Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Parallel application frameworks”

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 343 records · Page 19

Implementing High Performance Remote Method Invocation in CCA

We report our effort in engineering a high performance remote method invocation (RMI) mechanism for the Common Component Architecture (CCA). This mechanism provides a highly efficient and easy-to-use mechanism for distributed computing in CCA, enabling CCA applications to effectively leverage parallel systems to accelerate computations. This work is built on the previous work of Babel RMI. Babel is a high performance language interoperability tool that is used in CCA for scientific application writers to share, reuse, and compose applications from software components written in different programming languages. Babel provides a transparent and flexible RMI framework for distributed computing. However, the existing Babel RMI implementation is built on top of TCP and does not provide the level of performance required to distribute fine-grained tasks. We observed that the main reason the TCP based RMI does not perform well is because it does not utilize the high performance interconnect hardware on a cluster efficiently. We have implemented a high performance RMI protocol, HPCRMI. HPCRMI achieves low latency by building on top of a low-level portable communication library, Aggregated Remote Message Copy Interface (ARMCI), and minimizing communication for each RMI call. Our design allows a RMI operation to be completed by only two RDMA operations. We also aggressively optimize our system to reduce copying. In this paper, we discuss the design and our experimental evaluation of this protocol. Our experimental results show that our protocol can improve RMI performance by an order of magnitude.

99 GENERAL AND MISCELLANEOUS↗

GLUE Code: A framework handling communication and interfaces between scales

Many scientific applications are inherently multiscale in nature. Such complex physical phenomena often require simultaneous execution and coordination of simulations spanning multiple time and length scales. This is possible by combining expensive small-scale simulations (such as molecular dynamics simulations) with larger scale simulations (such continuum limit/hydro solvers) to allow for considerably larger systems using task and data parallelism. However, the granularity of the tasks can be very large and often leads to load imbalance. Traditionally, we use approximations to streamline the computation of the more costly interactions and this introduces trade-offs between simulation cost and accuracy. In recent years, the available computational power and the advances in machine learning have made computing these scale-bridging interactions and multiscale simulations more feasible. One driving application has been plasma modeling in inertial confinement fusion (ICF), which is fundamentally multiscale in nature. This requires deep understanding of how to extrapolate microscopic information into macroscopically relevant scales. For example, in ICF one needs an accurate understanding of the connection between experimental observables and the underlying microphysics. The properties of the larger scales are often affected by the microscale behavior incorporated usually into the equations of state and ionic and electronic transport coefficients (Liboff, 1959; Rinderknecht et al., 2014; Rosenberg et al., 2015; Ross et al., 2017). Instead of incorporating this information using reliable molecular dynamics (MD) simulations, one often needs to use theoretical models, due to the inability of MD to reach engineering scales (Glosli et al., 2007; Marinak et al., 1998). One approach to resolve this issue is by coupling two MD simulations of different scales via force interpolation, e.g., the AdResS method (Krekeler et al., 2018; Nagarajan et al., 2013). Another approach, which we will pursue in the scope of this work, is by enabling scale bridging between MD simulations and meso/macro-scale models through the development and support of application programming interfaces that these different applications can interact through.

54 ENVIRONMENTAL SCIENCES↗

In Situ Indexing and Query Processing of AMR Data

We have developed the first, holistic, Adaptive Mesh Refinement (AMR) capable, in situ query framework to ensure the tractability of the current- and next-generation AMR simulation analysis on the path to extreme scale. We have addressed key issues pertaining to in situ analysis, memory hierarchy, generalized parallel indexing, index-compression, and real-time access pattern detection. We have published the following research papers in high profile peer reviewed computer science conferences: (a) Parallel In Situ Detection of Connected Components in Adaptive Mesh Refinement Data; (b) AMR-aware In Situ Indexing and Scalable Querying; (c) Pattern-driven dynamic data layout optimization; (d) In situ Storage Layout Optimization for AMR Spatio-temporal Read Accesses; (e) AMRZone: A Runtime AMR Data Sharing Framework for Scientific Applications; (f) Exploring Memory Hierarchy and Network Topology for Runtime AMR Data Sharing Across Scientific Applications.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

