Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Parallel Supercomputers”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

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

At least 397 records · Page 22

Reducing the Time to Tune Parallel Dense Linear Algebra Routines with Partial Execution and Performance Modeling

We present a modeling framework to accurately predict time to run dense linear algebra calculation. We report the framework’s accuracy in a number of varied computational environments such as shared memory multicore systems, clusters, and large supercomputing installations with tens of thousands of cores. We also test the accuracy for various algorithms, each of which having a different scaling properties and tolerance to low-bandwidth/high-latency interconnects. The predictive accuracy is very good and on the order of measurement accuracy which makes the method suitable for both dedicated and non-dedicated environments. We also present a practical application of our model to reduce the time required to tune and optimize large parallel runs whose time is dominated by linear algebra computations. We show practical examples of how to apply the methodology to avoid common pitfalls and reduce the influence of measurement errors and the inherent performance variability.

Linear systems↗

Online data analysis and reduction: An important co-design motif for extreme-scale computers

A growing disparity between supercomputer computation speeds and I/O rates means that it is rapidly becoming infeasible to analyze supercomputer application output only after that output has been written to a file system. Instead, data-generating applications must run concurrently with data reduction and/or analysis operations, with which they exchange information via high-speed methods such as interprocess communications. The resulting parallel computing motif, online data analysis and reduction (ODAR), has important implications for both application and HPC systems design. Here we introduce the ODAR motif and its co-design concerns, describe a co-design process for identifying and addressing those concerns, present tools that assist in the co-design process, and present case studies to illustrate the use of the process and tools in practical settings.

Data Analysis↗

Stochastic Simulation of Uranium Migration at the Hanford 300 Area

The persistence of a hexavalent uranium [U(VI)] plume in the subsurface at the Hanford 300 Area is of concern to the U.S. Department of Energy (DOE). This work focusses on the quantification of groundwater flow and subsequent U(VI) transport uncertainty due to heterogeneity in the sediment permeability at the site. U(VI) migration at the Hanford 300 Area is simulated with multiple realizations of stochastically-generated high resolution permeability fields and comparisons are made of cumulative water and U(VI) flux to the Columbia River. The massively parallel code PFLOTRAN developed under the DOE SciDAC-2 project is employed on up to 40,960 processor cores on DOE's petascale Jaguar supercomputer to simultaneously execute 10 transient, variably-saturated groundwater flow and U(VI) transport simulations within 3D random permeability fields using the code's multi-realization simulation capability. Simulation results demonstrate that the cumulative U(VI) flux to the Columbia River is less responsive to fine scale heterogeneity in permeability and more sensitive to the distribution of permeability within the river hyporheic zone and mean permeability of larger scale geologic structures at the site.

58 GEOSCIENCES↗

Optimizing the hypre solver for manycore and GPU architectures

The solution of large-scale combustion problems with codes such as Uintah on modern computer architectures requires the use of multithreading and GPUs to achieve performance. Uintah uses a low-Mach number approximation that requires iteratively solving a large system of linear equations. The Hypre iterative solver has solved such systems in a scalable way for Uintah, but the use of OpenMP with Hypre leads to at least slowdown due to OpenMP overheads. The proposed solution uses the MPI Endpoints within Hypre, where each team of threads acts as a different MPI rank. This approach minimizes OpenMP synchronization overhead and performs as fast or (up to 1.44) faster than Hypre's MPI-only version, and allows the rest of Uintah to be optimized using OpenMP. The profiling of the GPU version of Hypre shows the bottleneck to be the launch overhead of thousands of micro-kernels. The GPU performance was improved by fusing these micro-kernels and was further optimized by using Cuda-aware MPI, resulting in an overall speedup of 1.16—1.44 compared to the baseline GPU implementation. The above optimization strategies were published in the International Conference on Computational Science 2020 [1]. This work extends the previously published research by carrying out the second phase of communication-centered optimizations in Hypre to improve its scalability on large-scale supercomputers. Additionally, this includes an efficient non-blocking inter-thread communication scheme, communication-reducing patch assignment, and expression of logical communication parallelism to a new version of the MPICH library that utilizes the underlying network parallelism [2]. The above optimizations avoid communication bottlenecks previously observed during strong scaling and improve performance by up to 2 on 256 nodes of Intel Knight's Landing processor.

