Search NASA⌕ Search

SEARCH · Search NASA

Results for “partitioned algorithm”

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 55 records · Page 3

Parallel adaptation of general three-dimensional hybrid meshes

A new parallel dynamic mesh adaptation and load balancing algorithm for general hybrid grids has been developed. The meshes considered in this work are composed of four kinds of elements; tetrahedra, prisms, hexahedra and pyramids, which poses a challenge to parallel mesh adaptation. Additional complexity imposed by the presence of multiple types of elements affects especially data migration, updates of local data structures and interpartition data structures. Efficient partition of hybrid meshes has been accomplished by transforming them to suitable graphs and using serial graph partitioning algorithms. Communication among processors is based on the faces of the interpartition boundary and the termination detection algorithm of Dijkstra is employed to ensure proper flagging of edges for refinement. An inexpensive dynamic load balancing strategy is introduced to redistribute work load among processors after adaptation. In particular, only the initial coarse mesh, with proper weighting, is balanced which yields savings in computation time and relatively simple implementation of mesh quality preservation rules, while facilitating coarsening of refined elements. Special algorithms are employed for (i) data migration and dynamic updates of the local data structures, (ii) determination of the resulting interpartition boundary and (iii) identification of the communication pattern of processors. Several representative applications are included to evaluate the method.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

A Linear-Complexity Tensor Butterfly Algorithm for Compressing High-Dimensional Oscillatory Integral Operators

This paper presents a multilevel tensor compression algorithm called tensor butterfly algorithm for efficiently representing large-scale and high-dimensional oscillatory integral operators, including Green's functions for wave equations and integral transforms such as Radon transforms and Fourier transforms. The proposed algorithm leverages a tensor extension of the so-called complementary low-rank property of existing matrix butterfly algorithms. The algorithm partitions the discretized integral operator tensor into subtensors of multiple levels and factorizes each subtensor at the middle level as a Tucker-type interpolative decomposition, whose factor matrices are formed in a multilevel fashion. For a d-dimensional (d > 1) integral operator discretized into a 2d-mode tensor with n2d entries, the overall CPU time and memory requirement scale as O(nd), in stark contrast to the O(nd log n) complexity of existing matrix algorithms such as matrix butterfly algorithms and fast Fourier transforms (FFTs), where n is the number of points per direction. When comparing with other tensor algorithms such as quantized tensor train (QTT), the proposed algorithm also shows superior CPU and memory performance for tensor contraction. Remarkably, the tensor butterfly algorithm can efficiently model high-frequency Green's function interactions between two unit cubes, each spanning 512 wavelengths per direction, which represents problems of scale over 512× larger than that existing butterfly algorithms can handle, with the same amount of computation resources. On the other hand, for a problem representing 64 wavelengths per direction, which is the largest size existing algebraic matrix algorithms can handle, our tensor butterfly algorithm exhibits 200x speedups and 30× memory reduction compared with existing ones. Moreover, the tensor butterfly algorithm also permits O(nd)-complexity FFTs and Radon transforms up to d = 6 dimensions.

Kielstra, P Michael↗

Exploiting Geometric Partitioning in Task Mapping for Parallel Computes

We present a new method for mapping applications' MPI tasks to cores of a parallel computer such that applications' communication time is reduced. We address the case of sparse node allocation, where the nodes assigned to a job are not necessarily located in a contiguous block nor within close proximity to each other in the network, although our methods generalize to contiguous allocations as well. The goal is to assign tasks to cores so that interdependent tasks are performed by "nearby' cores, thus lowering the distance messages must travel, the amount of congestion in the network, and the overall cost of communication. Our new method applies a geometric partitioning algorithm to both the tasks and the processors, and assigns task parts to the corresponding processor parts. We also present a number of algorithmic optimizations that exploit specific features of the network or application. We show that, for the structured finite difference mini-application MiniGhost, our mapping methods reduced communication time up to 75% relative to MiniGhost's default mapping on 128K cores of a Cray XK7 with sparse allocation. For the atmospheric modeling code E3SM/HOMME, our methods reduced communication time up to 31% on 32K cores of an IBM BlueGene/Q with contiguous allocation.

