Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “optimization of nuclear systems”

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 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Performant automatic differentiation of local coupled cluster theories: Response properties and ab initio molecular dynamics

In this work, we introduce a differentiable implementation of the local natural orbital coupled cluster (LNO-CC) method within the automatic differentiation framework of the PySCFAD package. The implementation is comprehensively tuned for enhanced performance, which enables the calculation of first-order static response properties on medium-sized molecular systems using coupled cluster theory with single, double, and perturbative triple excitations [CCSD(T)]. We evaluate the accuracy of our method by benchmarking it against the canonical CCSD(T) reference for nuclear gradients, dipole moments, and geometry optimizations. In addition, we demonstrate the possibility of property calculations for chemically interesting systems through the computation of bond orders and Mössbauer spectroscopy parameters for a [NiFe]-hydrogenase active site model, along with the simulation of infrared spectra via ab initio LNO-CC molecular dynamics for a protonated water hexamer.

Chemistry↗
DOE OSTI2021

Artificial Intelligence for Multiphysics Nuclear Design Optimization with Additive Manufacturing

The geometric flexibility of additively manufactured metals and ceramics generates a very large and open design space that requires advanced modeling and simulation tools for physics simulations and the rigorous definition of design problems. This effort deploys artificial intelligence (AI) and machine learning (ML) algorithms to understand the design space, evaluate potential designs, and more efficiently generate optimized results. The Transformational Challenge Reactor (TCR) program is leveraging advances in several scientific areas—including materials, manufacturing, sensors and control systems, data analytics, and high-fidelity modeling and simulation—to accelerate the design, manufacturing, qualification, and deployment of advanced nuclear energy systems. Through a manufacturing-informed design approach, the TCR program seeks to integrate digital data for rapid nuclear innovation; accelerate the adoption of advances in manufacturing, materials, and computational sciences for nuclear applications; and dramatically reduce deployment costs and timelines for new nuclear reactor technologies. This report documents efforts under the TCR program to leverage advanced modeling and simulation techniques driven by AI/ML algorithms on high-performance computing (HPC) systems to yield more optimized TCR core designs. A multiphysics ML surrogate model was developed to run on the HPC architectures. The surrogate model is trained on high-fidelity simulation data of coupled neutronics and thermofluidics and is used to quickly evaluate thousands of candidate core designs in parallel, which drives the evolution of the cooling channel shapes to minimize temperature peaking and material stress. Outcomes from these activities provide design information and feedback into the core design efforts.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2022

Reducing the Overnight Capital Cost of Advanced Reactors Using Equipment-Level Seismic Protective Systems

Consideration of the effects of earthquake shaking on the design and construction of nuclear power plants adds substantially to the overnight capital cost, with anecdotal estimates as high as 35+%, attributed to additional construction materials, need for one-off and sub-optimal designs of equipment due to conflicting design choices, the high cost of seismic qualification of equipment, and regulatory review. Safety-critical equipment in large light water reactors is designed and qualified for seismic demands imposed by the supporting reactor building, optimal mechanical designs are not possible, and designs of a given piece of equipment may vary with height above grade. Similar negative impacts are expected for advanced reactors unless the seismic design paradigm is changed. The overarching goal of this transformational MEITNER project, which involved a multidisciplinary engineering team and designers of three fundamentally different advanced reactors, was to adapt proven seismic isolation and damping technologies to operationalize modular protective systems for safety-class equipment inside advanced reactor buildings. Such seismic protective systems would be tightly integrated into design development for reactor support systems and balance-of-plant construction. The adoption of the technology, which is widely used in non-nuclear sectors, would simplify plant design, enable the use of standardized equipment and buildings, optimized for operational performance, and reduce plant size and weight. The need for site-specific equipment would be eliminated, enabling identical equipment to be used across multiple plants sited across the US and economies of scale, and catalyzing new interest and investment. The equipment-based protective systems would allow siting of advanced reactors in regions of high seismic hazard.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2023

Development of a compact fast-neutron spectrometer for nuclear emergency response applications