97 MATHEMATICS AND COMPUTING↗

CosmoFlow: Using Deep Learning to Learn the Universe at Scale

Deep learning is a promising tool to determine the physical model that describes our universe. To handle the considerable computational cost of this problem, we present CosmoFlow: a highly scalable deep learning application built on top of the TensorFlow framework. CosmoFlow uses efficient implementations of 3D convolution and pooling primitives, together with improvements in threading for many element-wise operations, to improve training performance on Intel ® Xeon Phi™ processors. We also utilize the Cray PE Machine Learning Plugin for efficient scaling to multiple nodes. We demonstrate fully synchronous data-parallel training on 8192 nodes of Cori with 77% parallel efficiency, achieving 3.5 Pflop/s sustained performance. To our knowledge, this is the first large-scale science application of the TensorFlow framework at supercomputer scale with fully-synchronous training. Here, these enhancements enable us to process large 3D dark matter distribution and predict the cosmological parameters Ω M , σ 8 and n s with unprecedented accuracy.

98 NUCLEAR DISARMAMENT, SAFEGUARDS, AND PHYSICAL P↗

Multiscale Molecular Simulations at the Petascale: Parallelization of Reactive Force Field Model for Blue Gene/Q (ALCF-2 Early Science Program Technical Report)

Our efforts in developing a highly parallel implementation of the reactive force field model, the Multi-state Empirical Valence Bond (MS-EVB) method, are discussed. We have introduced multi-threading, a state decomposition parallelism, and replica exchange capabilities. Example calculations are presented to demonstrate the improved productivity and scalability of the new code on the Blue Gene/Q supercomputer, Mira. We show that we are now able to successfully scale to half of Mira and have the potential to scale even further.

Lange, Adiran W.↗

nbodykit: An Open-source, Massively Parallel Toolkit for Large-scale Structure

We present nbodykit, an open-source, massively parallel Python toolkit for analyzing large-scale structure (LSS) data. Using Python bindings of the Message Passing Interface, we provide parallel implementations of many commonly used algorithms in LSS. nbodykit is both an interactive and scalable piece of scientific software, performing well in a supercomputing environment while still taking advantage of the interactive tools provided by the Python ecosystem. Existing functionality includes estimators of the power spectrum, two- and three-point correlation functions, a friends-of-friends grouping algorithm, mock catalog creation via the halo occupation distribution technique, and approximate N-body simulations via the FastPM scheme. The package also provides a set of distributed data containers, insulated from the algorithms themselves, that enables nbodykit to provide a unified treatment of both simulation and observational data sets. nbodykit can be easily deployed in a high-performance computing environment, overcoming some of the traditional difficulties of using Python on supercomputers. We provide performance benchmarks illustrating the scalability of the software. The modular, component-based approach of nbodykit allows researchers to easily build complex applications using its tools. The package is extensively documented at http://nbodykit.readthedocs.io, which also includes an interactive set of example recipes for new users to explore. As open-source software, we hope nbodykit provides a common framework for the community to use and develop in confronting the analysis challenges of future LSS surveys.

79 ASTRONOMY AND ASTROPHYSICS↗

Parallel k-Means Clustering for Quantitative Ecoregion Delineation Using Large Data Sets

Identification of geographic ecoregions has long been of interest to environmental scientists and ecologists for identifying regions of similar ecological and environmental conditions. Such classifications are important for predicting suitable species ranges, for stratification of ecological samples, and to help prioritize habitat preservation and remediation efforts. Hargrove and Hoffman (1999, 2009) have developed geographical spatio-temporal clustering algorithms and codes and have successfully applied them to a variety of environmental science domains, including ecological regionalization; environmental monitoring network design; analysis of satellite-, airborne-, and ground-based remote sensing, and climate model-model and model-measurement intercomparison. With the advances in state-of-the-art satellite remote sensing and climate models, observations and model outputs are available at increasingly high spatial and temporal resolutions. Long time series of these high resolution datasets are extremely large in size and growing. Analysis and knowledge extraction from these large datasets are not just algorithmic and ecological problems, but also pose a complex computational problem. This paper focuses on the development of a massively parallel multivariate geographical spatio-temporal clustering code for analysis of very large datasets using tens of thousands processors on one of the fastest supercomputers in the world.