97 MATHEMATICS AND COMPUTING↗

A GROUP FINDING ALGORITHM FOR MULTIDIMENSIONAL DATA SETS

We describe a density-based hierarchical group finding algorithm capable of identifying structures and substructures of any shape and density in multidimensional data sets where each dimension can be a numeric attribute with arbitrary measurement scale. This has applications in a wide variety of fields from finding structures in galaxy redshift surveys, to identifying halos and subhalos in N-body simulations and group finding in Local Group chemodynamical data sets. In general, clustering schemes require an a priori definition of a metric (a non-negative function that gives the distance between two points in a space) and the quality of clustering depends upon this choice. The general practice is to use a constant global metric which is optimal only if the clusters in the data are self-similar. For complex data configurations even the most finely tuned constant global metric turns out to be suboptimal. Moreover, the correct choice of metric also becomes increasingly important as the number of dimensions increase. To address these problems, we present an entropy-based binary space partitioning algorithm which uses a locally adaptive metric for each data point. The metric is employed to calculate the density at each point and a list of its nearest neighbors, and this information is then used to form a hierarchy of groups. Finally, the ratio of maximum to minimum density of points in a group is used to estimate the significance of the groups. Setting a threshold on this significance can effectively screen out groups arising due to Poisson noise and helps organize the groups into meaningful clusters. For a data set of N points, the algorithm requires only O(N) space and O(N(log N){sup 3}) time which makes it ideally suitable for analyzing large data sets. As an example, we apply the algorithm to identify structures in a simulated stellar halo using the full six-dimensional phase space coordinates.

79 ASTRONOMY AND ASTROPHYSICS↗

Aboveground and belowground contributions to ecosystem respiration in a temperate deciduous forest

In this study, we developed a three-way carbon dioxide (CO 2 ) flux-partitioning algorithm that separates net ecosystem exchange (NEE) into aboveground plant respiration (R above ), belowground root and soil respiration (R below ), and gross primary production (GPP). We applied this algorithm to a coupled dataset of continuous chamber-measured soil respiration and eddy covariance (EC)-measured NEE of CO 2 in an oak-hickory (Quercus-Carya) deciduous broadleaf forest from 2006 to 2015. We found that on annual time scale, R below dominated over R above with the former accounting for 66.9–86.4% and the latter 13.6–33.1%, of the total ecosystem respiration (R eco ). The ratio of R below to R above varied seasonally, ranging from 1.77 to 7.25 in growing season, and 1.02 to 4.57 in non-growing season. The temperature sensitivity (E 0 ) of R below was significantly higher than that of R above , and E 0 of R eco responded differently to air and soil temperature. Over the whole study period, annual mean R above , R below , and GPP were 243, 806, and 1170 g C m –2 , respectively, with annual R eco accounting for 89.6% of GPP, of which 68.8% was lost as R below and 20.8% lost as R above , and leaving only 10% of the carbon fixation in ecosystems. Furthermore, these estimates, however, did not consider potential light inhibition of leaf respiration. If we accept the presence of light inhibition, then the daytime three-way partitioning method would underestimate annual R above by 20.4% whereas the nighttime method would overestimate R above by 23.9% and GPP by 4.7%, compared with estimates accounting for light inhibition in leaves.

54 ENVIRONMENTAL SCIENCES↗

Geometric Mapping of Tasks to Processors on Parallel Computers with Mesh or Torus Networks

We present a new method for reducing parallel applications' communication time by mapping their MPI tasks to processors in a way that lowers the distance messages travel and the amount of congestion in the network. Assuming geometric proximity among the tasks is a good approximation of their communication interdependence, we use a geometric partitioning algorithm to order both the tasks and the processors, assigning task parts to the corresponding processor parts. In this way, interdependent tasks are assigned to “nearby” cores in the network. We also present a number of algorithmic optimizations that exploit specific features of the network or application to further improve the quality of the mapping. We specifically address the case of sparse node allocation, where the nodes assigned to a job are not necessarily located in a contiguous block nor within close proximity to each other in the network. Furthermore, our methods generalize to contiguous allocations as well, and results are shown for both contiguous and non-contiguous allocations. We show that, for the structured finite difference mini-application MiniGhost, our mapping methods reduced communication time up to 75% relative to MiniGhost's default mapping on 128K cores of a Cray XK7 with sparse allocation. For the atmospheric modeling code E3SM/HOMME, our methods reduced communication time up to 31% on 16K cores of an IBM BlueGene/Q with contiguous allocation.