We have developed a Compact Fast Neutron Spectrometer (CFNS) for passive assay of special nuclear material (SNM) through the observation of fast neutrons. The CFNS consists of eight organic glass scintillators (OGS) coupled to silicon photomultipliers and a waveform digitizer, which are integrated within a human-portable box. The CFNS determines the neutron energy profile by spectrum unfolding using the Maximum-Likelihood Expectation Maximization method. The detector acquisition system was optimized to have a dynamic range of up to 10 MeV neutron energy. Bulk special nuclear material (SNM) measurements from the National Criticality Experiments Research Center were analyzed for SNM validation/examination. Additionally, the results show that the CFNS can be used to distinguish between fission and (α, n) neutron emitters, regardless of intervening material type (Cu and polyethylene) and thickness, by taking the ratio of neutron counts at different regions in the unfolded energy spectrum. Additionally, by fitting an exponential curve to the unfolded energy spectrum of PuO 2 and Pu neutron emitters, the CFNS showed the ability of distinguishing between pure Pu oxide, pure Pu metal and mixed oxide-metal configurations.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2020

Constrained Bayesian Optimization of Criticality Experiments at LLNL [Slides]

The design of criticality experiments is typically an iterative process that employs a Monte Carlo transport code. The goal is to find a design that optimizes some variable, like the sensitivity of a response to a cross section, while simultaneously ensuring criticality. The high fidelity of the Monte Carlo code is a great asset, but it makes exploring the design space computationally expensive. Herein, we present how a constrained Bayesian optimization algorithm can be used to efficiently design a criticality experiment. It uses Gaussian processes as a surrogate model to probe the design space and to reduce the number of code executions that are needed to find the optimum. We demonstrate constrained Bayesian optimization with a Pu-239/polyethylene solution system and a TEX experiment that is designed for critical ity safety validation of a nuclear waste model at the Hanford Site. For both systems, a global optimum was found within 75 Monte Carlo simulations.

42 ENGINEERING↗
DOE OSTI2020

Constrained Bayesian Optimization of Criticality Experiments [Slides]

The design of criticality experiments is typically an iterative process that employs a Monte Carlo transport code. The goal is to find a design that optimizes some variable, like the sensitivity of a response to a cross section, while simultaneously ensuring criticality. The high fidelity of the Monte Carlo code is a great asset, but it makes exploring the design space computationally expensive. Herein, we present how a constrained Bayesian optimization algorithm can be used to efficiently design a criticality experiment. It uses Gaussian processes as a surrogate model to probe the design space and to reduce the number of code executions that are needed to find the optimum. We demonstrate constrained Bayesian optimization with a Pu-239/polyethylene solution system and a TEX experiment that is designed for criticality safety validation of a nuclear waste model at the Hanford Site. For both systems, a global optimum was found within 75 Monte Carlo simulations.

42 ENGINEERING↗
DOE OSTI2021

A Hybrid Energy System Workflow for Energy Portfolio Optimization

This manuscript develops a workflow, driven by data analytics algorithms, to support the optimization of the economic performance of an Integrated Energy System. The goal is to determine the optimum mix of capacities from a set of different energy producers (e.g., nuclear, gas, wind and solar). A stochastic-based optimizer is employed, based on Gaussian Process Modeling, which requires numerous samples for its training. Each sample represents a time series describing the demand, load, or other operational and economic profiles for various types of energy producers. These samples are synthetically generated using a reduced order modeling algorithm that reads a limited set of historical data, such as demand and load data from past years. Numerous data analysis methods are employed to construct the reduced order models, including, for example, the Auto Regressive Moving Average, Fourier series decomposition, and the peak detection algorithm. All these algorithms are designed to detrend the data and extract features that can be employed to generate synthetic time histories that preserve the statistical properties of the original limited historical data. The optimization cost function is based on an economic model that assesses the effective cost of energy based on two figures of merit: the specific cash flow stream for each energy producer and the total Net Present Value. An initial guess for the optimal capacities is obtained using the screening curve method. The results of the Gaussian Process model-based optimization are assessed using an exhaustive Monte Carlo search, with the results indicating reasonable optimization results. The workflow has been implemented inside the Idaho National Laboratory’s Risk Analysis and Virtual Environment (RAVEN) framework. The main contribution of this study addresses several challenges in the current optimization methods of the energy portfolios in IES: First, the feasibility of generating the synthetic time series of the periodic peak data; Second, the computational burden of the conventional stochastic optimization of the energy portfolio, associated with the need for repeated executions of system models; Third, the inadequacies of previous studies in terms of the comparisons of the impact of the economic parameters. The proposed workflow can provide a scientifically defendable strategy to support decision-making in the electricity market and to help energy distributors develop a better understanding of the performance of integrated energy systems.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2024

Leveraging Optimal Sparse Sensor Placement to Aggregate a Network of Digital Twins for Nuclear Subsystems

