Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Network partitioning”

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 199 records · Page 11

Scalable balanced training of conditional generative adversarial neural networks on image data

Here, we propose a distributed approach to train deep convolutional generative adversarial neural network (DC-CGANs) models. Our method reduces the imbalance between generator and discriminator by partitioning the training data according to data labels, and enhances scalability by performing a parallel training where multiple generators are concurrently trained, each one of them focusing on a single data label. Performance is assessed in terms of inception score, Fréchet inception distance, and image quality on MNIST, CIFAR10, CIFAR100, and ImageNet1k datasets, showing a significant improvement in comparison to state-of-the-art techniques to training DC-CGANs. Weak scaling is attained on all the four datasets using up to 1000 processes and 2000 NVIDIA V100 GPUs on the OLCF supercomputer Summit.

97 MATHEMATICS AND COMPUTING↗

DRIPS: Dynamic Rebalancing of Pipelined Streaming Applications on CGRAs

Coarse-grained reconfigurable arrays (CGRAs) provide higher flexibility than application-specific integrated circuits (ASICs) and higher efficiency than fine-grained reconfigurable devices such as Field Programmable Gate Arrays (FPGAs). However, CGRAs are generally designed to support offloading of a single kernel. While their design, based on communicating functional units, appears to naturally suit data streaming applications composed of multiple cooperating kernels, current approaches only statically partition the resources across application kernels. However, emerging streaming applications at the edge (scientific instruments, sensor networks, network processing) perform much more than digital signal processing and often are data and input dependent. This leads to extremely variable kernel execution times, severely impacting the throughput of the entire pipeline if resources are only statically allocated. Therefore, in this paper, we propose DRIPS — a coarse-grained, dynamically, and partially reconfigurable array for data-dependent streaming applications. We present a unified compiler framework to facilitate the mapping of a given streaming application onto the DRIPS CGRA architecture. The experimental results show that DRIPS achieves an average throughput improvement of 1.46$\times$ across a set of representative applications over a statically partitioned solution. The additional area overhead to enable dynamic rebalancing consumes 16.34% of the entire area for a 5x5 CGRA prototype.

Tan, Cheng↗

Dynamic Configuration of the CMS Data Acquisition Cluster

The CMS Data Acquisition cluster, which runs around 10000 applications, is configured dynamically at run time. XML configuration documents determine what applications are executed on each node and over what networks these applications communicate. Through this mechanism the DAQ System may be adapted to the required performance, partitioned in order to perform (test-) runs in parallel, or re-structured in case of hardware faults. This paper presents the configuration procedure and the CMS DAQ Confi gurator tool, which is used to generate comprehensive configurations of the CMS DAQ system based on a high-level description given by the user. Using a database of configuration templates and a database containing a detailed model of hardware modules, dat a and control links, nodes and the network topology, the tool automatically determines which applications are needed, on which nodes they should run, and over which networks the event traffic will flow. The tool computes application parameters and generat es the XML configuration documents and the configuration of the run-control system. The performance of the configuration procedure and the tool as well as operational experience during CMS commissioning and the first LHC runs are discussed.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Vegetation type is an important predictor of the arctic summer land surface energy budget

Despite the importance of high-latitude surface energy budgets (SEBs) for land-climate interactions in the rapidly changing Arctic, uncertainties in their prediction persist. Here, we harmonize SEB observations across a network of vegetated and glaciated sites at circumpolar scale (1994–2021). Our variance-partitioning analysis identifies vegetation type as an important predictor for SEB-components during Arctic summer (June-August), compared to other SEB-drivers including climate, latitude and permafrost characteristics. Differences among vegetation types can be of similar magnitude as between vegetation and glacier surfaces and are especially high for summer sensible and latent heat fluxes. The timing of SEB-flux summer-regimes (when daily mean values exceed 0 Wm –2 ) relative to snow-free and -onset dates varies substantially depending on vegetation type, implying vegetation controls on snow-cover and SEB-flux seasonality. Our results indicate complex shifts in surface energy fluxes with land-cover transitions and a lengthening summer season, and highlight the potential for improving future Earth system models via a refined representation of Arctic vegetation types.