97 MATHEMATICS AND COMPUTING↗

Association Between White Blood Cell Count Following Radiation Therapy With Radiation Pneumonitis in Non-Small Cell Lung Cancer

Purpose: Radiation pneumonitis (RP) is an inflammatory response to radiation therapy (RT). We assessed the association between RP and white blood cell (WBC) count, an established metric of systemic inflammation, after RT for non-small cell lung cancer. Methods and Materials: We retrospectively analyzed 366 patients with non-small cell lung cancer who received ≥60 Gy as definitive therapy. The primary endpoint was whether WBC count after RT (defined as 2 weeks through 3 months after RT completion) was associated with grade ≥3 or grade ≥2 RP. Median lung volume receiving ≥20 Gy (V{sub 20}) was 31%, and post-RT WBC counts ranged from 1.7 to 21.2 × 10{sup 3} WBCs/μL. Odds ratios (ORs) associating clinical variables and post-RT WBC counts with RP were calculated via logistic regression. A recursive-partitioning algorithm was used to define optimal post-RT WBC count cut points. Results: Post-RT WBC counts were significantly higher in patients with grade ≥3 RP than without (P<.05). Optimal cut points for post-RT WBC count were found to be 7.4 and 8.0 × 10{sup 3}/μL for grade ≥3 and ≥2 RP, respectively. Univariate analysis revealed significant associations between post-RT WBC count and grade ≥3 (n=46, OR=2.6, 95% confidence interval [CI] 1.4‒4.9, P=.003) and grade ≥2 RP (n=164, OR=2.0, 95% CI 1.2‒3.4, P=.01). This association held in a stepwise multivariate regression. Of note, V{sub 20} was found to be significantly associated with grade ≥2 RP (OR=2.2, 95% CI 1.2‒3.4, P=.01) and trended toward significance for grade ≥3 RP (OR=1.9, 95% CI 1.0-3.5, P=.06). Conclusions: Post-RT WBC counts were significantly and independently associated with RP and have potential utility as a diagnostic or predictive marker for this toxicity.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Performance evaluation of a parallel sparse lattice Boltzmann solver

We develop a performance prediction model for a parallelized sparse lattice Boltzmann solver and present performance results for simulations of flow in a variety of complex geometries. A special focus is on partitioning and memory/load balancing strategy for geometries with a high solid fraction and/or complex topology such as porous media, fissured rocks and geometries from medical applications. The topology of the lattice nodes representing the fluid fraction of the computational domain is mapped on a graph. Graph decomposition is performed with both multilevel recursive-bisection and multilevel k-way schemes based on modified Kernighan-Lin and Fiduccia-Mattheyses partitioning algorithms. Performance results and optimization strategies are presented for a variety of platforms, showing a parallel efficiency of almost 80% for the largest problem size. A good agreement between the performance model and experimental results is demonstrated.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

A Data-Driven Approach to Nation-Scale Building Energy Modeling

In 2019, 125 million U.S. residential and commercial buildings consumed $412 billion in energy bills. These buildings currently consume 40% of the nation's primary energy, 73% of electricity, 80% of energy during peak electric grid use, and responsible for 39% of greenhouse gas emissions [14]. Urban-scale building energy modeling has grown significantly in the past decade, allowing individual campuses or communities of buildings to be modeled, simulated, and cost-effective solutions for intelligent management to be identified and implemented. While traditionally limited to individual counties and usually less than 2,000 buildings, the Automatic Building Energy Modeling (AutoBEM) soft-ware suite has been developed to process unconventional, nation-scale data sources to generate unique OpenStudio and EnergyPlus models of each building. Through the use of High Performance Computing (HPC) resources, every U.S. building has been simulated. This paper showcases the data layout, node partitioning, algorithmic approaches, and analytic results that were used to create, share, and analyze 124.4 million U.S. building models.

Berres, Andy↗

Muster: Massively Scalable Clustering

Muster is a framework for scalable cluster analysis. It includes implementations of classic K-Medoids partitioning algorithms, as well as infrastructure for making these algorithms run scalably on very large systems. In particular, Muster contains algorithms such as CAPEK (described in reference 1) that are capable of clustering highly distributed data sets in-place on a hundred thousand or more processes.