54 ENVIRONMENTAL SCIENCES↗

Scalable Earthquake Simulation on Petascale Supercomputers

NPetascale simulations are needed to understand the rupture and wave dynamics of the largest earthquakes at shaking frequencies required to engineer safe structures (> 1 Hz). Toward this goal, we have developed a highly scalable, parallel application (AWP-ODC) that has achieved “M8”: a full dynamical simulation of a magnitude-8 earthquake on the southern San Andreas fault up to 2 Hz. M8 was calculated using a uniform mesh of 436 billion 40-m 3 cubes to represent the three-dimensional crustal structure of Southern California, in a 800 km by 400 km area, home to over 20 million people. This production run producing 360 sec of wave propagation sustained 220 Tflop/s for 24 hours on NCCS Jaguar using 223,074 cores. As the largest-ever earthquake simulation, M8 opens new territory for earthquake science and engineering - the physics-based modeling of the largest seismic hazards with the goal of reducing their potential for loss of life and property.

Cui, Yifeng↗

A Foundation for the Accurate Prediction of the Soft Error Vulnerability of Scientific Applications

Understanding the soft error vulnerability of supercomputer applications is critical as these systems are using ever larger numbers of devices that have decreasing feature sizes and, thus, increasing frequency of soft errors. As many large scale parallel scientific applications use BLAS and LAPACK linear algebra routines, the soft error vulnerability of these methods constitutes a large fraction of the applications overall vulnerability. This paper analyzes the vulnerability of these routines to soft errors by characterizing how their outputs are affected by injected errors and by evaluating several techniques for predicting how errors propagate from the input to the output of each routine. The resulting error profiles can be used to understand the fault vulnerability of full applications that use these routines.

99 GENERAL AND MISCELLANEOUS↗

A fully coupled two-level Schwarz preconditioner based on smoothed aggregation for the transient multigroup neutron diffusion equations

The neutron multigroup diffusion equations is an important approximation of the neutron transport equation, and has been widely used for studying the motion of neutrons and their interactions with the stationary background materials. Solving the multigroup diffusion equations faces challenging because many variables are coupled together and a lot of computing resources are required. In this paper, we focus on the development of a scalable parallel algorithm framework for the system of equations arising from the discretization of the multigroup diffusion equations using a finite element method in space and using an implicit scheme in time. The parallel agorithm framweork consists of an inexact Jacobian-free Newton for the nonlinear equations, a Krylov subspace method for the Jacobian system and a parallel Schwarz preconditioner for the acceleration of the algorithm convergence. A coarse space based on smoothed aggregation is introduced to construct a two-level method to improve the parallel performance of the Schwarz preconditioner. We numerically shown that compared with the one-level method, the proposed two-level method works significantly better in terms of the number of linear iterations and the compute time for a system of equations with millions of unknowns and a supercomputer with thousands of processors.

97 MATHEMATICS AND COMPUTING↗

Developing Ultrahigh-Resolution E3SM Land Model for GPU Systems

Designing and refactoring complex scientific code, such as the E3SM land model (ELM), for new computing architectures is challenging. This paper presents design strategies and technical approaches to develop a data-oriented, GPU-ready ELM model using compiler directives (OpenACC/OpenMP). We first analyze the datatypes and processes in the original ELM code. Then we present design considerations for ultrahigh-resolution ELM (uELM) development for massive GPU systems. These techniques include the global data-oriented simulation workflow, domain partition, code porting and data copy, memory reduction, parallel loop restructure and flattening, and race condition detection. We implemented the first version of uELM using OpenACC targeting the NVidia GPUs in the Summit supercomputer at Oak Ridge National Laboratory. During the implementation, we developed a software tool (named SPEL) to facilitate code generation, verification, and performance tuning using these techniques. The first uELM implementation for Nvidia GPUs on Summit delivered promising results: 1) over 98% of the ELM code was automatically generated and tuned by scripts. Most ELM modules had better computational performances than the original ELM code for CPUs. The GPU-ready uELM is more scalable than the CPU code on fully-loaded Summit nodes. Example profiling results from several modules are also presented to illustrate the performance improvements and race condition detection. The lessons learned and toolkit developed in the study are also suitable for further uELM deployment using OpenMP on the first US exascale computer, Frontier, equipped with AMD CPUs and GPUs.

