SEARCH · Engineering Papers

Results for “Data parallelism”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 55 records · Page 3

All sources NASA NTRS DOE OSTI

DOE OSTI2020

Scalable Data-Intensive Geocomputation: A Design for Real-Time Continental Flood Inundation Mapping

The convergence of data-intensive and extreme-scale computing enables an integrated software and data ecosystem for scientific discovery. Developments in this realm will fuel transformative research in data-driven interdisciplinary domains. Geocomputation provides computing paradigms in Geographic Information Systems (GIS) for interactive computing of geographic data, processes, models, and maps. Because GIS is data-driven, the computational scalability of a geocomputation workflow is directly related to the scale of the GIS data layers, their resolution and extent, as well as the velocity of the geo-located data streams to be processed. Unique in high user interactivity and low end-to-end latency requirements, geocomputation applications will dramatically benefit from the convergence of high-end data analytics (HDA) and high-performance computing (HPC). The application level challenge, however, is to identify and eliminate computational bottlenecks that arise along a geocomputation workflow. Indeed, poor scalability at any of the workflow components is detrimental to the entire end-to-end pipeline. Here, we study a large geocomputation use case in flood inundation mapping that handles multiple national-scale geospatial datasets and targets low end-to-end latency. We discuss benefits and challenges for harnessing both HDA and HPC for data-intensive geospatial data processing and intensive numerical modeling of geographic processes. We propose an HDA+HPC geocomputation architecture design that couples HDA (e.g., Spark)-based spatial data handling and HPC-based parallel data modeling. Key techniques for coupling HDA and HPC to bridge the two different software stacks are reviewed and discussed.

Liu, Yan↗

DOE OSTI2020

Improving Performance of M-to-N Processing and Data Redistribution in In Transit Analysis and Visualization

In an in transit setting, a parallel data producer, such as a numerical simulation, runs on one set of ranks M, while a data consumer, such as a parallel visualization application, runs on a different set of ranks N. One of the central challenges in this in transit setting is to determine the mapping of data from the set of M producer ranks to the set of N consumer ranks. This is a challenging problem for several reasons, such as the producer and consumer codes potentially having different scaling characteristics and different data models. The resulting mapping from M to N ranks can have a significant impact on aggregate application performance. In this work, we present an approach for performing this M-to-N mapping in a way that has broad applicability across a diversity of data producer and consumer applications. We evaluate its design and performance with a study that runs at high concurrency on a modern HPC platform. By leveraging design characteristics, which facilitate an “intelligent” mapping from M-to-N, we observe significant performance gains are possible in terms of several different metrics, including time-to-solution and amount of data moved.

Loring, Burlen↗

DOE OSTI2020

Improving Performance of M-to-N Processing and Data Redistribution in In Transit Analysis and Visualization

In an in transit setting, a parallel data producer, such as a numerical simulation, runs on one set of ranks M, while a data consumer, such as a parallel visualization application, runs on a different set of ranks N: One of the central challenges in this in transit setting is to determine the mapping of data from the set of M producer ranks to the set of N consumer ranks. This is a challenging problem for several reasons, such as the producer and consumer codes potentially having different scaling characteristics and different data models. The resulting mapping from M to N ranks can have a significant impact on aggregate application performance. In this work, we present an approach for performing this M-to-N mapping in a way that has broad applicability across a diversity of data producer and consumer applications. We evaluate its design and performance with a study that runs at high concurrency on a modern HPC platform. By leveraging design characteristics, which facilitate an ''intelligent'' mapping from M-to-N, we observe significant performance gains are possible in terms of several different metrics, including time-to-solution and amount of data moved.