Source record↗

Futility

The Futility package contains the following: 1) Definition of the size of integers and real numbers; 2) A generic Unit test harness; 3) Definitions for some basic extensions to the Fortran language: arbitrary length strings, a parameter list construct, exception handlers, command line processor, timers; 4) Geometry definitions: point, line, plane, box, cylinder, polyhedron; 5) File wrapper functions: standard Fortran input/output files, Fortran binary files, HDF5 files; 6) Parallel wrapper functions: MPI, and Open MP abstraction layers, partitioning algorithms; 7) Math utilities: BLAS, Matrix and Vector definitions, Linear Solver methods and wrappers for other TPLs (PETSC, MKL, etc), preconditioner classes; 8) Misc: random number generator, water saturation properties, sorting algorithms.

Collins, BenjaminS.↗

TAP - Tools for Adaptive Partitioning v. 0.99 Beta

TAP is a set of tools which are essential for conducting research on adaptive partitioners. The basic premise is that a single partitioner may not be a good choice for adaptive mesh simulations; rather one must match a partitioner (obtained from a partitioning package like Zoltan, ParMetis etc) with the mesh being partitioned. TAP provides the tools that can judge the suitability of a partitioning algorithm to a given mesh.

Steensland, Johan↗

Vertically Resolved Convective–Stratiform Echo-Type Identification and Convectivity Retrieval for Vertically Pointing Radars

Using data from the airborne HIAPER Cloud Radar (HCR), a partitioning algorithm (ECCO-V) that provides vertically resolved convectivity and convective versus stratiform radar-echo classification is developed for vertically pointing radars. The algorithm is based on the calculation of reflectivity and radial velocity texture fields that measure the horizontal homogeneity of cloud and precipitation features. The texture fields are translated into convectivity, a numerical measure of the convective or stratiform nature of each data point. The convective–stratiform classification is obtained by thresholding the convectivity field. Subcategories of low, mid-, and high stratiform, shallow, mid-, deep, and elevated convective, and mixed echoes are introduced, which are based on the melting-layer and divergence-level altitudes. As the algorithm provides vertically resolved classifications, it is capable of identifying different types of vertically layered echoes, and convective features that are embedded in stratiform cloud layers. Its robustness was tested on data from four HCR field campaigns that took place in different meteorological and climatological regimes. The algorithm was adapted for use in spaceborne and ground-based radars, proving its versatility, as it is adaptable not only to different radar types and wavelengths, but also different research applications.

54 ENVIRONMENTAL SCIENCES↗

Processing Particle Data Flows with SmartNICs

Many distributed applications implement complex data flows and need a flexible mechanism for routing data between producers and consumers. Recent advances in programmable network interface cards, or SmartNICs, represent an opportunity to offload data-flow tasks into the network fabric, thereby freeing the hosts to perform other work. System architects in this space face multiple questions about the best way to leverage SmartNICs as processing elements in data flows. In this paper, we advocate the use of Apache Arrow as a foundation for implementing data-flow tasks on SmartNICs. We report on our experiences adapting a partitioning algorithm for particle data to Apache Arrow and measure the on-card processing performance for the BlueField-2 SmartNIC. Our experiments confirm that the BlueField-2’s (de)compression hardware can have a significant impact on in-transit workflows where data must be unpacked, processed, and repacked.

97 MATHEMATICS AND COMPUTING↗

TMVOC-MP: a parallel numerical simulator for Three-PhaseNon-isothermal Flows of Multicomponent Hydrocarbon Mixtures inporous/fractured media