Schwartz, Peter↗

Distributed Multi-GPU Community Detection on Exascale Computing Platforms

Community detection is a fundamental operation in graph mining, and by uncovering hidden structures and patterns within complex systems it helps solve fundamental problems pertaining to social networks, such as information diffusion, epidemics, and recommender systems. Scaling graph algorithms for massive networks becomes challenging on modern distributed-memory multi-GPU (Graphics Processing Unit) systems due to limitations such as irregular memory access patterns, load imbalances, higher communication-computation ratios, and cross-platform support. We present a novel algorithm HiPDPL-GPU (distributed parallel Louvain) to address these challenges. We conduct experiments involving different partitioning techniques to achieve optimized performance of HiPDPL-GPU on the two largest supercomputers: Frontier and Summit. Remarkably, HiPDPL-GPU processes a graph with 4.2 billion edges in less than 3 minutes using 1024 GPUs. Qualitatively performance of HiPDPL-GPU is similar or better compared to other state-of-the-art CPU- and GPU-based implementations. While prior GPU implementations have predominantly employed CUDA, our first-of-its-kind implementation for community detection is cross-platform, accommodating both AMD and NVIDIA GPUs.

graph algorithms, high performance comptuing↗

WARP

WARP, which can stand for ``Weaving All the Random Particles,'' is a three-dimensional (3D) continuous energy Monte Carlo neutron transport code developed at UC Berkeley to efficiently execute on NVIDIA graphics processing unit (GPU) platforms. WARP accelerates Monte Carlo simulations while preserving the benefits of using the Monte Carlo method, namely, that very few physical and geometrical simplifications are applied. WARP is able to calculate multiplication factors, neutron flux distributions (in both space and energy), and fission source distributions for time-independent neutron transport problems. It can run in both criticality or fixed source modes, but fixed source mode is currently not robust, optimized, or maintained in the newest version. WARP can transport neutrons in unrestricted arrangements of parallelepipeds, hexagonal prisms, cylinders, and spheres. The goal of developing WARP is to investigate algorithms that can grow into a full-featured, continuous energy, Monte Carlo neutron transport code that is accelerated by running on GPUs. The crux of the effort is to make Monte Carlo calculations faster while producing accurate results. Modern supercomputers are commonly being built with GPU coprocessor cards in their nodes to increase their computational efficiency and performance. GPUs execute efficiently on data-parallel problems, but most CPU codes, including those for Monte Carlo neutral particle transport, are predominantly task-parallel. WARP uses a data-parallel neutron transport algorithm to take advantage of the computing power GPUs offer.

Bergmann, RyanM. [University of California - Berke↗

WARP

WARP, which can stand for ``Weaving All the Random Particles,'' is a three-dimensional (3D) continuous energy Monte Carlo neutron transport code developed at UC Berkeley to efficiently execute on NVIDIA graphics processing unit (GPU) platforms. WARP accelerates Monte Carlo simulations while preserving the benefits of using the Monte Carlo method, namely, that very few physical and geometrical simplifications are applied. WARP is able to calculate multiplication factors, neutron flux distributions (in both space and energy), and fission source distributions for time-independent neutron transport problems. It can run in both criticality or fixed source modes, but fixed source mode is currently not robust, optimized, or maintained in the newest version. WARP can transport neutrons in unrestricted arrangements of parallelepipeds, hexagonal prisms, cylinders, and spheres. The goal of developing WARP is to investigate algorithms that can grow into a full-featured, continuous energy, Monte Carlo neutron transport code that is accelerated by running on GPUs. The crux of the effort is to make Monte Carlo calculations faster while producing accurate results. Modern supercomputers are commonly being built with GPU coprocessor cards in their nodes to increase their computational efficiency and performance. GPUs execute efficiently on data-parallel problems, but most CPU codes, including those for Monte Carlo neutral particle transport, are predominantly task-parallel. WARP uses a data-parallel neutron transport algorithm to take advantage of the computing power GPUs offer.