Loring, Burlen↗

DOE OSTI2021

Minimizing development costs for efficient many-core visualization using MCD3

Scientific visualization software increasingly needs to support many-core architectures. However, development time is a significant challenge due to the breadth and diversity of both visualization algorithms and architectures. With this work, we introduce a development environment for visualization algorithms on many-core devices that extends the traditional data-parallel primitive (DPP) approach with several existing constructs and an important new construct: meta-DPPs. We refer to our approach as MCD 3 — Meta-DPPs, Convenience routines, Data management, DPPs, and Devices. The twin goals of MCD 3 are to reduce developer time and to deliver efficient performance on many-core architectures, and our evaluation considers both of these goals. For development time, we study 57 algorithms implemented in the VTK-m software library and determine that MCD 3 leads to significant savings. For efficient performance, we survey ten studies looking at individual algorithms and determine that the MCD 3 hardware-agnostic approach leads to performance comparable to hardware-specific approaches: sometimes better, sometimes worse, and better in the aggregate. In total, we find that MCD 3 is an effective approach for scientific visualization libraries to support many-core architectures.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2024

Efficient distributed continual learning for steering experiments in real-time

Deep learning has emerged as a powerful method for extracting valuable information from large volumes of data. However, when new training data arrives continuously (i.e., is not fully available from the beginning), incremental training suffers from catastrophic forgetting (i.e., new patterns are reinforced at the expense of previously acquired knowledge). Training from scratch each time new training data becomes available would result in extremely long training times and massive data accumulation. Rehearsal-based continual learning has shown promise for addressing the catastrophic forgetting challenge, but research to date has not addressed performance and scalability. To fill this gap, we propose an approach based on a distributed rehearsal buffer that efficiently complements data-parallel training on multiple GPUs to achieve high accuracy, short runtime, and scalability. It leverages a set of buffers (local to each GPU) and uses several asynchronous techniques for updating these local buffers in an embarrassingly parallel fashion, all while handling the communication overheads necessary to augment input minibatches using unbiased, global sampling. We further propose a generalization of rehearsal buffers to support both classification and generative learning tasks, as well as more advanced rehearsal strategies (notably Dark Experience Replay, leveraging knowledge distillation). We illustrate this approach with a real-life HPC streaming application from the domain of ptychographic image reconstruction. Furthermore, we run extensive experiments on up to 128 GPUs of the ThetaGPU supercomputer to compare our approach with baselines representative of training-from-scratch (the upper bound in terms of accuracy) and incremental training (the lower bound). Results show that rehearsal-based continual learning achieves a top-5 validation accuracy close to the upper bound, while simultaneously exhibiting a runtime close to the lower bound.

Asynchronous data management↗

DOE OSTI2020

Calibration and modification of the constitutive model for the mechanical response of P91

The objective of this first milestone was to calibrate and if necessary, modify a constitutive model predicting the primary and secondary creep response of P91. The overall path to be followed consists of first calibrating this constitutive model against experimental data available from the literature (this milestone). In parallel, data acquired by indentation tests will be used to calibrate the model (this involves transforming the indentation data into equivalent uniaxial tests). An overall comparison of model parameters will allow assessing to what extent these rapid tests can be used to calibrate models predicting the overall mechanical response of the bulk. Given that COVID naturally hindered the gathering of new experimental data, the choice was made to ensure that LANL would have sufficient funds in year two to perform the bulk of the new calibration work. The following report details (i) how and which of the bulk experimental data was selected, (ii) constitutive model formulation, (iii) predictions of the constitutive model: effects of imposed stress, temperature and precipitate content, (iv) future work.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