54 ENVIRONMENTAL SCIENCES↗

Improved Multi-Partition Method for Line-Based Iteration Schemes

Regular 3-dimensional multi-partitioning has been shown to be an efficient domain decomposition method for the parallelization of ADI-type algorithms on MIMD architectures. This paper discusses further improvements that can be made to the scheme that increase the granularity and reduce the communication density. These improvements, which are illustrated by simulation and parallel benchmark results, make multi-partitioning the method of choice on systems with relatively poor communication capabilities, such as networks of workstations, or on massively parallel machines with very fast processors, such as the IBM SP2.

Smith, Merritt H.↗

brulilo, Version 0.x

brulilo is a Python package for building and evolving thermonuclear reaction networks, such as those used in astrophysical explosion calculations. It is very much a current work in progress, and much of the intended functionality is still under development. The main goal of brulilo is an easy-to-use code for quick calculations or post-processing, as well as a learning tool to test various implementations of solvers on both CPUs and GPUs. Currently, brulilo parses an input text file that specifies the reactions to be included in the network. The notation of the inputs are in the typical astrophysical notation; i.e. hydrogen burning to deuterium in the sun, "p + p + e --> d + neutrino", is given as "p(p,)d." Each reaction is added to a list comprising the reaction network, and once completed, the network contains a list of all isotopes and reactions involved. brulilo then parses publicly available nuclear data (binding energy, spin, partition function data, etc.) for each isotope from the WebNucleo group's data at nucleo.ces.clemson.edu. Fits to reaction rate data for each reaction are obtained from the publicly available data provided by the Joint Institute for Nuclear Astrophysics' Reaclib database. Reverse reactions are calculated from detailed balance. The reaction network then consists of a sparse system of mathematically stiff ordinary differential equations (ODEs). Physically, "stiffness" here means that there are some reaction rates that are occurring much faster than others; the ratio of the fast to slow often being greater than 10^15 in astrophysics. Numerical integration of such systems typically involves implicit integration schemes, which brulilo will leverage from the scientific computing Python package, Numpy, as well as implement its own. There has recently been work from researchers at Oak Ridge National Lab who have found good results from applying quasi-equilibrium and quasi-asymptotic conditions to the stiff system of ODEs, which effectively remove some of the stiffness and allow for efficient explicit integration techniques to be used. The original intent of brulilo was to implement these stiffness-alleviating techniques with explicit integrators and compare the performance to traditional implicit integrations of the full stiff system. This is still underway, as the code is very much in an alpha-release state. Furthermore, explicit integrators are often much easier to parallelize than their implicit counterparts. brulilo will implement parallelization of these techniques, leveraging both the Python implementation of MPI, mpi4py, as well as highly parallelized versions targeted at GPUs with PyOpenCL and/or PyCUDA.