Bergmann, Ryan M.↗

Petascale Parallelization of the Gyrokinetic Toroidal Code

The Gyrokinetic Toroidal Code (GTC) is a global, three-dimensional particle-in-cell application developed to study microturbulence in tokamak fusion devices. The global capability of GTC is unique, allowing researchers to systematically analyze important dynamics such as turbulence spreading. In this work we examine a new radial domain decomposition approach to allow scalability onto the latest generation of petascale systems. Extensive performance evaluation is conducted on three high performance computing systems: the IBM BG/P, the Cray XT4, and an Intel Xeon Cluster. Overall results show that the radial decomposition approach dramatically increases scalability, while reducing the memory footprint - allowing for fusion device simulations at an unprecedented scale. After a decade where high-end computing (HEC) was dominated by the rapid pace of improvements to processor frequencies, the performance of next-generation supercomputers is increasingly differentiated by varying interconnect designs and levels of integration. Understanding the tradeoffs of these system designs is a key step towards making effective petascale computing a reality. In this work, we examine a new parallelization scheme for the Gyrokinetic Toroidal Code (GTC) [?] micro-turbulence fusion application. Extensive scalability results and analysis are presented on three HEC systems: the IBM BlueGene/P (BG/P) at Argonne National Laboratory, the Cray XT4 at Lawrence Berkeley National Laboratory, and an Intel Xeon cluster at Lawrence Livermore National Laboratory. Overall results indicate that the new radial decomposition approach successfully attains unprecedented scalability to 131,072 BG/P cores by overcoming the memory limitations of the previous approach. The new version is well suited to utilize emerging petascale resources to access new regimes of physical phenomena.

42 ENGINEERING↗

OpenMP Parallelization and Optimization of Graph-Based Machine Learning Algorithms

In this paper, we investigate the OpenMP parallelization and optimization of two novel data classification algorithms. The new algorithms are based on graph and PDE solution techniques and provide significant accuracy and performance advantages over traditional data classification algorithms in serial mode. The methods leverage the Nystrom extension to calculate eigenvalue/eigenvectors of the graph Laplacian and this is a self-contained module that can be used in conjunction with other graph-Laplacian based methods such as spectral clustering. We use performance tools to collect the hotspots and memory access of the serial codes and use OpenMP as the parallelization language to parallelize the most time-consuming parts. Where possible, we also use library routines. We then optimize the OpenMP implementations and detail the performance on traditional supercomputer nodes (in our case a Cray XC30), and test the optimization steps on emerging testbed systems based on Intel’s Knights Corner and Landing processors. We show both performance improvement and strong scaling behavior. Finally, a large number of optimization techniques and analyses are necessary before the algorithm reaches almost ideal scaling.

97 MATHEMATICS AND COMPUTING↗

Summary Report of Working Group 2: Computation

The working group on computation addressed three physics areas: (i) plasma-based accelerators (laser-driven and beam-driven), (ii) high gradient structure-based accelerators, and (iii) electron beam sources and transport [1]. Highlights of the talks in these areas included new models of breakdown on the microscopic scale, new three-dimensional multipacting calculations with both finite difference and finite element codes, and detailed comparisons of new electron gun models with standard models such as PARMELA. The group also addressed two areas of advances in computation: (i) new algorithms, including simulation in a Lorentz-boosted frame that can reduce computation time orders of magnitude, and (ii) new hardware architectures, like graphics processing units and Cell processors that promise dramatic increases in computing power. Highlights of the talks in these areas included results from the first large-scale parallel finite element particle-in-cell code (PIC), many order-of-magnitude speedup of, and details of porting the VPIC code to the Roadrunner supercomputer. The working group featured two plenary talks, one by Brian Albright of Los Alamos National Laboratory on the performance of the VPIC code on the Roadrunner supercomputer, and one by David Bruhwiler of Tech-X Corporation on recent advances in computation for advanced accelerators. Highlights of the talk by Albright included the first one trillion particle simulations, a sustained performance of 0.3 petaflops, and an eight times speedup of science calculations, including back-scatter in laser-plasma interaction. Highlights of the talk by Bruhwiler included simulations of 10 GeV accelerator laser wakefield stages including external injection, new developments in electromagnetic simulations of electron guns using finite difference and finite element approaches.

43 PARTICLE ACCELERATORS↗