DOE OSTI2023

2022 American Conference on Neutron Scattering (ACNS 2022)

The 11th American Conference on Neutron Scattering (ACNS 2022) will be held on June 5-9, 2022, in Boulder, CO. The Conference will provide essential information on the breadth and depth of current neutron-related research worldwide. Hosted by the Neutron Scattering Society of America, this year’s Conference will feature a combination of invited and contributed talks, poster sessions, and tutorials. Topics of the conference are: Advances in Neutron Facilities, Instrumentation and Software: Developments in sources, instrumentation, sample environments and control software. Hard Condensed Matter: Magnetism, correlated metals, quantum/topological materials, superconductors, ferroelectrics, multiferroics, glasses, and disorder phenomena. Submissions outlining examples of neutron scattering in industrial and engineering applications involving hard condensed matter systems are also encouraged. Soft Matter: Neutron studies of soft materials and related fields including in situ and in operando studies. Polymers, surfactants, emulsions, gels, nanoparticles, colloidal suspensions and more. Submissions of computational studies or applications of machine learning beneficial to neutron scattering experiments, as well as examples of neutron scattering in industrial and engineering applications are strongly encouraged. Biology, Biophysics and Biotechnology: Neutron studies of biological and biologically relevant systems. Proteins, bio membranes, biological assemblies, natural materials, nucleic acids, drug-delivery platforms and biomedical systems. Submissions of computational studies or applications of machine learning beneficial to biological neutron scattering experiments, as well as examples of neutron scattering in applied research involving biological systems, are strongly encouraged. Materials Chemistry and Energy: Neutron-based studies of functional materials and materials for energy applications. Examples include porous materials such as metal organic frameworks (MOFs), zeolites; phosphors; novel pigments; electrolytes; catalysts; ionic conductors/cathode materials; photovoltaic materials (hybrid perovskites); thermoelectrics; magnetocalorics/electrocalorics. Structural Materials and Engineering: Neutron scattering studies of materials and engineering processes including structural materials, concrete and metals, as well as engineering processes including combustion, corrosion, additive manufacturing, and others. Neutron Physics: Fundamental physical studies of the neutron and related areas. Emerging Applications in Neutron Scattering: Machine Learning and Data Science: Advances in computing power have contributed to rapidly evolving machine learning and data science fields that can be leveraged to the benefit of the neutron scattering community. The purpose of this session is to highlight recent advances in machine learning and data science and to serve as the foundation of a parallel data and computation track highlighting computation advances and applications in neutron scattering throughout the conference.

36 MATERIALS SCIENCE↗

DOE OSTI2021

Performance of Julia for High Energy Physics Analyses

We argue that the Julia programming language is a compelling alternative to currently more common implementations in Python and C++ for common data analysis workflows in high energy physics. We compare the speed of implementations of different workflows in Julia with those in Python and C++. Furthermore, our studies show that the Julia implementations are competitive for tasks that are dominated by computational load rather than data access. For work that is dominated by data access, we demonstrate an application with concurrent file reading and parallel data processing.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2021

System and method of storing and analyzing information

A system and method of storing and analyzing information is disclosed. The system includes a compiler layer to convert user queries to data parallel executable code. The system further includes a library of multithreaded algorithms, processes, and data structures. The system also includes a multithreaded runtime library for implementing compiled code at runtime. The executable code is dynamically loaded on computing elements and contains calls to the library of multithreaded algorithms, processes, and data structures and the multithreaded runtime library.

Feo, John T.↗

DOE OSTI2022

Exploring the Use of Novel Spatial Accelerators in Scientific Applications

Driven by the need to find alternative accelerators which can viably replace GPUs in next-generation Supercomputing systems, this paper proposes a methodology to enable agile application/hardware co-design. The application-first methodology provides the ability to come up with design of accelerators while working with real-world workloads, available accelerators, and system software. The iterative design process targets a set of kernels in a workload for performance estimates that can prune the design space for later phases of detailed architectural evaluations. To this effect, in this paper, a novel data-parallel device model is introduced that simulates the latency of performance-sensitive operations in an accelerator including data transfers and kernel computation using multi-core CPUs. The use of off-the-shelf simulators, such as pre-RTL simulator Aladdin or multiple tools available for exploring the design of deep neural network accelerators (e.g., Timeloop) is demonstrated for evaluation of various accelerator designs using applications with realistic inputs. Examples of multiple device configurations that are instantiable in a system are explored to evaluate the performance benefit of deploying novel accelerators. The proposed device is integrated with a programming model and system software to potentially explore the impacts of high-level programming languages/compilers and low-level effects such as task scheduling on multiple accelerators. We analyze our methodology for a set of applications that represent high-performance computing (HPC) and graph analytics. The applications include a computational chemistry kernel realized using tensor contractions, triangle counting, GraphSAGE and Breadth-first Search. These applications include kernels such as dense matrix-dense matrix multiplication, sparse matrix-spare matrix multiplication, and sparse matrix-dense vector multiplication. Our results indicate potential performance benefits and insights for system design by including accelerators that realize these kernels along-side general purpose accelerators.

