Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “heterogeneous computing”

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 379 records · Page 21
All sourcesNASA NTRSDOE OSTI
DOE OSTI2013

Accelerating Subsurface Transport Simulation on Heterogeneous Clusters

Reactive transport numerical models simulate chemical and microbiological reactions that occur along a flowpath. These models have to compute reactions for a large number of locations. They solve the set of ordinary differential equations (ODEs) that describes the reaction for each location through the Newton-Raphson technique. This technique involves computing a Jacobian matrix and a residual vector for each set of equation, and then solving iteratively the linearized system by performing Gaussian Elimination and LU decomposition until convergence. STOMP, a well known subsurface flow simulation tool, employs matrices with sizes in the order of 100x100 elements and, for numerical accuracy, LU factorization with full pivoting instead of the faster partial pivoting. Modern high performance computing systems are heterogeneous machines whose nodes integrate both CPUs and GPUs, exposing unprecedented amounts of parallelism. To exploit all their computational power, applications must use both the types of processing elements. For the case of subsurface flow simulation, this mainly requires implementing efficient batched LU-based solvers and identifying efficient solutions for enabling load balancing among the different processors of the system. In this paper we discuss two approaches that allows scaling STOMP's performance on heterogeneous clusters. We initially identify the challenges in implementing batched LU-based solvers for small matrices on GPUs, and propose an implementation that fulfills STOMP's requirements. We compare this implementation to other existing solutions. Then, we combine the batched GPU solver with an OpenMP-based CPU solver, and present an adaptive load balancer that dynamically distributes the linear systems to solve between the two components inside a node. We show how these approaches, integrated into the full application, provide speed ups from 6 to 7 times on large problems, executed on up to 16 nodes of a cluster with two AMD Opteron 6272 and a Tesla M2090 per node.

GPU↗
DOE OSTI2012

Long-term Differences in Tillage and Land Use Affect Intra-aggregate Pore Heterogeneity

Recent advances in computed tomography provide measurement tools to study internal structures of soil aggregates at micrometer resolutions and to improve our understanding of specific mechanisms of various soil processes. Fractal analysis is one of the data analysis tools that can be helpful in evaluating heterogeneity of the intra-aggregate internal structures. The goal of this study was to examine how long-term tillage and land use differences affect intra-aggregate pore heterogeneity. The specific objectives were: (i) to develop an approach to enhance utility of box-counting fractal dimension in characterizing intra-aggregate pore heterogeneity; (ii) to examine intra-aggregate pores in macro-aggregates (4-6 mm in size) using the computed tomography scanning and fractal analysis, and (iii) to compare heterogeneity of intra-aggregate pore space in aggregates from loamy Alfisol soil subjected to 20 yr of contrasting management practices, namely, conventional tillage (chisel plow) (CT), no-till (NT), and native succession vegetation (NS). Three-dimensional images of the intact aggregates were obtained with a resolution of 14.6 {micro}m at the Advanced Photon Source, Argonne National Laboratory, Argonne, IL. Proposed box-counting fractal dimension normalization was successfully implemented to estimate heterogeneity of pore voxel distributions without bias associated with different porosities in soil aggregates. The aggregates from all three studied treatments had higher porosity associated with large (>100 {micro}m) pores present in their centers than in their exteriors. Pores 15 to 60 {micro}m were equally abundant throughout entire aggregates but their distributions were more heterogeneous in aggregate interiors. The CT aggregates had greater numbers of pores 15 to 60 {micro}m than NT and NS. Distribution of pore voxels belonging to large pores was most heterogeneous in the aggregates from NS, followed by NT and by CT. This result was consistent with presence of very large pores associated with former root channels in NT and NS aggregates. Our findings indicate that mechanisms of macro-aggregate formation might differ in their importance in different land use and management practices.