AGEING PROCEDURES ON LITHIUM BATTERIES IN AN INTERNATIONAL COLLABORATION CONTEXT

The widespread introduction of electrically-propelled vehicles is currently part of many political strategies and introduction plans. These new vehicles, ranging from limited (mild) hybrid to plug-in hybrid to fully-battery powered, will rely on a new class of advanced storage batteries, such as those based on lithium, to meet different technical and economical targets. The testing of these batteries to determine the performance and life in the various applications is a time-consuming and costly process that is not yet well developed. There are many examples of parallel testing activities that are poorly coordinated, for example, those in Europe, Japan and the US. These costs and efforts may be better leveraged through international collaboration, such as that possible within the framework of the International Energy Agency. Here, a new effort is under development that will establish standardized, accelerated testing procedures and will allow battery testing organizations to cooperate in the analysis of the resulting data. This paper reviews the present state-of-the-art in accelerated life testing in Europe, Japan and the US. The existing test procedures will be collected, compared and analyzed with the goal of international collaboration.

25 ENERGY STORAGE↗

Reactor Pressure Vessel Integrity Assessments with the Grizzly Aging Simulation Code

Grizzly is a simulation tool being developed at Idaho National Laboratory (INL) as part of the US Department of Energy’s Light Water Reactor Sustainability program to provide improved safety assessments of systems, components, and structures in nuclear power plants subjected to age-related degradation. Its goal is to provide an improved scientific basis for decisions surrounding license renewal, which would permit operation of commercial nuclear power plants beyond 60 years. Grizzly is based on INL’s MOOSE framework, which enables multiphysics simulations in a parallel computing environment. It will address a wide variety of aging issues in nuclear power plant systems, components, and structures, modelling both the aging processes and the ability of age-degraded components to perform safely. The reactor pressure vessel (RPV) was chosen as the initial application for Grizzly. Grizzly solves tightly coupled equations of heat conduction and solid mechanics to simulate the global response of the RPV to accident conditions, and uses submodels to represent regions with pre-existing flaws. Domain integrals are used to calculate stress intensity factors on those flaws. A physically based empirical model is used to evaluate material embrittlement, and is used to evaluate whether crack growth would occur. Grizzly can represent the RPV in 2D or 3D, allowing it to evaluate effects that require higher dimensionality models to capture. Work is underway to use lower length scale models of material evolution to inform engineering models of embrittlement. This paper demonstrates an application of Grizzly to RPV failure assessment, and summarizes on-going work.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

DaCe - Data Centric Parallel Programming

DaCe is a parallel programming framework that takes code in Python/NumPy and other programming languages, and maps it to high-performance CPU, GPU, and FPGA programs, which can be optimized to achieve state-of-the-art. Internally, DaCe uses the Stateful DataFlow multiGraph (SDFG) data-centric intermediate representation: A transformable, interactive representation of code based on data movement. Since the input code and the SDFG are separate, it is possible to optimize a program without changing its source, so that it stays readable. On the other hand, transformations are customizable and user-extensible, so they can be written once and reused in many applications. With data-centric parallel programming, we enable direct knowledge transfer of performance optimization, regardless of the application or the target processor.