Malone, Chris↗

brulilo, Version 0.x

brulilo is a Python package for building and evolving thermonuclear reaction networks, such as those used in astrophysical explosion calculations. It is very much a current work in progress, and much of the intended functionality is still under development. The main goal of brulilo is an easy-to-use code for quick calculations or post-processing, as well as a learning tool to test various implementations of solvers on both CPUs and GPUs. Currently, brulilo parses an input text file that specifies the reactions to be included in the network. The notation of the inputs are in the typical astrophysical notation; i.e. hydrogen burning to deuterium in the sun, "p + p + e --> d + neutrino", is given as "p(p,)d." Each reaction is added to a list comprising the reaction network, and once completed, the network contains a list of all isotopes and reactions involved. brulilo then parses publicly available nuclear data (binding energy, spin, partition function data, etc.) for each isotope from the WebNucleo group's data at nucleo.ces.clemson.edu. Fits to reaction rate data for each reaction are obtained from the publicly available data provided by the Joint Institute for Nuclear Astrophysics' Reaclib database. Reverse reactions are calculated from detailed balance. The reaction network then consists of a sparse system of mathematically stiff ordinary differential equations (ODEs). Physically, "stiffness" here means that there are some reaction rates that are occurring much faster than others; the ratio of the fast to slow often being greater than 10^15 in astrophysics. Numerical integration of such systems typically involves implicit integration schemes, which brulilo will leverage from the scientific computing Python package, Numpy, as well as implement its own. There has recently been work from researchers at Oak Ridge National Lab who have found good results from applying quasi-equilibrium and quasi-asymptotic conditions to the stiff system of ODEs, which effectively remove some of the stiffness and allow for efficient explicit integration techniques to be used. The original intent of brulilo was to implement these stiffness-alleviating techniques with explicit integrators and compare the performance to traditional implicit integrations of the full stiff system. This is still underway, as the code is very much in an alpha-release state. Furthermore, explicit integrators are often much easier to parallelize than their implicit counterparts. brulilo will implement parallelization of these techniques, leveraging both the Python implementation of MPI, mpi4py, as well as highly parallelized versions targeted at GPUs with PyOpenCL and/or PyCUDA.

Malone, Chris↗

Distributed simulation using a real-time shared memory network

The Advanced Control Technology Branch of the NASA Lewis Research Center performs research in the area of advanced digital controls for aeronautic and space propulsion systems. This work requires the real-time implementation of both control software and complex dynamical models of the propulsion system. We are implementing these systems in a distributed, multi-vendor computer environment. Therefore, a need exists for real-time communication and synchronization between the distributed multi-vendor computers. A shared memory network is a potential solution which offers several advantages over other real-time communication approaches. A candidate shared memory network was tested for basic performance. The shared memory network was then used to implement a distributed simulation of a ramjet engine. The accuracy and execution time of the distributed simulation was measured and compared to the performance of the non-partitioned simulation. The ease of partitioning the simulation, the minimal time required to develop for communication between the processors and the resulting execution time all indicate that the shared memory network is a real-time communication technique worthy of serious consideration.

Simon, Donald L.↗

Modeling the contributions of global air temperature, synoptic-scale phenomena and soil moisture to near-surface static energy variability using artificial neural networks

Abstract. The static energy content of the atmosphere is increasing on a global scale, but exhibits important subglobal and subregional scales of variability and is a useful parameter for integrating the net effect of changes in the partitioning of energy at the surface and for improving understanding of the causes of so-called warming holes (i.e., locations with decreasing daily maximum air temperatures ( T ) or increasing trends of lower magnitude than the global mean). Further, measures of the static energy content (herein the equivalent potential temperature, θ e ) are more strongly linked to excess human mortality and morbidity than air temperature alone, and have great relevance in understanding causes of past heat-related excess mortality and making projections of possible future events that are likely to be associated with negative human health and economic consequences. New nonlinear statistical models for summertime daily maximum and minimum θ e are developed and used to advance understanding of drivers of historical change and variability over the eastern USA. The predictor variables are an index of the daily global mean temperature, daily indices of the synoptic-scale meteorology derived from T and specific humidity ( Q ) at 850 and 500 hPa geopotential heights ( Z ), and spatiotemporally averaged soil moisture ( SM ). SM is particularly important in determining the magnitude of θ e over regions that have previously been identified as exhibiting warming holes, confirming the key importance of SM in dictating the partitioning of net radiation into sensible and latent heat and dictating trends in near-surface T and θ e . Consistent with our a priori expectations, models built using artificial neural networks (ANNs) out-perform linear models that do not permit interaction of the predictor variables (global T , synoptic-scale meteorological conditions and SM ). This is particularly marked in regions with high variability in minimum and maximum θ e , where more complex models built using ANN with multiple hidden layers are better able to capture the day-to-day variability in θ e and the occurrence of extreme maximum θ e . Over the entire domain, the ANN with three hidden layers exhibits high accuracy in predicting maximum θ e > 347 K. The median hit rate for maximum θ e >347 K is > 0.60, while the median false alarm rate is ≈ 0.08.

