) form of the Boltzmann transport equation on unstructured, arbitrary polyhedral meshes. It supports high-fidelity simulations involving steady-state, eigenvalue, and adjoint problems for neutral particles (e.g., neutrons, photons, multi-particles), using the multigroup approximation in energy. OpenSn combines angular discretization via discrete ordinates with a discontinuous Galerkin finite element method (DGFEM) in space, enabling accurate resolution of transport physics on arbitrary polyhedral cells, included locally refined spatial grids. It includes multiple angular quadrature types, including locally refined angular quadratures. Written in modern C++ with a Python API, OpenSn runs efficiently on platforms ranging from laptops to supercomputers. The transport sweep algorithm is implemented using a task-based, directed-acyclic-graph (DAG) approach for each angle and supports asynchronous parallelism across thousands of MPI ranks. Group-set aggregation improves compute intensity, and synthetic acceleration techniques (e.g., diffusion synthetic acceleration, second-moment method) enhance solver convergence. OpenSn has been verified on reactor physics problems and demonstrated excellent weak and strong scaling performance on more than 32,768 processes, making it a versatile and robust platform for large-scale transport simulations in complex geometries.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Methodologies and Tools for Tuning Parallel Programs: 80% Art, 20% Science, and 10% Luck

The need for computing power has forced a migration from serial computation on a single processor to parallel processing on multiprocessors. However, without effective means to monitor (and analyze) program execution, tuning the performance of parallel programs becomes exponentially difficult as program complexity and machine size increase. In the past few years, the ubiquitous introduction of performance tuning tools from various supercomputer vendors (Intel's ParAide, TMC's PRISM, CRI's Apprentice, and Convex's CXtrace) seems to indicate the maturity of performance instrumentation/monitor/tuning technologies and vendors'/customers' recognition of their importance. However, a few important questions remain: What kind of performance bottlenecks can these tools detect (or correct)? How time consuming is the performance tuning process? What are some important technical issues that remain to be tackled in this area? This workshop reviews the fundamental concepts involved in analyzing and improving the performance of parallel and heterogeneous message-passing programs. Several alternative strategies will be contrasted, and for each we will describe how currently available tuning tools (e.g. AIMS, ParAide, PRISM, Apprentice, CXtrace, ATExpert, Pablo, IPS-2) can be used to facilitate the process. We will characterize the effectiveness of the tools and methodologies based on actual user experiences at NASA Ames Research Center. Finally, we will discuss their limitations and outline recent approaches taken by vendors and the research community to address them.

Yan, Jerry C.↗

A Fully Coupled Model for Electromechanics of the Heart

We present a fully coupled electromechanical model of the heart. The model integrates cardiac electrophysiology and cardiac mechanics through excitation-induced contraction and deformation-induced current. Numerical schemes based on finite element were implemented in a supercomputer. Numerical examples were presented using a thin cardiac tissue and a dog ventricle with realistic geometry. Performance of the parallel simulation scheme was studied. The model provides a useful tool to understand cardiovascular dynamics.

42 ENGINEERING↗

ArborX

ArborX library tackles a problem of efficiently finding geometric objects that are close in space. Variations of this problem, such as finding the nearest neighbors of a point, or finding all objects within a certain distance, are inherent components of applications in many fields. The data may be large so that solving the problem efficiently may require significant computational resources, such as multiple processors or accelerators such as general purpose GPUs. ArborX' main advantage in its ability to solve large problems efficiently utilizing a combination of distributed and on-node parallelism. ArborX can be run efficiently on a wide variety of hardware, including GPUs from different vendors, which distinguishes it from other available libraries which typically choose only few of these. The other advantage is that it supports both types of user problems: spatial problems (useful for intersections and finding objects within certain distance), and nearest neighbor problems. ArborX also supports flexible interface in its interaction with a user. Particularly, it allows a user to call user's own function on a positive match, a functionality not rarely available in other libraries. ArborX implements construction and traversal algorithms using efficient tree structures, such as bounding volume hierarchy (BVH). At its core, it uses linear BVH for its low construction cost and sufficient quality. ArborX is written using C++, and is parallelized using the message passing interface (MPI) for the distributed communication, and the Kokkos library for on-node parallelism. This approach allows ArborX to be run on a wide variety of hardware, from common laptops and desktops to supercomputers while using the same codebase. ArborX also implements several advanced algorithms using geometric search, such as density-based clustering algorithm DBSCAN.

ECP↗

ArborX 2.0

ArborX library tackles a problem of efficiently finding geometric objects that are close in space. Variations of this problem, such as finding the nearest neighbors of a point, or finding all objects within a certain distance, are inherent components of applications in many fields. The data may be large so that solving the problem efficiently may require significant computational resources, such as multiple processors or accelerators such as general purpose GPUs. ArborX' main advantage in its ability to solve large problems efficiently utilizing a combination of distributed and on-node parallelism. ArborX can be run efficiently on a wide variety of hardware, including GPUs from different vendors, which distinguishes it from other available libraries which typically choose only few of these. The other advantage is that it supports both types of user problems: spatial problems (useful for intersections and finding objects within certain distance), and nearest neighbor problems. ArborX also supports flexible interface in its interaction with a user. Particularly, it allows a user to call user's own function on a positive match, a functionality not rarely available in other libraries. ArborX implements construction and traversal algorithms using efficient tree structures, such as bounding volume hierarchy (BVH). At its core, ArborX uses linear BVH for its low construction cost and sufficient quality. ArborX implements both spatial and nearest-neighbor traversal algorithms. ArborX also provides several clustering algorithms (minimum spanning tree, DBSCAN, HDBSCAN*), interpolation using minimum least squares and ray tracing. ArborX is written using C++, and is parallelized using the message passing interface (MPI) for the distributed communication, and the Kokkos library for on-node parallelism. This approach allows ArborX to be run on a wide variety of hardware, from common laptops and desktops to supercomputers while using the same codebase.