Ben-Nun, Tal↗

A Full-Stack Exploration of Language-Based Parallelism in Fortran 2023

This poster explores native parallel features in Fortran 2023 through the lens of supporting applications with libraries, compilers, and parallel runtimes. The language revision informally named Fortran 2008 introduced parallelism in the form of Single Program Multiple Data (SPMD) execution with two broad feature sets: (1) loop-level parallelism via do concurrent and (2) a Partitioned Global Address Space (PGAS) comprised of distributed “coarray” data structures. Fortran’s native parallelism has demonstrated high performance [1] and reduced the burden of inserting what sometimes amounts to more directives than code. Several compilers support both feature sets, typically by translating do concurrent into serial do loops annotated by parallel directives and by translating SPMD/PGAS features into direct calls to a communication library. Our research focuses primarily on two questions: (1) can the compiler’s parallel runtime library be developed in the language being compiled (Fortran) and (2) can we define an interface to the runtime that liberates compilers from being hardwired to one runtime and vice versa. We are answering these questions by developing the Parallel Runtime Interface for Fortran (PRIF) [2] and the Co-Array Fortran Framework of Efficient Interfaces to Network Environments (Caffeine) [3]. Caffeine is initially targeting adoption by LLVM Flang, a new open-source Fortran compiler developed by a broad community in industry, academia, and government labs. We are also exploring the use of these features in Inference-Engine, a deep learning library designed to facilitate neural network training and inference for high-performance computing applications written in modern Fortran.

Rasmussen, Katherine↗

Integration of a Decentralized Linear-Quadratic-Gaussian Control into GSFC's Universal 3-D Autonomous Formation Flying Algorithm

A decentralized control is investigated for applicability to the autonomous formation flying control algorithm developed by GSFC for the New Millenium Program Earth Observer-1 (EO-1) mission. This decentralized framework has the following characteristics: The approach is non-hierarchical, and coordination by a central supervisor is not required; Detected failures degrade the system performance gracefully; Each node in the decentralized network processes only its own measurement data, in parallel with the other nodes; Although the total computational burden over the entire network is greater than it would be for a single, centralized controller, fewer computations are required locally at each node; Requirements for data transmission between nodes are limited to only the dimension of the control vector, at the cost of maintaining a local additional data vector. The data vector compresses all past measurement history from all the nodes into a single vector of the dimension of the state; and The approach is optimal with respect to standard cost functions. The current approach is valid for linear time-invariant systems only. Similar to the GSFC formation flying algorithm, the extension to linear LQG time-varying systems requires that each node propagate its filter covariance forward (navigation) and controller Riccati matrix backward (guidance) at each time step. Extension of the GSFC algorithm to non-linear systems can also be accomplished via linearization about a reference trajectory in the standard fashion, or linearization about the current state estimate as with the extended Kalman filter. To investigate the feasibility of the decentralized integration with the GSFC algorithm, an existing centralized LQG design for a single spacecraft orbit control problem is adapted to the decentralized framework while using the GSFC algorithm's state transition matrices and framework. The existing GSFC design uses both reference trajectories of each spacecraft in formation and by appropriate choice of coordinates and simplified measurement modeling is formulated as a linear time-invariant system. Results for improvements to the GSFC algorithm and a multiple satellite formation will be addressed. The goal of this investigation is to progressively relax the assumptions that result in linear time-invariance, ultimately to the point of linearization of the non-linear dynamics about the current state estimate as in the extended Kalman filter. An assessment will then be made about the feasibility of the decentralized approach to the realistic formation flying application of the EO-1/Landsat 7 formation flying experiment.

Folta, David C.↗

Scalable parallel methods for monolithic coupling in fluid-structure interaction with application to blood flow modeling

We introduce and study numerically a scalable parallel finite element solver for the simulation of blood flow in compliant arteries. The incompressible Navier-Stokes equations are used to model the fluid and coupled to an incompressible linear elastic model for the blood vessel walls. Our method features an unstructured dynamic mesh capable of modeling complicated geometries, an arbitrary Lagrangian-Eulerian framework that allows for large displacements of the moving fluid domain, monolithic coupling between the fluid and structure equations, and fully implicit time discretization. Simulations based on blood vessel geometries derived from patient-specific clinical data are performed on large supercomputers using scalable Newton-Krylov algorithms preconditioned with an overlapping restricted additive Schwarz method that preconditions the entire fluid-structure system together. The algorithm is shown to be robust and scalable for a variety of physical parameters, scaling to hundreds of processors and millions of unknowns.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