43 PARTICLE ACCELERATORS↗
DOE OSTI2016

Computationally Probing the Performance of Hybrid, Heterogeneous, and Homogeneous Iridium-Based Catalysts for Water Oxidation

An attractive strategy to improve the performance of water oxidation catalysts would be to anchor a homogeneous molecular catalyst on a heterogeneous solid surface to create a hybrid catalyst. The idea of this combined system is to take advantage of the individual properties of each of the two catalyst components. We use Density Functional Theory to determine the stability and activity of a model hybrid water oxidation catalyst consisting of a dimeric Ir complex attached on the IrO 2 (110) surface through two oxygen atoms. We find that homogeneous catalysts can be bound to its matrix oxide without losing significant activity. Hence, designing hybrid systems that benefit from both the high tunability of activity of homogeneous catalysts and the stability of heterogeneous systems seems feasible.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

GPU-accelerated DNS of compressible turbulent flows

Here, this paper explores strategies to transform an existing CPU-based high-performance computational fluid dynamics solver, HyPar, for compressible flow simulations on emerging exascale heterogeneous (CPU+GPU) computing platforms. The scientific motivation for developing a GPU-enhanced version of HyPar is to simulate canonical turbulent flows at the highest resolution possible on such platforms. We show that optimizing memory operations and thread blocks results in 200x speedup of computationally intensive kernels compared with a CPU core. Using multiple GPUs and CUDA-aware MPI communication, we demonstrate both strong and weak scaling of our GPU-based HyPar implementation on the NVIDIA Volta V100 GPUs. We simulate the decay of homogeneous isotropic turbulence in a triply periodic box on grids with up to 1024 3 points (5.3 billion degrees of freedom) and on up to 1,024 GPUs. We compare the wall times for CPU-only and CPU+GPU simulations. The results presented in the paper are obtained on the Summit and Lassen supercomputers at Oak Ridge and Lawrence Livermore National Laboratories, respectively.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Secure Federated Learning Across Heterogeneous Cloud and High-Performance Computing Resources: A Case Study on Federated Fine-Tuning of LLaMA 2

Federated learning enables multiple data owners to collaboratively train robust machine learning models without transferring large or sensitive local datasets by only sharing the parameters of the locally trained models. Here, in this article, we elaborate on the design of our Advanced Privacy-Preserving Federated Learning (APPFL) framework, which streamlines end-to-end secure and reliable federated learning experiments across cloud computing facilities and high-performance computing resources by leveraging Globus Compute, a distributed function as a service platform, and Amazon Web Services. We further demonstrate the use case of APPFL in fine-tuning an LLaMA 2 7B model using several cloud resources and supercomputers.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2017

rvGAHP: push-based job submission using reverse SSH connections. In: WORKS '17 Proceedings of the 12th Workshop on Workflows in Support of Large-Scale Science, Article No. 3

Computational science researchers running large-scale scientific workflow applications often want to run their workflows on the largest available compute systems to improve time to solution. Workflow tools used in distributed, heterogeneous, high performance computing environments typically rely on either a push-based or a pull-based approach for resource provisioning from these compute systems. However, many large clusters have moved to two-factor authentication for job submission, making traditional automated push-based job submission impossible. On the other hand, pull-based approaches such as pilot jobs may lead to increased complexity and a reduction in node-hour efficiency. In this paper, we describe a new, efficient approach based on HTCondor-G called reverse GAHP (rvGAHP) that allows us to push jobs using reverse SSH submissions with better efficiency than pull-based methods. We successfully used this approach to perform a large probabilistic seismic hazard analysis study using SCEC's CyberShake workflow in March 2017 on the Titan Cray XK7 hybrid system at Oak Ridge National Laboratory.

Callaghan, Scott↗
DOE OSTI2021

Using computational singular perturbation as a diagnostic tool in ODE and DAE systems: a case study in heterogeneous catalysis

We have extended the computational singular perturbation (CSP) method to differential algebraic equation (DAE) systems and demonstrated its application in a heterogeneous-catalysis problem. The extended method obtains the CSP basis vectors for DAEs from a reduced Jacobian matrix that takes the algebraic constraints into account. Here we use a canonical problem in heterogeneous catalysis, the transient continuous stirred tank reactor (T-CSTR), for illustration. The T-CSTR problem is modelled fundamentally as an ordinary differential equation (ODE) system, but it can be transformed to a DAE system if one approximates typically fast surface processes using algebraic constraints for the surface species. We demonstrate the application of CSP analysis for both ODE and DAE constructions of a T-CSTR problem, illustrating the dynamical response of the system in each case. We also highlight the utility of the analysis in commenting on the quality of any particular DAE approximation built using the quasi-steady state approximation (QSSA), relative to the ODE reference case.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2018