54 ENVIRONMENTAL SCIENCES↗

Sensitivity evaluation of dynamic speckle activity measurements using clustering methods

We evaluate and compare the use of competitive neural networks, self-organizing maps, the expectation-maximization algorithm, K-means, and fuzzy C-means techniques as partitional clustering methods, when the sensitivity of the activity measurement of dynamic speckle images needs to be improved. The temporal history of the acquired intensity generated by each pixel is analyzed in a wavelet decomposition framework, and it is shown that the mean energy of its corresponding wavelet coefficients provides a suited feature space for clustering purposes. The sensitivity obtained by using the evaluated clustering techniques is also compared with the well-known methods of Konishi-Fujii, weighted generalized differences, and wavelet entropy. The performance of the partitional clustering approach is evaluated using simulated dynamic speckle patterns and also experimental data.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Prognostic Indexes for Brain Metastases: Which Is the Most Powerful?

Purpose: The purpose of the present study was to compare the prognostic indexes (PIs) of patients with brain metastases (BMs) treated with whole brain radiotherapy (WBRT) using an artificial neural network. This analysis is important, because it evaluates the prognostic power of each PI to guide clinical decision-making and outcomes research. Methods and Materials: A retrospective prognostic study was conducted of 412 patients with BMs who underwent WBRT between April 1998 and March 2010. The eligibility criteria for patients included having undergone WBRT or WBRT plus neurosurgery. The data were analyzed using the artificial neural network. The input neural data consisted of all prognostic factors included in the 5 PIs (recursive partitioning analysis, graded prognostic assessment [GPA], basic score for BMs, Rotterdam score, and Germany score). The data set was randomly divided into 300 training and 112 testing examples for survival prediction. All 5 PIs were compared using our database of 412 patients with BMs. The sensibility of the 5 indexes to predict survival according to their input variables was determined statistically using receiver operating characteristic curves. The importance of each variable from each PI was subsequently evaluated. Results: The overall 1-, 2-, and 3-year survival rate was 22%, 10.2%, and 5.1%, respectively. All classes of PIs were significantly associated with survival (recursive partitioning analysis, P < .0001; GPA, P < .0001; basic score for BMs, P = .002; Rotterdam score, P = .001; and Germany score, P < .0001). Comparing the areas under the curves, the GPA was statistically most sensitive in predicting survival (GPA, 86%; recursive partitioning analysis, 81%; basic score for BMs, 79%; Rotterdam, 73%; and Germany score, 77%; P < .001). Among the variables included in each PI, the performance status and presence of extracranial metastases were the most important factors. Conclusion: A variety of prognostic models describe the survival of patients with BMs to a more or less satisfactory degree. Among the 5 PIs evaluated in the present study, GPA was the most powerful in predicting survival. Additional studies should include emerging biologic prognostic factors to improve the sensibility of these PIs.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Diversity-oriented synthesis of polymer membranes with ion solvation cages