AI, codesign, Accelerated Computing, Modeling and ↗

DOE OSTI2026

FLYING SERVING: On-the-Fly Parallelism Switching for Large Language Model Serving

Production LLM serving must simultaneously deliver high throughput, low latency, and sufficient context capacity under non-stationary traffic and mixed request requirements. Data parallelism (DP) maximizes throughput by running independent replicas, while tensor parallelism (TP) reduces per-request latency and pools memory for long-context inference. However, existing serving stacks typically commit to a static parallelism configuration at deployment; adapting to bursts, priorities, or long-context requests is often disruptive and slow. We present Flying Serving, a vLLM-based system that enables online DP-TP switching without restarting engine workers. Flying Serving makes reconfiguration practical by virtualizing the state that would otherwise force data movement: (i) a zero-copy Model Weights Manager that exposes TP shard views on demand, (ii) a KV Cache Adaptor that preserves request KV state across DP/TP layouts, (iii) an eagerly initialized Communicator Pool to amortize collective setup, and (iv) a deadlock-free scheduler that coordinates safe transitions under execution skew. Across three popular LLMs and realistic serving scenarios, Flying Serving improves performance by up to 4.79 × under high load and 3.47 × under low load while supporting latency- and memory-driven requests.

Gao, Shouwei [ORNL]↗

DOE OSTI2022

Metaplastic and energy-efficient biocompatible graphene artificial synaptic transistors for enhanced accuracy neuromorphic computing

CMOS-based computing systems that employ the von Neumann architecture are relatively limited when it comes to parallel data storage and processing. In contrast, the human brain is a living computational signal processing unit that operates with extreme parallelism and energy efficiency. Although numerous neuromorphic electronic devices have emerged in the last decade, most of them are rigid or contain materials that are toxic to biological systems. In this work, we report on biocompatible bilayer graphene-based artificial synaptic transistors (BLAST) capable of mimicking synaptic behavior. The BLAST devices leverage a dry ion-selective membrane, enabling long-term potentiation, with ~50 aJ/µm 2 switching energy efficiency, at least an order of magnitude lower than previous reports on two-dimensional material-based artificial synapses. The devices show unique metaplasticity, a useful feature for generalizable deep neural networks, and we demonstrate that metaplastic BLASTs outperform ideal linear synapses in classic image classification tasks. With switching energy well below the 1 fJ energy estimated per biological synapse, the proposed devices are powerful candidates for bio-interfaced online learning, bridging the gap between artificial and biological neural networks.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2023

Massively scalable Kerr comb-driven silicon photonic link

Abstract The growth of computing needs for artificial intelligence and machine learning is critically challenging data communications in today’s data-centre systems. Data movement, dominated by energy costs and limited ‘chip-escape’ bandwidth densities, is perhaps the singular factor determining the scalability of future systems. Using light to send information between compute nodes in such systems can dramatically increase the available bandwidth while simultaneously decreasing energy consumption. Through wavelength-division multiplexing with chip-based microresonator Kerr frequency combs, independent information channels can be encoded onto many distinct colours of light in the same optical fibre for massively parallel data transmission with low energy. Although previous high-bandwidth demonstrations have relied on benchtop equipment for filtering and modulating Kerr comb wavelength channels, data-centre interconnects require a compact on-chip form factor for these operations. Here we demonstrate a massively scalable chip-based silicon photonic data link using a Kerr comb source enabled by a new link architecture and experimentally show aggregate single-fibre data transmission of 512 Gb s −1 across 32 independent wavelength channels. The demonstrated architecture is fundamentally scalable to hundreds of wavelength channels, enabling massively parallel terabit-scale optical interconnects for future green hyperscale data centres.