Nuclear power plants (NPPs) require continuous monitoring of various systems, structures, and components to ensure safe and efficient operations. The critical safety testing of new fuel compositions and the analysis of the effects of power transients on core temperatures can be achieved through modeling and simulations. They capture the dynamics of the physical phenomenon associated with failure modes and facilitate the creation of digital twins (DTs). Accurate reconstruction of fields of interest (e.g., temperature, pressure, velocity) from sensor measurements is crucial to establish a two-way communication between physical experiments and models. Sensor placement is highly constrained in most nuclear subsystems due to challenging operating conditions and inherent spatial limitations. This study develops optimized data-driven sensor placements for full-field reconstruction within reactor and steam generator subsystems of NPPs. Optimized constrained sensors reconstruct field of interest within a tri-structural isotropic (TRISO) fuel irradiation experiment, a lumped parameter model of a nuclear fuel test rod and a steam generator. The optimization procedure leverages reduced-order models of flow physics to provide a highly accurate full-field reconstruction of responses of interest, noise-induced uncertainty quantification and physically feasible sensor locations. Accurate sensor-based reconstructions establish a foundation for the digital twinning of subsystems, culminating in a comprehensive DT aggregate of an NPP.

42 ENGINEERING↗
DOE OSTI2025

A Review of the Research and Development of Brayton Cycle Technology in Nuclear Power Applications with a Focus on Compressor Technology

This study reviews the integration of Brayton Cycle (BC) systems in nuclear power generation, emphasizing their potential to enhance thermal efficiency and operational flexibility over traditional Rankine Cycle (RC) systems. Key working fluids, such as helium (He), supercritical carbon dioxide (sCO 2 ), nitrogen (N 2 ), and air, are evaluated for their performance, efficiency, and compatibility with nuclear systems. He is recognized for its high thermal conductivity and inertness at elevated temperatures, while sCO 2 demonstrates advantages in compactness and efficiency in midrange temperatures. This article also highlights the importance of compressor designs in optimizing BC performance and reviews, available compressor technologies. Axial and centrifugal compressor designs enable efficient gas compression while managing the thermal and mechanical stresses associated with high-pressure operations in nuclear systems. Combined with variable geometry components and advanced materials, these technologies address the challenges posed by varying load conditions. Despite the promising features of BC systems, several challenges persist, including high leakage rates and material degradation under extreme conditions, which necessitate robust sealing technologies and thorough testing. The insights gained from operational experiences at facilities, such as the Oberhausen II plant and the High-Temperature He Test Facility (HHV), underscore the complexities involved in designing high-temperature gas turbines for nuclear applications. This review concludes that as the nuclear industry evolves, BC systems hold significant promise for contributing to a sustainable energy future, particularly in the context of small modular reactors (SMRs) and microreactors. Further exploration of combined cycle configurations that combine BCs with RCs may enhance overall efficiency and flexibility in power generation.

22 - GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2024

Optimizing Hydride Stability in U-ZrH x Nuclear Fuel: The “Goldilocks Radius”

Nuclear-powered microreactors show great promise for opening new nuclear energy markets due to the flexibility offered by their rapid/streamlined in-factory fabrication, transportability, and self-regulating nature. The economic benefits of any commercialized nuclear reactor, however, rely on the system’s ability to produce large amounts of heat and efficiently convert that heat into electrical power reliably for long periods of time. Uranium-zirconium hydride (U-ZrH x ) is currently being considered for compact reactor designs because it is a well-known nuclear fuel system that is self-moderating, but this fuel, which has historically been used for research reactors, has not been optimized for commercial power production. Here, this paper analyzes the hydride stability of standard 304 stainless steel–clad U-ZrH x fuel under commercially relevant conditions. Fuel element design parameters, including physical dimensions, as-fabricated hydrogen content, burnup, peak fuel temperature, temperature gradient, operational fuel cycle duration, and volumetric heat generation rate, are discussed with a focus on hydrogen distribution and phase stability within the fuel element. Hydride stability declines more rapidly as the coolant temperature, burnup, and fuel cycle duration increase. Using a fuel-cladding gap material with heat transfer properties superior to air, such as helium or sodium, is essential to prolonging fuel hydride stability. The fuel’s physical dimensions are also important. At very small fuel diameters, the H/Zr ratio in the fuel meat decreases too rapidly due to the hydrogen content’s dependence on fuel meat volume. Conversely, the fuel meat temperature and temperature gradient exacerbate hydrogen loss at very large fuel diameters. We find that the most important parameter to consider when optimizing the hydride stability of U-ZrH x fuel is the relationship between the fuel meat radius and the power density in the fuel. A simple equation is empirically determined that relates the “Goldilocks radius,” that is, the fuel radius for which the H/Zr ratio is most stable, to the power density in the fuel.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2024