Microporous polymers feature shape-persistent free volume elements (FVEs), which are permeated by small molecules and ions when used as membranes for chemical separations, water purification, fuel cells, and batteries. It remains a significant challenge to identify FVEs with analyte specificity, due to difficulties in generating microporous polymer libraries with sufficient diversity for screening their properties. Here, we describe a diversity-oriented synthetic (DOS) strategy for microporous polymer membranes from which we identified those whose FVEs serve as solid solvation cages for lithium ions (Li + ). Furthermore, key elements of our strategy included diversification of bis(catechol)-type monomers via multi-component Mannich reactions to introduce Li+-coordinating functionality within individual FVEs, topology-enforcing polymerizations for generating macromolecular skeletal diversity for networking FVEs into different pore architectures, and several classes of on-polymer reactions for diversifying pore geometries and dielectric properties. Lead candidate polymer membranes featuring explicit ion solvation cages exhibited both higher ionic conductivity and higher cation transference number than control membranes where FVEs were aspecific, which indicates conventional bounds for membrane permeability and selectivity for ion transport can be overcome.4 These advantages are tied to enhanced Li + partitioning from the electrolyte when the cages are present, higher diffusion barriers for anions within the pores, and network-enforced restrictions on the number of solvent molecules bound to Li+ by comparison to the bulk electrolyte, which reduces the effective mass of the working ion. Such membranes show promise as anode-stabilizing interlayers in high-voltage lithium-metal batteries for electric mobility.

42 ENGINEERING↗

D2NO: Efficient handling of heterogeneous input function spaces with distributed deep neural operators

Neural operators have been applied in various scientific fields, such as solving parametric partial differential equations, dynamical systems with control, and inverse problems. However, challenges arise when dealing with input functions that exhibit heterogeneous properties, requiring multiple sensors to handle functions with minimal regularity. To address this issue, discretization-invariant neural operators have been used, allowing the sampling of diverse input functions with different sensor locations. However, existing frameworks still require an equal number of sensors for all functions. We propose a novel distributed approach to further relax the discretization requirements and solve the heterogeneous dataset challenges. Our method involves partitioning the input function space and processing individual input functions using independent and separate neural networks. A centralized neural network is used to handle shared information across all output functions. This distributed methodology reduces the number of gradient descent back-propagation steps, improving efficiency while maintaining accuracy. Here, we demonstrate that the corresponding neural network is a universal approximator of continuous nonlinear operators and present three numerical examples to validate its performance.

97 MATHEMATICS AND COMPUTING↗

Widespread inhibition of daytime ecosystem respiration

The global land surface absorbs about a third of anthropogenic emissions each year, due to the difference between two key processes: ecosystem photosynthesis and respiration. Despite the importance of these two processes, it is not possible to measure either at the ecosystem scale during the daytime. Eddy-covariance measurements are widely used as the closest 'quasi-direct' ecosystem-scale observation from which to estimate ecosystem photosynthesis and respiration. Recent research, however, suggests that current estimates may be biased by up to 25%, due to a previously unaccounted for process: the inhibition of leaf respiration in the light. Yet the extent of inhibition remains debated, and implications for estimates of ecosystem-scale respiration and photosynthesis remain unquantified. Here, we quantify an apparent inhibition of daytime ecosystem respiration across the global FLUXNET eddy-covariance network and identify a pervasive influence that varies by season and ecosystem type. We develop partitioning methods that can detect an apparent ecosystem-scale inhibition of daytime respiration and find that diurnal patterns of ecosystem respiration might be markedly different than previously thought. The results call for the re-evaluation of global terrestrial carbon cycle models and also suggest that current global estimates of photosynthesis and respiration may be biased, some on the order of magnitude of anthropogenic fossil fuel emissions.

54 ENVIRONMENTAL SCIENCES↗

Lipid sponge droplets as programmable synthetic organelles

Living cells segregate molecules and reactions in various subcellular compartments known as organelles. Spatial organization is likely essential for expanding the biochemical functions of synthetic reaction systems, including artificial cells. Many studies have attempted to mimic organelle functions using lamellar membrane-bound vesicles. However, vesicles typically suffer from highly limited transport across the membranes and an inability to mimic the dense membrane networks typically found in organelles such as the endoplasmic reticulum. Here, we describe programmable synthetic organelles based on highly stable nonlamellar sponge phase droplets that spontaneously assemble from a single-chain galactolipid and nonionic detergents. Due to their nanoporous structure, lipid sponge droplets readily exchange materials with the surrounding environment. In addition, the sponge phase contains a dense network of lipid bilayers and nanometric aqueous channels, which allows different classes of molecules to partition based on their size, polarity, and specific binding motifs. The sequestration of biologically relevant macromolecules can be programmed by the addition of suitably functionalized amphiphiles to the droplets. We demonstrate that droplets can harbor functional soluble and transmembrane proteins, allowing for the colocalization and concentration of enzymes and substrates to enhance reaction rates. Droplets protect bound proteins from proteases, and these interactions can be engineered to be reversible and optically controlled. Our results show that lipid sponge droplets permit the facile integration of membrane-rich environments and self-assembling spatial organization with biochemical reaction systems.