Cabana

Physics simulations that use particles in their algorithms are ubiquitous on modern high performance computers. Cabana is a C++ library for developing particle-based simulations on heterogeneous high performance computing platforms. Cabana provides general data structures and algorithms for common particle operations in a way that is both performant and portable on HPC systems. This includes particle data structure allocatable and configurable on GPU and CPU architectures, and threaded and distributed algorithms for sorting particles, halo exchanges, neighbor list generation, and other basic particle operations.

Slattery, StuartR↗
DOE OSTI2021

SBIR Phase I Final Report, TACO: Distributed and Heterogeneous Sparse Compiler

Tensor algebra is a powerful tool for computing, but writing optimized codes that operate on sparse tensors can be very complex. This project enables a Tensor Algebra Compiler (TACO) that simplifies this task from man-years to man-days and extends TACO to support complex and large distributed systems. This report details the hypotheses, approaches used, and findings in this project.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2018

A hydrological emulator for global applications – HE v1.0.0

Abstract. While global hydrological models (GHMs) are very useful in exploring water resources and interactions between the Earth and human systems, their use often requires numerous model inputs, complex model calibration, and high computation costs. To overcome these challenges, we construct an efficient open-source and ready-to-use hydrological emulator (HE) that can mimic complex GHMs at a range of spatial scales (e.g., basin, region, globe). More specifically, we construct both a lumped and a distributed scheme of the HE based on the monthly abcd model to explore the tradeoff between computational cost and model fidelity. Model predictability and computational efficiency are evaluated in simulating global runoff from 1971 to 2010 with both the lumped and distributed schemes. The results are compared against the runoff product from the widely used Variable Infiltration Capacity (VIC) model. Our evaluation indicates that the lumped and distributed schemes present comparable results regarding annual total quantity, spatial pattern, and temporal variation of the major water fluxes (e.g., total runoff, evapotranspiration) across the global 235 basins (e.g., correlation coefficient r between the annual total runoff from either of these two schemes and the VIC is >0.96), except for several cold (e.g., Arctic, interior Tibet), dry (e.g., North Africa) and mountainous (e.g., Argentina) regions. Compared against the monthly total runoff product from the VIC (aggregated from daily runoff), the global mean Kling–Gupta efficiencies are 0.75 and 0.79 for the lumped and distributed schemes, respectively, with the distributed scheme better capturing spatial heterogeneity. Notably, the computation efficiency of the lumped scheme is 2 orders of magnitude higher than the distributed one and 7 orders more efficient than the VIC model. A case study of uncertainty analysis for the world's 16 basins with top annual streamflow is conducted using 100 000 model simulations, and it demonstrates the lumped scheme's extraordinary advantage in computational efficiency. Our results suggest that the revised lumped abcd model can serve as an efficient and reasonable HE for complex GHMs and is suitable for broad practical use, and the distributed scheme is also an efficient alternative if spatial heterogeneity is of more interest.

Liu, Yaling↗
DOE OSTI2022

The revised FLORIDyn model: implementation of heterogeneous flow and the Gaussian wake