ORCA Software Quality Assurance Plan

This document summarizes the software quality assurance (SQA) planning activities conducted for the Optimization of Real-time Capacity Allocation (ORCA) plug-in. It outlines the approaches and document structures adopted to maintain high standards of software quality. It also gives examples of certain SQA documents.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2023

Measurement of Photoneutrons from Depleted Uranium and Comparison Study Using MCNPX-PoliMi

High-energy photon interrogation is a nondestructive technique that is used to detect special nuclear materials and characterize nuclear waste. The development of such systems is complex and requires Monte Carlo simulations to optimize system performance. Monte Carlo simulations rely on various scattering, absorption, and photonuclear cross-section data. While the scattering and absorption cross-section data have been extensively studied and validated with experiments, the results obtained from photonuclear simulations are often found to underpredict measured results, indicating uncertainties in the photonuclear cross sections themselves. Thus, there is a need for new measured results that can be used to quantify underpredictions in simulations using photonuclear cross-section data. In the present work, we interrogated depleted uranium with a 9-MV electron linac and detected photoneutrons with trans-stilbene organic scintillators. The measurement of photoneutrons with organic scintillators is challenging due to the presence of the intense photon flux, which causes issues such as pulse pile-up, detector saturation, and poor signal-to-background ratio. To mitigate these challenges, we used iron and polyethylene shielding of varying thicknesses around the depleted uranium target and a neural network–based digital pulse processing algorithm to recover neutron and photon information from piled-up events. Our goal was to compare the measured photoneutron count rate with the simulated rate obtained using the MCNPX-PoliMi transport code. For a light output window of 0.28 to 2.67 MeVee (1.66- to 6.85-MeV proton recoil energy), we found that the simulated count rate obtained using the ENDF/B-VII photonuclear cross-section library underpredicts the measured rate by 32.8% 3.2%. Additionally, we compared the simulated and measured photoneutron light output distributions. For the least thicknesses of shielding, the simulation was found to underpredict measurements in the 0.70- to 2.67-MeVee light output window. For the greatest thicknesses of shielding, the simulation was found to underpredict the measurement across the entire light output window of 0.28 to 2.67 MeVee.

98 NUCLEAR DISARMAMENT, SAFEGUARDS, AND PHYSICAL P↗
DOE OSTI2023

Tools and Methods for Optimization of Nuclear Plant Outages

Refueling outages are one of the most challenging phases in a nuclear power plants (NPP) operating cycle. Refueling outages are extremely costly for a NPP due to the large amount of required resources and because of lost revenue due to plant being off the grid. Outage durations have steadily decreased across the industry over that last few decades primarily due to improved planning and coordination, but there are still many plants that struggle to meet the performance metrics accomplished by other utilities. Schedule resilience is one of the issues. NPP outages require scheduling thousands of activities in a duration of around 30 days on average. Outage staff begin working on the schedule more than a year ahead of the outage start and make every effort to build a robust schedule. Despite the robust and detailed planning, once the outage starts, numerous emergent issues typically appear along with schedule delays requiring continuous replanning and adjusting. When schedule disruption occurs during an outage, plant staff make urgent efforts to recover but often not able to maintain the planned outage duration. These outage delays can cost a utility several million dollars per day. Tools that could help outage schedulers create a more resilient schedule and allow them to optimally reschedule emergent work could significantly reduce outage delays. One key aspect of creating a resilient schedule is to have accurate estimates for activity duration. Another important outage scheduling capability is the ability to schedule emergent work with minimal disruption. The Optimization of Outage Activities project under the Risk-Informed Systems Analysis Pathway (RISA) sponsored by Department of Energy (DOE) Light Water Reactor Sustainability (LWRS) program focuses on developing tools and methods to support nuclear power plants with optimization of outage schedules. The goal of the outage optimization is the completion of all planned and emergent outage activities as fast as possible while maintaining highest level of safety. This report describes the initial development of tools to support outage management that leverage computational and machine learning methods developed in other RISA and LWRS projects.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2020

Constrained Bayesian optimization of criticality experiments

The design of criticality experiments is typically an iterative process that employs a Monte Carlo transport code. The goal is to find a design that optimizes some variable, like the sensitivity of a response to a cross section, while simultaneously ensuring criticality. The high fidelity of the Monte Carlo code is a great asset, but it makes exploring the design space computationally expensive. Herein, we present how a constrained Bayesian optimization algorithm can be used to efficiently design a criticality experiment. It uses Gaussian processes as a surrogate model to probe the design space and to reduce the number of code executions that are needed to find the optimum. Furthermore, we demonstrate constrained Bayesian optimization with a Pu-239/polyethylene solution system and a TEX experiment that is designed for criticality safety validation of a nuclear waste model at the Hanford Site. For both systems, a global optimum was found within 75 Monte Carlo simulations.