59 BASIC BIOLOGICAL SCIENCES↗

Designing Scalable PGAS Communication Subsystems on Cray Gemini Interconnect

The Cray Gemini Interconnect has been recently introduced as a next generation network architecture for building multi-petaflop supercomputers. Cray XE6 systems including LANL Cielo, NERSC Hopper, ORNL Titan and proposed NCSA BlueWaters leverage the Gemini Interconnect as their primary Interconnection network. At the same time, programming models such as the Message Passing Interface (MPI) and Partitioned Global Address Space (PGAS) models such as Unified Parallel C (UPC) and Co-Array Fortran (CAF) have become available on these systems. Global Arrays is a popular PGAS model used in a variety of application domains including hydrodynamics, chemistry and visualization. Global Arrays uses Aggregate Re- mote Memory Copy Interface (ARMCI) as the communication runtime system for Remote Memory Access communication. This paper presents a design, implementation and performance evaluation of scalable and high performance communication subsystems on Cray Gemini Interconnect using ARMCI. The design space is explored and time-space complexities of commu- nication protocols for one-sided communication primitives such as contiguous and uniformly non-contiguous datatypes, atomic memory operations (AMOs) and memory synchronization is presented. An implementation of the proposed design (referred as ARMCI-Gemini) demonstrates the efficacy on communication primitives, application kernels such as LU decomposition and full applications such as Smooth Particle Hydrodynamics (SPH) application.

ARMCI↗

Evaluating Spatial Accelerator Architectures with Tiled Matrix-Matrix Multiplication.

There is a growing interest in custom spatial accelerators for machine learning applications. These accelerators employ a spatial array of processing elements (PEs) interacting via custom buffer hierarchies and networks-on-chip. The efficiency of these accelerators comes from employing optimized dataflow (i.e., spatial/temporal partitioning of data across the PEs and fine-grained scheduling) strategies to optimize data reuse. The focus of this work is to evaluate these accelerator architectures using a tiled general matrix-matrix multiplication (GEMM) kernel. To do so, we develop a framework that finds optimized mappings (dataflow and tile sizes) for a tiled GEMM for a given spatial accelerator and workload combination, leveraging an analytical cost model for runtime and energy. Finally, our evaluations over five spatial accelerators demonstrate that the tiled GEMM mappings systematically generated by our framework achieve high performance on various GEMM workloads and accelerators.

42 ENGINEERING↗

Evaluating Spatial Accelerator Architectures with Tiled Matrix-Matrix Multiplication

There is a growing interest in custom spatial accelerators for machine learning applications. These accelerators employ a spatial array of processing elements (PEs) interacting via custom buffer hierarchies and networks-on-chip. The efficiency of these accelerators comes from employing optimized dataflow (i.e., spatial/temporal partitioning of data across the PEs and fine-grained scheduling) strategies to optimize data reuse. The focus of this work is to evaluate these accelerator architectures using a tiled general matrix-matrix multiplication (GEMM) kernel. To do so, we develop a framework that finds optimized mappings (dataflow and tile sizes) for a tiled GEMM for a given spatial accelerator and workload combination, leveraging an analytical cost model for runtime and energy. Our evaluations over five spatial accelerators demonstrate that the tiled GEMM mappings systematically generated by our framework achieve high performance on various GEMM workloads and accelerators.

43 PARTICLE ACCELERATORS↗