Prokopenko, Andrey [Oak Ridge National Laboratory ↗

HydraGNN v4.0

The new version of HydraGNN v4.0 provides additional core capabilities, such as: Inclusion of multi-body atomistic cluster expansion MACE, polarizable atom interaction neural network PAINN, and equivariant principal neighborhood aggregation (PNAEq) among the message passing layers supported -Inclusion of graph transformers to directly model long-range interactions between nodes that are distant in the graph topology Integration of graph transformers with message passing layers by combining the graph embedding generated by the two mechanisms, which allows for an improved expressivity of the HydraGNN architecture Improved re-implementation of multi-task learning (MTL) to allow its use for stabilized training across imbalanced, multi-source, multi-fidelity data Introduction of multi-task parallelism, a newly proposed type of model parallelism specifically for MTL architectures, which allows to dispatch different output decoding heads to different GPU devices Integration of multi-task parallelism with pre-existing distributed data parallelism to enable a 2D parallelization for distributed training Improved portability of the distributed training across Intel GPUs, which has been testes on ALCF exascale supercomputer Aurora Inclusion of 2-level fine-grained energy profilers portable across NVIDIA, AMD, and Intel GPUs to monitor the power and energy consumption associated with different functions executed by the HydraGNN code during data pre-load and training Restructuring of previous examples and inclusion of new sets of examples to illustrate the download, preprocess, and training of HydraGNN models on new large-scale open-source datasets for atomistic materials modeling (e.g., Alexandria, Transition1x, OMat24, OMol25)

Lupo Pasini, Massimiliano [Oak Ridge National Labo↗

PCMS: Parallel Coupler For Multimodel Simulations

This paper presents the Parallel Coupler for Multimodel Simulations (PCMS), a new GPU accelerated generalized coupling framework for coupling simulation codes on leadership class supercomputers. PCMS includes distributed control and field mapping methods for up to five dimensions. For field mapping PCMS can utilize discretization and field information to accommodate physics constraints. PCMS is demonstrated with a coupling of the gyrokinetic microturbulence code XGC with a Monte Carlo neutral transport code DEGAS2 and with a 5D distribution function coupling of an energetic particle transport code (GNET) to a gyrokinetic microturbulence code (GTC). Weak scaling is also demonstrated on up to 2,080 GPUs of Frontier with a weak scaling efficiency of 85%.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Parallel Navier-Stokes computations on shared and distributed memory architectures

We study a high order finite difference scheme to solve the time accurate flow field of a jet using the compressible Navier-Stokes equations. As part of our ongoing efforts, we have implemented our numerical model on three parallel computing platforms to study the computational, communication, and scalability characteristics. The platforms chosen for this study are a cluster of workstations connected through fast networks (the LACE experimental testbed at NASA Lewis), a shared memory multiprocessor (the Cray YMP), and a distributed memory multiprocessor (the IBM SPI). Our focus in this study is on the LACE testbed. We present some results for the Cray YMP and the IBM SP1 mainly for comparison purposes. On the LACE testbed, we study: (1) the communication characteristics of Ethernet, FDDI, and the ALLNODE networks and (2) the overheads induced by the PVM message passing library used for parallelizing the application. We demonstrate that clustering of workstations is effective and has the potential to be computationally competitive with supercomputers at a fraction of the cost.

Hayder, M. Ehtesham↗

Efficacy of Code Optimization on Cache-based Processors

The current common wisdom in the U.S. is that the powerful, cost-effective supercomputers of tomorrow will be based on commodity (RISC) micro-processors with cache memories. Already, most distributed systems in the world use such hardware as building blocks. This shift away from vector supercomputers and towards cache-based systems has brought about a change in programming paradigm, even when ignoring issues of parallelism. Vector machines require inner-loop independence and regular, non-pathological memory strides (usually this means: non-power-of-two strides) to allow efficient vectorization of array operations. Cache-based systems require spatial and temporal locality of data, so that data once read from main memory and stored in high-speed cache memory is used optimally before being written back to main memory. This means that the most cache-friendly array operations are those that feature zero or unit stride, so that each unit of data read from main memory (a cache line) contains information for the next iteration in the loop. Moreover, loops ought to be 'fat', meaning that as many operations as possible are performed on cache data-provided instruction caches do not overflow and enough registers are available. If unit stride is not possible, for example because of some data dependency, then care must be taken to avoid pathological strides, just ads on vector computers. For cache-based systems the issues are more complex, due to the effects of associativity and of non-unit block (cache line) size. But there is more to the story. Most modern micro-processors are superscalar, which means that they can issue several (arithmetic) instructions per clock cycle, provided that there are enough independent instructions in the loop body. This is another argument for providing fat loop bodies. With these restrictions, it appears fairly straightforward to produce code that will run efficiently on any cache-based system. It can be argued that although some of the important computational algorithms employed at NASA Ames require different programming styles on vector machines and cache-based machines, respectively, neither architecture class appeared to be favored by particular algorithms in principle. Practice tells us that the situation is more complicated. This report presents observations and some analysis of performance tuning for cache-based systems. We point out several counterintuitive results that serve as a cautionary reminder that memory accesses are not the only factors that determine performance, and that within the class of cache-based systems, significant differences exist.

VanderWijngaart, Rob F.↗