High Energy Electromagnetic Particle Transportation on the GPU

We present massively parallel high energy electromagnetic particle transportation through a finely segmented detector on a Graphics Processing Unit (GPU). Simulating events of energetic particle decay in a general-purpose high energy physics (HEP) detector requires intensive computing resources, due to the complexity of the geometry as well as physics processes applied to particles copiously produced by primary collisions and secondary interactions. The recent advent of hardware architectures of many-core or accelerated processors provides the variety of concurrent programming models applicable not only for the high performance parallel computing, but also for the conventional computing intensive application such as the HEP detector simulation. The components of our prototype are a transportation process under a non-uniform magnetic field, geometry navigation with a set of solid shapes and materials, electromagnetic physics processes for electrons and photons, and an interface to a framework that dispatches bundles of tracks in a highly vectorized manner optimizing for spatial locality and throughput. Core algorithms and methods are excerpted from the Geant4 toolkit, and are modified and optimized for the GPU application. Program kernels written in C/C++ are designed to be compatible with CUDA and OpenCL and with the aim to be generic enough for easy porting to future programming models and hardware architectures. To improve throughput by overlapping data transfers with kernel execution, multiple CUDA streams are used. Issues with floating point accuracy, random numbers generation, data structure, kernel divergences and register spills are also considered. Performance evaluation for the relative speedup compared to the corresponding sequential execution on CPU is presented as well.

Canal, P.↗

Modeling Chemical Reactions in Classical Molecular Dynamics Simulations

An algorithm capable of incorporating multi-step reaction mechanisms into atomistic molecular dynamics (MD) simulations using traditional fixed valence force fields is proposed and implemented within the framework of LAMMPS (Large-scale Atomic Molecular Massively Parallel Simulator). This extension, referred to as fix bond/react, enables bonding topology modifications during a running MD simulation using pre- and post-reaction bonding templates to carry out a pre-specified reaction. Candidate reactants are first identified by interatomic separation, followed by the application of a generalized topology matching algorithm to confirm they match the pre-reaction template. This is followed by a topology conversion to match the post-reaction template and a dynamic relaxation to minimize high energy configurations. Two case studies, the condensation polymerization of nylon 6,6 and the formation of a highly-crosslinked epoxy, are simulated to demonstrate the robustness, stability, and speed of the algorithm. Improvements which could increase its utility are discussed.

Jacob R Gissinger↗

An Approach to Solving Enclosure Radiation Problems in A Multi-Physics Context

Thermal protection system analysis of complex features or damage sites can sometimes require modeling of high temperature enclosures. Implementing efficient and accurate view-factor algorithms required to model such problems is complex. The current work leverages the Non-equilibrium Radiation (NERO) software, which solves the radiation transport equation in a finite-volume scheme, to alleviating challenges often faced with view-factor calculations. By assuming heat transfer occurs only between grey bodies and that the medium is non-participating, computational cost of the method is significantly reduced. The enclosure physics are modeled through emitting and reflecting boundary conditions in NERO. The emitted radiative flux is dependent on the wall temperature which is a solution to the material response, obtained from Icarus, in this context. The Ares framework manages the time-advancement and exchange of the necessary data between the solvers. The surface energy balance is modified to account for the enclosure terms within the material response boundary condition. The methodology was verified against analytical solutions including radiating parallel plates, a hollow cylinder (shown in Fig. 1), and a hemisphere. Application of the methodology to inform the design of components of the Dragonfly system will be shown.

Ablation↗

In‐memory integration of existing software components for parallel adaptive unstructured mesh workflows

Summary Reliable mesh‐based simulations are needed to solve complex engineering problems. Mesh adaptivity can increase reliability by reducing discretization errors but requires multiple software components to exchange information. Often, components exchange information by reading and writing a common file format. This file‐based approach becomes a problem on massively parallel computers where filesystem bandwidth is a critical performance bottleneck. Our approach using data streams and component interfaces avoids the filesystem bottleneck. In this paper, we present these techniques and their use for coupling mesh adaptivity to the PHASTA computational fluid dynamics solver, the Albany multi‐physics framework, and the Omega3P linear accelerator frequency analysis applications. Performance results are reported on up to 16,384 cores of an Intel Knights Landing‐based system.