Rizzo, Anthony (ORCID:000000034752797X)↗

DOE OSTI2024

Scheduling and Performance of Asynchronous Tasks in Fortran 2018 with FEATS

Most parallel scientific programs contain compiler directives (pragmas) such as those from OpenMP (Hermanns in Parallel programming in Fortran 95 using openMP, 2002. School of Aeronautical Engineering, Universidad Politécnica de Madrid, España, 2011), explicit calls to runtime library procedures such as those implementing the Message Passing Interface (MPI) (in A message-passing interface standard version 4.0, 2021. https://www.mpi-forum.org/docs/mpi-4.0/mpi40-report.pdf), or compiler-specific language extensions such as those provided by CUDA (Ruetsch and Fatica in CUDA Fortran for scientists and engineers: best practices for efficient CUDA Fortran programming, Elsevier, 2013). By contrast, the recent Fortran standards empower developers to express parallel algorithms without directly referencing lower-level parallel programming models (Numrich in Parallel programming with co-arrays, CRC Press, 2018, and Curcic in Modern Fortran: building efficient parallel applications, Manning Publications, 2020). Fortran’s parallel features place the language within the Partitioned Global Address Space (PGAS) class of programming models. When writing programs that exploit data parallelism, application developers often find it straightforward to develop custom parallel algorithms. Problems involving complex, heterogeneous, staged calculations, however, pose much greater challenges. Such applications require careful coordination of tasks in a manner that respects dependencies prescribed by a directed acyclic graph. When rolling one’s own solution proves difficult, extending a customizable framework becomes attractive. Further, the paper presents the design, implementation, and use of the Framework for Extensible Asynchronous Task Scheduling (FEATS), which we believe to be the first task scheduling tool written in modern Fortran. We describe the benefits and compromises associated with choosing Fortran as the implementation language, and we propose ways in which future Fortran standards can best support the use case in this paper.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2023

Framework for Extensible, Asynchronous Task Scheduling (FEATS) in Fortran

Most parallel scientific programs contain compiler directives (pragmas) such as those from OpenMP, explicit calls to runtime library procedures such as those implementing the Message Passing Interface (MPI), or compiler-specific language extensions such as those provided by CUDA. By contrast, the recent Fortran standards empower developers to express parallel algorithms without directly referencing lower-level parallel programming models. Fortran’s parallel features place the language within the Partitioned Global Address Space (PGAS) class of programming models. When writing programs that exploit data-parallelism, application developers often find it straightforward to develop custom parallel algorithms. Problems involving complex, heterogeneous, staged calculations, however, pose much greater challenges. Such applications require careful coordination of tasks in a manner that respects dependencies prescribed by a directed acyclic graph. When rolling one’s own solution proves difficult, extending a customizable framework becomes attractive. The paper presents the design, implementation, and use of the Framework for Extensible Asynchronous Task Scheduling (FEATS), which we believe to be the first task-scheduling tool written in modern Fortran. We describe the benefits and compromises associated with choosing Fortran as the implementation language, and we propose ways in which future Fortran standards can best support the use case in this paper.

Richardson, Brad↗

DOE OSTI2025

Scale-up Unlearnable Examples Learning with High-performance Computing

Recent advancements in AI models, like ChatGPT, are structured to retain user interactions, which could inadvertently include sensitive healthcare data. In the healthcare field, particularly when radiologists use AI-driven diagnostic tools hosted on online platforms, there is a risk that medical imaging data may be repurposed for future AI training without explicit consent, spotlighting critical privacy and intellectual property concerns around healthcare data usage. Addressing these privacy challenges, a novel approach known as Unlearnable Examples (UEs) has been introduced, aiming to make data unlearnable to deep learning models. A prominent method within this area, called Unlearnable Clustering (UC), has shown improved UE performance with larger batch sizes but was previously limited by computational resources (e.g., a single workstation). To push the boundaries of UE performance with theoretically unlimited resources, we scaled up UC learning across various datasets using Distributed Data Parallel (DDP) training on the Summit supercomputer. Our goal was to examine UE efficacy at high-performance computing (HPC) levels to prevent unauthorized learning and enhance data security, particularly exploring the impact of batch size on UE’s unlearnability. Utilizing the robust computational capabilities of the Summit, extensive experiments were conducted on diverse datasets such as Pets, MedMNist, Flowers, and Flowers102. Our findings reveal that both overly large and overly small batch sizes can lead to performance instability and affect accuracy. However, the relationship between batch size and unlearnability varied across datasets, highlighting the necessity for tailored batch size strategies to achieve optimal data protection. The use of Summit’s high-performance GPUs, along with the efficiency of the DDP framework, facilitated rapid updates of model parameters and consistent training across nodes. Our results underscore the critical role of selecting appropriate batch sizes based on the specific characteristics of each dataset to prevent learning and ensure data security in deep learning applications. The source code is publicly available at https: // github. com/ hrlblab/ UE_ HPC .