In this paper, a new version of the FLOw Redirection and Induction Dynamics (FLORIDyn) model is presented. The new model uses the three-dimensional parametric Gaussian FLORIS model and can provide dynamic wind farm simulations at a low computational cost under heterogeneous and changing wind conditions. Both FLORIS and FLORIDyn are parametric models which can be used to simulate wind farms, evaluate controller performance and can serve as a control-oriented model. One central element in which they differ is in their representation of flow dynamics: FLORIS neglects these and provides a computationally very cheap approximation of the mean wind farm flow. FLORIDyn defines a framework which utilizes this low computational cost of FLORIS to simulate basic wake dynamics. This is achieved by creating so-called observation points (OPs) at each time step at the rotor plane which inherit the turbine state. In this work, we develop the initial FLORIDyn framework further considering multiple aspects. The underlying FLORIS wake model is replaced by a Gaussian wake model. The distribution and characteristics of the OPs are adapted to account for the new parametric model but also to take complex flow conditions into account. To achieve this, a mathematical approach is developed to combine the parametric model and the changing, heterogeneous world conditions and link them with each OP. We also present a computationally lightweight wind field model to allow for a simulation environment in which heterogeneous flow conditions are possible. FLORIDyn is compared to Simulator for Offshore Wind Farm Applications (SOWFA) simulations in three- and nine-turbine cases under static and changing environmental conditions. The results show a good agreement with the timing of the impact of upstream state changes on downstream turbines. They also show a good agreement in terms of how wakes are displaced by wind direction changes and when the resulting velocity deficit is experienced by downstream turbines. A good fit of the mean generated power is ensured by the underlying FLORIS model. In the three-turbine case, FLORIDyn simulates 4 s simulation time in 24.49 ms computational time. The resulting new FLORIDyn model proves to be a computationally attractive and capable tool for model-based dynamic wind farm control.

17 WIND ENERGY↗
DOE OSTI2024

PanDA: Production and Distributed Analysis System

The Production and Distributed Analysis (PanDA) system is a data-driven workload management system engineered to operate at the LHC data processing scale. The PanDA system provides a solution for scientific experiments to fully leverage their distributed heterogeneous resources, showcasing scalability, usability, flexibility, and robustness. The system has successfully proven itself through nearly two decades of steady operation in the ATLAS experiment, addressing the intricate requirements such as diverse resources distributed worldwide at about 200 sites, thousands of scientists analyzing the data remotely, the volume of processed data beyond the exabyte scale, dozens of scientific applications to support, and data processing over several billion hours of computing usage per year. PanDA’s flexibility and scalability make it suitable for the High Energy Physics community and wider science domains at the Exascale. Beyond High Energy Physics, PanDA’s relevance extends to other big data sciences, as evidenced by its adoption in the Vera C. Rubin Observatory and the sPHENIX experiment. As the significance of advanced workflows continues to grow, PanDA has transformed into a comprehensive ecosystem, effectively tackling challenges associated with emerging workflows and evolving computing technologies. The paper discusses PanDA’s prominent role in the scientific landscape, detailing its architecture, functionality, deployment strategies, project management approaches, results, and evolution into an ecosystem.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1987

Concerning the relationship between evapotranspiration and soil moisture

The relationship between the evapotranspiration and soil moisture during the drying, supply-limited phase is studied. A second scaling parameter, based on the evapotranspirational supply and demand concept of Federer (1982), is defined; the parameter, referred to as the threshold evapotranspiration, occurs in vegetation-covered surfaces just before leaf stomata close and when surface tension restricts moisture release from bare soil pores. A simple model for evapotranspiration is proposed. The effects of natural soil heterogeneities on evapotranspiration computed from the model are investigated. It is observed that the natural variability in soil moisture, caused by the heterogeneities, alters the relationship between regional evapotranspiration and the area average soil moisture.

Wetzel, Peter J.↗
NASA NTRS1991

Measuring the effects of heterogeneity on distributed systems

Distributed computer systems in daily use are becoming more and more heterogeneous. Currently, much of the design and analysis studies of such systems assume homogeneity. This assumption of homogeneity has been mainly driven by the resulting simplicity in modeling and analysis. A simulation study is presented which investigated the effects of heterogeneity on scheduling algorithms for hard real time distributed systems. In contrast to previous results which indicate that random scheduling may be as good as a more complex scheduler, this algorithm is shown to be consistently better than a random scheduler. This conclusion is more prevalent at high workloads as well as at high levels of heterogeneity.