Smith, Cameron W.↗

Neural Network Molecular Dynamics at Scale

Neural network molecular dynamics (NNMD) simulations could revolutionize atomistic modeling of materials with quantum-mechanical accuracy at a fraction of computational cost. However, popular NNMD frameworks are generally implemented for a single computing node, and conventional energy-based NN models still suffer from large time-to-solution (T2S), prohibiting the application of NNMD to challenging materials simulations encompassing large spatiotemporal scales. Consequently, no leadership-scale NNMD simulation has thus far been reported. Here, we present a scalable parallel NNMD software (RXMD-NN) based on our scalable reactive molecular dynamics simulation engine named RXMD. RXMD-NN has achieved high scalability up to 786,432 IBM BlueGene/Q cores involving 1.7 billion atoms. Furthermore, we have achieved 4.6-fold reduction of T2S by using a novel network that directly predicts atomic forces from feature vectors. Reduced T2S has for the first time allowed the study of large-scale off-stoichiometry effect in a widely used phase change material, Ge 2 Se 2 Te 5 , thereby resolving its "firstsharp diffraction peak mystery".

fist sharp diffraction peak↗

Scalable Deep Learning via I/O Analysis and Optimization

Scalable deep neural network training has been gaining prominence because of the increasing importance of deep learning in a multitude of scientific and commercial domains. Consequently, a number of researchers have investigated techniques to optimize deep learning systems. Much of the prior work has focused on runtime and algorithmic enhancements to optimize the computation and communication. Despite these enhancements, however, deep learning systems still suffer from scalability limitations, particularly with respect to data I/O. This situation is especially true for training models where the computation can be effectively parallelized, leaving I/O as the major bottleneck. In fact, our analysis shows that I/O can take up to 90% of the total training time. Thus, in this article, we first analyze LMDB, the most widely used I/O subsystem of deep learning frameworks, to understand the causes of this I/O inefficiency. Based on our analysis, we propose LMDBIO—an optimized I/O plugin for scalable deep learning. LMDBIO includes six novel optimizations that together address the various shortcomings in existing I/O for deep learning. Lastly, our experimental results show that LMDBIO significantly outperforms LMDB in all cases and improves overall application performance by up to 65-fold on a 9,216-core system.

97 MATHEMATICS AND COMPUTING↗

Taming parallel I/O complexity with auto-tuning

We present an auto-tuning system for optimizing I/O performance of HDF5 applications and demonstrate its value across platforms, applications, and at scale. The system uses a genetic algorithm to search a large space of tunable parameters and to identify effective settings at all layers of the parallel I/O stack. The parameter settings are applied transparently by the auto-tuning system via dynamically intercepted HDF5 calls. To validate our auto-tuning system, we applied it to three I/O benchmarks (VPIC, VORPAL, and GCRM) that replicate the I/O activity of their respective applications. We tested the system with different weak-scaling configurations (128, 2048, and 4096 CPU cores) that generate 30 GB to 1 TB of data, and executed these configurations on diverse HPC platforms (Cray XE6, IBM BG/P, and Dell Cluster). In all cases, the auto-tuning framework identified tunable parameters that substantially improved write performance over default system settings. In conclusion, we consistently demonstrate I/O write speedups between 2x and 100x for test configurations.

97 MATHEMATICS AND COMPUTING↗

NORSAR Final Scientific Report Adaptive Waveform Correlation Detectors for Arrays: Algorithms for Autonomous Calibration