Zhu, Yanfan [Vanderbilt University, Nashville, TN,↗

DOE OSTI2022

The Astronomy Commons Platform: A Deployable Cloud-based Analysis Platform for Astronomy

Abstract We present a scalable, cloud-based science platform solution designed to enable next-to-the-data analyses of terabyte-scale astronomical tabular data sets. The presented platform is built on Amazon Web Services (over Kubernetes and S3 abstraction layers), utilizes Apache Spark and the Astronomy eXtensions for Spark for parallel data analysis and manipulation, and provides the familiar JupyterHub web-accessible front end for user access. We outline the architecture of the analysis platform, provide implementation details and rationale for (and against) technology choices, verify scalability through strong and weak scaling tests, and demonstrate usability through an example science analysis of data from the Zwicky Transient Facility’s 1Bn+ light-curve catalog. Furthermore, we show how this system enables an end user to iteratively build analyses (in Python) that transparently scale processing with no need for end-user interaction. The system is designed to be deployable by astronomers with moderate cloud engineering knowledge, or (ideally) IT groups. Over the past 3 yr, it has been utilized to build science platforms for the DiRAC Institute, the ZTF partnership, the LSST Solar System Science Collaboration, and the LSST Interdisciplinary Network for Collaboration and Computing, as well as for numerous short-term events (with over 100 simultaneous users). In a live demo instance, the deployment scripts, source code, and cost calculators are accessible. 4 4 http://hub.astronomycommons.org/

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2024

Distributed-Memory Sparse Deep Neural Network Inference Using Global Arrays

Partitioned Global Address Space (PGAS) models exhibit tremendous promise in developing efficient and productive distributed-memory parallel applications. They have been used extensively in scientific computations due to conveniently offering a ``shared-memory''-like model and convenient interfaces that separate communication with synchronization. Traditionally, PGAS communication models have been applied to dense/contiguously distributed data, but most modern applications depict varied levels of sparsity. Existing PGAS models require certain adaptations to support distributed sparse computations, since associated computations often require matrix arithmetic, in addition to data movement. The Global Arrays toolkit from Pacific Northwest National Laboratory (PNNL) is one of the earliest PGAS models to combine one-sided data communication and distributed matrix operations and is still used in the popular NWChem quantum chemistry suite. Recently, we have expanded the Global Arrays toolkit to support common sparse operations, like sparse matrix-dense matrix multiplies (SpMM), sparse matrix-sparse matrix multiplication (SpGEMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM). As it turns out, these operations are the bedrock of sparse Deep Learning (DL); sparse deep neural networks and Graph Neural Networks (GNNs) have gained increasing attention recently in achieving speedups on training and inference with reduced memory footprints. Unlike scientific applications in High Performance Computing (HPC), modern (distributed-memory capable) DL toolkits often rely on non-standardized and closed-source vendor software optimizations, creating challenges in software-hardware co-design at scale. Our goal is to support a variety of distributed-memory sparse matrix operations and helper functions in the newly created Sparse Global Arrays (SGA), such that it is possible to build portable and productive Machine Learning scenarios for algorithm/software and hardware codesign purposes. Contemporary data-parallel schemes for training/inference are undergoing a major overhaul since model replication limits scalability and causes resource inefficiencies. As such, we have adopted tensor parallelism in decomposing the model and inputs, to mitigate memory issues. Current implementation is built on top of MPI and uses CPUs to maximize the portability across the platforms.

Distributed computing, machine learning↗