42 ENGINEERING↗
DOE OSTI2024

FORCE Update 2024

The Framework for Optimization of Resources and Economics (FORCE) tool suite is the U.S. Department of Energy’s Nuclear Integrated Energy Systems (IES) Program flagship tool suite for technoeconomic IES analysis of IES. This tool suite is useful for analysis designed to evaluate and improve the technoeconomics of energy production systems, particularly for systems including nuclear technology. In this report, we document the development activity for the FORCE tool suite to extend its capabilities as performed during fiscal year 2024. In addition to reliability and accessibility, capability is one of the three standards guiding the development of the FORCE tool suite and the software codes that are its constituent parts. Extending the capabilities of the FORCE tool suite allows analysis both within the IES program as well as industry, university, and laboratory partners to perform analysis with more accuracy, insight, and impactful narrative. Four areas of capability development were the focus of activity this year: economic parameter uncertainty quantification, multiresolution analysis, components-to-optimization workflow automation, and statespace construction workflows for real-time optimal control. In economic parameter uncertainty quantification, the ability of HERON to capture risk due to scenarios (weather and energy demand uncertainty) was expanded to also include uncertainties in financial parameters such as capital cost or operation and maintenance costs. By including these sources of uncertainty, which are sometimes very large compared with scenario uncertainty, HERON is better able to capture the risk posed by investment in various IES technology. Because of this, analysts can also consider the reduction in risks that can be realized by choice of some technologies. In multiresolution analysis, development activity extended on work completed previously. In fiscal year 2023, methods for decomposing time series signals, such as demand, solar and wind availability, and price profiles, were analyzed and down-selected to those most effective at splitting signals into different resolutions. These resolutions allow considering the influence of different energy demand and supply behaviors across different time scales. For example, energy demand might be divided into seasonal, weekly, and hourly profiles. In fiscal year 2024, this preliminary work was extended and implemented within the Risk Analysis Virtual Environment (RAVEN) risk and uncertainty analysis platform, which is used throughout the FORCE framework. This development of the “multi-resolution time series analysis” (MR-TSA) module in RAVEN allows training synthetic history generators on complex time series. These synthetic history generators can then be used in HERON for generating scenarios that represent possible market and weather scenarios that can be analyzed on different time scales. We envision completing this work in the future, implementing multiresolution dispatch optimization strategies that can make the most beneficial use of these stratified time histories. In components-to-optimization workflow development, workflows for translating user inputs of components into algorithms for algebraic optimization were selected and implemented. Similar algorithms within the Holistic Energy Resource Optimization Network (HERON) were separated from the main code base of HERON and gathered with the components-to-optimization workflows in the new Dispatch Optimization Variable Engine (DOVE) software library. This modularization allows FORCE users to analyze dispatch optimization and energy system duty cycles independently of HERON, which previously was a burdensome task. Additionally, these dispatch optimization algorithms, set up in an independent library, can now be used across all software applications within FORCE, especially including the real-time optimal control software Optimization of Real-time Capacity Allocation (ORCA). Allowing FORCE software to share dispatch optimization algorithms within a single library allows for improved software maintenance and reliability. In statespace characterization workflow development, alternative workflows for optimizing dispatch with additional technical accuracy was the focus, particularly to improve the real-time optimization decision making in ORCA. Using algorithms and workflows initially developed for the Feasible Actuator Range Modifier (FARM), workflows for determining the statespace representation of IES were identified and demonstrated. The resulting dispatch optimization required a more robust optimization algorithm than that originally used in HERON (and moved to DOVE), which required adding an alternate workflow to DOVE that can more accurately match the behavior of physical systems using a partial differential equation representation. In conclusion, capability developments in the FORCE tool suite in fiscal year 2024 have improved the ability of the FORCE tool suite to perform

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2021

Machine learning-enabled prediction of chemical durability of A 2 B 2 O 7 pyrochlore and fluorite