Correlation detection is a relatively new approach in seismology that offers significant advantages in increased sensitivity and event screening over standard energy detection algorithms. The basic concept is that a representative event waveform is used as a template (i.e. matched filter) that is correlated against a continuous, possibly multichannel, data stream to detect new occurrences of that same signal. These algorithms are therefore effective at detecting repeating events, such as explosions and aftershocks at a specific location. This final report summarizes the results of a three-year cooperative project undertaken by NORSAR and Lawrence Livermore National Laboratory. The overall objective has been to develop and test a new advanced, automatic approach to seismic detection using waveform correlation. The principal goal is to develop an adaptive processing algorithm. By this we mean that the detector is initiated using a basic set of reference ('master') events to be used in the correlation process, and then an automatic algorithm is applied successively to provide improved performance by extending the set of master events selectively and strategically. These additional master events are generated by an independent, conventional detection system. A periodic analyst review will then be applied to verify the performance and, if necessary, adjust and consolidate the master event set. A primary focus of this project has been the application of waveform correlation techniques to seismic arrays. The basic procedure is to perform correlation on the individual channels, and then stack the correlation traces using zero-delay beam forming. Array methods such as frequency-wavenumber analysis can be applied to this set of correlation traces to help guarantee the validity of detections and lower the detection threshold. In principle, the deployment of correlation detectors against seismically active regions could involve very large numbers of very specific detectors. To meet this challenge, we have examined two strategies: (1) use of subspace detectors, a multi-dimensional extension of correlators, which allow representation and detection of signals exhibiting some degree of variation; and (2) autonomous calibration of many subspace and correlation detectors in an adaptive detection framework, subject to analyst review. Because correlation detectors are relatively new to seismology, a significant amount of research on how to tune these detectors has been needed to address later calibration efforts that will arise as they are adopted for operational use. We have approached these challenges by carrying out a number of case studies, encompassing various monitoring scenarios such as earthquake aftershock sequences and swarms, recurring mining explosions, other types of explosions, and rockbursts. We have studied several different geographical regions (the European Arctic, Central Asia, and the western United States). We have drawn on available Ground Truth data in assessing the results of the various processing schemes. In all cases, we have benefited from the high-quality seismic arrays or networks available in these regions, and we have thus been able to evaluate the performance of array-based correlation processing under a variety of conditions. The main results of the project are summarized as follows: (1) Array-based waveform correlation has been demonstrated to lower significantly detection thresholds in comparison with standard single-channel waveform correlation. (2) Frequency-wavenumber analysis of the correlation traces on a small-aperture array provides an effective method for screening out a certain category of false alarms, and can therefore be used to improve detector sensitivity by lowering the threshold for automatic array detection. (3) We have developed and tested a framework for autonomous correlation detection. The framework comprises a set of conventional (STA/LTA) detectors on a collection of array beams, augmented by correlation and subspace detectors. The detectors are applied in parallel. (4) We have experimentally tested the processing of a large aftershock sequence using various transformations of the signals from the main event to construct what we denote an incoherent correlation detector or characteristic function correlation detector. (5) An important application of multi-channel waveform correlation is to identify instances of erroneous instrumental timing. We provide examples of detected timing errors on single channels of an array, as well as on a remote three-component station.

58 GEOSCIENCES↗

A programmable shared-memory system for an array of processing-in-memory devices

Processing in memory (PIM), the concept of integrating processing directly with memory has been attracting a lot of attention, since PIM can assist in overcoming the throughput limitation caused by data movement between CPU and memory. The challenge, however, is that it requires the programmers to have a deep understanding of the PIM architecture to maximize the benefits such as data locality and parallel thread execution on multiple PIM devices. In this study, we present AnalyzeThat, a programmable shared-memory system for parallel data processing with PIM devices. Thematic to AnalyzeThat is a rich PIM-aware data structure (PADS), which is an encapsulation that integrally ties together the data, the analysis tasks and the runtime needed to interface with the PIM device array. The PADS abstraction provides (i) a sophisticated key-value data container that allows programmers to easily store data on multiple PIMs, (ii) a suite of parallel operations with which users can easily implement data analysis applications, and (iii) a runtime, hidden to programmers, which provides the mechanisms needed to overlay both the data and the tasks on the PIM device array in an intelligent fashion, based on PIM-specific information collected from the hardware. We have developed a PIM emulation framework called AnalyzeThat. In conclusion, our experimental evaluation with representative data analytics applications suggests that the proposed system can significantly reduce the PIM programming effort without losing its technology benefits.

97 MATHEMATICS AND COMPUTING↗