El-Toweissy, Mohamed↗
NASA NTRS2017

Scale Dependence of Cirrus Horizontal Heterogeneity Effects on TOA Measurements: MODIS Brightness Temperatures in the Thermal Infrared - Part I

This paper presents a study on the impact of cirrus cloud heterogeneities on MODIS simulated thermal infrared (TIR) brightness temperatures (BTs) at the top of the atmosphere (TOA) as a function of spatial resolution from 50 meters to 10 kilometers. A realistic 3-D (three-dimensional) cirrus field is generated by the 3DCLOUD model (average optical thickness of 1.4, cloudtop and base altitudes at 10 and 12 kilometers, respectively, consisting of aggregate column crystals of D (sub eff) equals 20 microns), and 3-D thermal infrared radiative transfer (RT) is simulated with the 3DMCPOL (3-D Monte Carlo Polarized) code. According to previous studies, differences between 3-D BT computed from a heterogenous pixel and 1-D (one-dimensional) RT computed from a homogeneous pixel are considered dependent at nadir on two effects: (i) the optical thickness horizontal heterogeneity leading to the plane-parallel homogeneous bias (PPHB); and the (ii) horizontal radiative transport (HRT) leading to the independent pixel approximation error (IPAE). A single but realistic cirrus case is simulated and, as expected, the PPHB mainly impacts the low-spatial resolution results (above approximately 250 meters), with averaged values of up to 5-7 K (thousand), while the IPAE mainly impacts the high-spatial resolution results (below approximately 250 meters) with average values of up to 1-2 K (thousand). A sensitivity study has been performed in order to extend these results to various cirrus optical thicknesses and heterogeneities by sampling the cirrus in several ranges of parameters. For four optical thickness classes and four optical heterogeneity classes, we have found that, for nadir observations, the spatial resolution at which the combination of PPHB and HRT effects is the smallest, falls between 100 and 250 meters. These spatial resolutions thus appear to be the best choice to retrieve cirrus optical properties with the smallest cloud heterogeneity-related total bias in the thermal infrared. For off-nadir observations, the average total effect is increased and the minimum is shifted to coarser spatial resolutions.

TOA↗
DOE OSTI2014

Time-Adjusted Internal Target Volume: A Novel Approach Focusing on Heterogeneity of Tumor Motion Based on 4-Dimensional Computed Tomography Imaging for Radiation Therapy Planning of Lung Cancer

Purpose: To consider nonuniform tumor motion within the internal target volume (ITV) by defining time-adjusted ITV (TTV), a volume designed to include heterogeneity of tumor existence on the basis of 4-dimensional computed tomography (4D-CT). Methods and Materials: We evaluated 30 lung cancer patients. Breath-hold CT (BH-CT) and free-breathing 4D-CT scans were acquired for each patient. The tumors were manually delineated using a lung CT window setting (window, 1600 HU; level, −300 HU). Tumor in BH-CT images was defined as gross tumor volume (GTV), and the sum of tumors in 4D-CT images was defined as ITV-4D. The TTV images were generated from the 4D-CT datasets, and the tumor existence probability within ITV-4D was calculated. We calculated the TTV{sub 80} value, which is the percentage of the volume with a tumor existence probability that exceeded 80% on ITV-4D. Several factors that affected the TTV{sub 80} value, such as the ITV-4D/GTV ratio or tumor centroid deviation, were evaluated. Results: Time-adjusted ITV images were acquired for all patients, and tumor respiratory motion heterogeneity was visualized. The median (range) ITV-4D/GTV ratio and median tumor centroid deviation were 1.6 (1.0-4.1) and 6.3 mm (0.1-30.3 mm), respectively. The median TTV{sub 80} value was 43.3% (2.9-98.7%). Strong correlations were observed between the TTV{sub 80} value and the ITV-4D/GTV ratio (R=−0.71) and tumor centroid deviation (R=−0.72). The TTV images revealed the tumor motion pattern features within ITV. Conclusions: The TTV images reflected nonuniform tumor motion, and they revealed the tumor motion pattern features, suggesting that the TTV concept may facilitate various aspects of radiation therapy planning of lung cancer while incorporating respiratory motion in the future.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2024

CHARM-SYCL & IRIS: A Tool Chain for Performance Portability on Extremely Heterogeneous Systems