Pyrochlore-structure type and its derivative in a general formula A 2 B 2 O 7 (A = rare earth elements and actinides; B = Ti, Sn, Zr, Hf, Pb, Si, etc.) display excellent structural flexibility and rich crystal chemistry as promising nuclear waste form materials capable of immobilizing actinides and fission products. It is essential to understand these materials’ chemical durability and element release of radionuclides in order to evaluate their performance in near-field environment. However, it is a formidable grand technological challenge to experimentally perform durability testing across hundreds of thousands of possibilities resulting from their extreme compositional complexities due to cation substitutions at both A and B-sites. In this work, we demonstrate a machine learning approach to determine the key materials parameters and structural characteristics governing the leaching behaviors from a small set of selected compositions as model systems, enabling a science-based prediction of their chemical durability that can be extended to a wide range of chemical compositions. The combination of four key structural characteristics and materials parameters, including ionic radius size difference , ionic potential difference , electronegativity difference , and lattice parameter , creates features an optimized prediction of the chemical durability. Two machine learning models, linear regression and Kernel ridge regression models, are trained on the randomly-split training dataset derived from the experimentally-determined elemental release rates, and subsequently tested on the testing dataset. The predicted leaching rates from both machine learning models show an excellent agreement with the experimental data, demonstrating the feasibility of rapidly evaluating the material properties of new compositions. These results highlight the immense potential of synergizing informatics through machine learning-based models and well-controlled experiments of selected model systems to accelerate materials design and discovery with optimized compositions and performance of promising materials for effective nuclear waste management.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2022

Regional Hybrid Energy Systems Technoeconomic Analysis

This presentation summarizes the final outputs from a multi-year HFTO-funded project that evaluated the potential for hybridized nuclear power plants to economically produce hydrogen. The focus of this presentation is on the interactions between the electricity system modeling and hydrogen system optimization and the resulting figures of merit. This project partnered with Idaho National Laboratory, Argonne National Laboratory, Electric Power Research Institute, and Xcel Energy.

ENERGY PLANNING, POLICY, AND ECONOMY,HYDROGEN↗
DOE OSTI2024

Efficient Parallelization of Irregular Applications on GPU Architectures

With the enlarging computation capacity of general Graphics Processing Units (GPUs), leveraging GPUs to accelerate parallel applications has become a critical topic in academia and industry. However, a wide range of irregular applications with the computation-/memory-intensive nature cannot easily achieve high GPU utilization. The challenges mainly involve the following aspects: first, data dependence leads to coarse-grained kernel and inefficient parallelism; second, heavy GPU memory usage may cause frequent memory evictions and extra overhead of I/O; third, specific computation patterns produce memory redundancies; last, workload balance and data reusability conjunctly benefit the overall performance, but there may exist a dynamic trade-off between them. Targeting these challenges, this dissertation proposes multiple optimizations to accelerate two real-world applications: many-body correlation functions to simulate nuclear physics in a large-scale scientific system; the other is the eALS-based matrix factorization recommendation system. To accelerate the calculations of many-body correlation functions, this dissertation presents three frameworks in GPU memory management and multi-GPU scheduling. Firstly, an optimized systematic GPU memory management framework, MemHC, utilizes a series of new memory reduction designs in GPU memory allocation, CPU/GPU communications, and GPU memory oversubscription. Secondly, an enhanced multi-GPU scheduling framework, MICCO, particularly by taking both data dimension (e.g., data reuse and data eviction) and computation dimension into account. MICCO designs a heuristic scheduling algorithm and a machine learning-based regression model to generate the optimal settings of a proposed new concept to manage the trade-off. Thirdly, a locality-aware multi-GPU scheduling framework. This scheduler leverages pipeline batch generation with a looking-ahead strategy by building local dependency graphs for memory transfer reduction and better data reuse, achieving up to 79.92% memory cost reduction and 1.67x speedup. To parallelize the eALS-based recommendation system, this dissertation proposes an efficient CPU/GPU heterogeneous recommendation system, HEALS. HEALS employs newly designed architecture-adaptive data formats to achieve load balance and good data locality on CPU and GPU. To mitigate the data dependence, HEALS presents a CPU/GPU collaboration model for both task parallelism and data parallelism with multiple kernel computation optimizations. In summary, this dissertation efficiently accelerates two typical irregular applications on GPUs by building four frameworks, including CPU/GPU collaboration, GPU memory management, and multi-GPU scheduling.

Wang, Qihan↗