Performance portability is becoming crucial as high-performance computing systems become increasingly heterogeneous. We have many options for CPUs and accelerators (e.g., GPUs) but also for non-Von Neumann architectures such as field-programmable gate arrays. This paper presents the CHARM-SYCL unified programming environment for multiple accelerator types as a performance-portable programming environment. It uses the IRIS library developed at Oak Ridge National Laboratory as the back end accelerator runtime. IRIS has a high-performance scheduler to distribute tasks across accelerators. This design allows us to run an application from the same source on multiple systems with multiple configurations. We provide three types of portability with CHARM-SYCL: Portable Workflow, Compiler and Runtime Portability, and Application and Performance Portability. We implement a Monte Carlo simulation benchmark code on the CHARM-SYCL execution environment and demonstrate that our programming environment can accommodate extremely heterogeneous systems.

Fujita, Norihisa↗
NASA NTRS2000

Information Power Grid: Distributed High-Performance Computing and Large-Scale Data Management for Science and Engineering

We use the term "Grid" to refer to distributed, high performance computing and data handling infrastructure that incorporates geographically and organizationally dispersed, heterogeneous resources that are persistent and supported. This infrastructure includes: (1) Tools for constructing collaborative, application oriented Problem Solving Environments / Frameworks (the primary user interfaces for Grids); (2) Programming environments, tools, and services providing various approaches for building applications that use aggregated computing and storage resources, and federated data sources; (3) Comprehensive and consistent set of location independent tools and services for accessing and managing dynamic collections of widely distributed resources: heterogeneous computing systems, storage systems, real-time data sources and instruments, human collaborators, and communications systems; (4) Operational infrastructure including management tools for distributed systems and distributed resources, user services, accounting and auditing, strong and location independent user authentication and authorization, and overall system security services The vision for NASA's Information Power Grid - a computing and data Grid - is that it will provide significant new capabilities to scientists and engineers by facilitating routine construction of information based problem solving environments / frameworks. Such Grids will knit together widely distributed computing, data, instrument, and human resources into just-in-time systems that can address complex and large-scale computing and data analysis problems. Examples of these problems include: (1) Coupled, multidisciplinary simulations too large for single systems (e.g., multi-component NPSS turbomachine simulation); (2) Use of widely distributed, federated data archives (e.g., simultaneous access to metrological, topological, aircraft performance, and flight path scheduling databases supporting a National Air Space Simulation systems}; (3) Coupling large-scale computing and data systems to scientific and engineering instruments (e.g., realtime interaction with experiments through real-time data analysis and interpretation presented to the experimentalist in ways that allow direct interaction with the experiment (instead of just with instrument control); (5) Highly interactive, augmented reality and virtual reality remote collaborations (e.g., Ames / Boeing Remote Help Desk providing field maintenance use of coupled video and NDI to a remote, on-line airframe structures expert who uses this data to index into detailed design databases, and returns 3D internal aircraft geometry to the field); (5) Single computational problems too large for any single system (e.g. the rotocraft reference calculation). Grids also have the potential to provide pools of resources that could be called on in extraordinary / rapid response situations (such as disaster response) because they can provide common interfaces and access mechanisms, standardized management, and uniform user authentication and authorization, for large collections of distributed resources (whether or not they normally function in concert). IPG development and deployment is addressing requirements obtained by analyzing a number of different application areas, in particular from the NASA Aero-Space Technology Enterprise. This analysis has focussed primarily on two types of users: the scientist / design engineer whose primary interest is problem solving (e.g. determining wing aerodynamic characteristics in many different operating environments), and whose primary interface to IPG will be through various sorts of problem solving frameworks. The second type of user is the tool designer: the computational scientists who convert physics and mathematics into code that can simulate the physical world. These are the two primary users of IPG, and they have rather different requirements. The results of the analysis of the needs of these two types of users provides a broad set of requirements that gives rise to a general set of required capabilities. The IPG project is intended to address all of these requirements. In some cases the required computing technology exists, and in some cases it must be researched and developed. The project is using available technology to provide a prototype set of capabilities in a persistent distributed computing testbed. Beyond this, there are required capabilities that are not immediately available, and whose development spans the range from near-term engineering development (one to two years) to much longer term R&D (three to six years). Additional information is contained in the original.

Johnston, William E.↗