TMVOC-MP is a massively parallel version of the TMVOC code (Pruess and Battistelli, 2002), a numerical simulator for three-phase non-isothermal flow of water, gas, and a multicomponent mixture of volatile organic chemicals (VOCs) in multidimensional heterogeneous porous/fractured media. TMVOC-MP was developed by introducing massively parallel computing techniques into TMVOC. It retains the physical process model of TMVOC, designed for applications to contamination problems that involve hydrocarbon fuels or organic solvents in saturated and unsaturated zones. TMVOC-MP can model contaminant behavior under 'natural' environmental conditions, as well as for engineered systems, such as soil vapor extraction, groundwater pumping, or steam-assisted source remediation. With its sophisticated parallel computing techniques, TMVOC-MP can handle much larger problems than TMVOC, and can be much more computationally efficient. TMVOC-MP models multiphase fluid systems containing variable proportions of water, non-condensible gases (NCGs), and water-soluble volatile organic chemicals (VOCs). The user can specify the number and nature of NCGs and VOCs. There are no intrinsic limitations to the number of NCGs or VOCs, although the arrays for fluid components are currently dimensioned as 20, accommodating water plus 19 components that may be either NCGs or VOCs. Among them, NCG arrays are dimensioned as 10. The user may select NCGs from a data bank provided in the software. The currently available choices include O{sub 2}, N{sub 2}, CO{sub 2}, CH{sub 4}, ethane, ethylene, acetylene, and air (a pseudo-component treated with properties averaged from N{sub 2} and O{sub 2}). Thermophysical property data of VOCs can be selected from a chemical data bank, included with TMVOC-MP, that provides parameters for 26 commonly encountered chemicals. Users also can input their own data for other fluids. The fluid components may partition (volatilize and/or dissolve) among gas, aqueous, and NAPL phases. Any combination of the three phases may present, and phases may appear and disappear in the course of a simulation. In addition, VOCs may be adsorbed by the porous medium, and may biodegrade according to a simple half-life model. Detailed discussion of physical processes, assumptions, and fluid properties used in TMVOC-MP can be found in the TMVOC user's guide (Pruess and Battistelli, 2002). TMVOC-MP was developed based on the parallel framework of the TOUGH2-MP code (Zhang et al. 2001, Wu et al. 2002). It uses the MPI (Message Passing Forum, 1994) for parallel implementation. A domain decomposition approach is adopted for the parallelization. The code partitions a simulation domain, defined by an unstructured grid, using partitioning algorithm from the METIS software package (Karypsis and Kumar, 1998). In parallel simulation, each processor is in charge of one part of the simulation domain for assembling mass and energy balance equations, solving linear equation systems, updating thermophysical properties, and performing other local computations. The local linear-equation systems are solved in parallel by multiple processors with the Aztec linear solver package (Tuminaro et al., 1999). Although each processor solves the linearized equations of subdomains independently, the entire linear equation system is solved together by all processors collaboratively via communication between neighboring processors during each iteration. Detailed discussion of the prototype of the data-exchange scheme can be found in Elmroth et al. (2001). In addition, FORTRAN 90 features are introduced to TMVOC-MP, such as dynamic memory allocation, array operation, matrix manipulation, and replacing 'common blocks' (used in the original TMVOC) with modules. All new subroutines are written in FORTRAN 90. Program units imported from the original TMVOC remain in standard FORTRAN 77. This report provides a quick starting guide for using the TMVOC-MP program. We suppose that the users have basic knowledge of using the original TMVOC code. The users can find the detailed technical description of the physical processes modeled, and the mathematical and numerical methods in the user's guide for TMVOC (Pruess and Battistelli, 2002).

58 GEOSCIENCES↗

REFINEMENT OF THE NEPHELINE DISCRIMINATOR: RESULTS OF A PHASE II STUDY

Twenty five glass compositions were selected for a Phase II study to assess the potential for reducing the conservatism in the nepheline discriminator. The glass compositions were restricted to regions that fell within the validation ranges of the DWPF PCCS models. In addition, the liquidus temperature model was used to restrict the glass compositions so that they could all be melted at the same temperature. The nepheline discriminator was used to force the glass compositions into regions where nepheline formation was predicted to occur. The glasses were fabricated in the laboratory and characterized for crystallization and chemical durability after both quenching and slow cooling. Chemical analysis showed that the fabricated glasses met the target compositions. Nepheline was identified in one of the quenched glasses and several of the CCC glasses. There was no clear relationship between the types of crystallization that occurred in a particular glass and its location on the Al{sub 2}O{sub 3}-Na{sub 2}O-SiO{sub 2} ternary diagram. A partitioning algorithm was used to identify trends in crystallization behavior based on glass composition. Generally, for the CCC glasses MnO influenced the crystallization of spinels and B{sub 2}O{sub 3} and SiO{sub 2} influenced the crystallization of nepheline. Measured durability responses varied from acceptable to unacceptable depending on the glass composition and type and extent of crystallization that occurred. It was not possible to identify any linear effects of composition on chemical durability performance for this set of study glasses. The results were not sufficient to recommend modification of the current nepheline discriminator at this time. It is recommended that the next series of experiments continue to focus not only on compositional regions where the PCCS models are considered applicable (i.e., the model validation ranges), but also be restricted to compositional regions where acceptable glasses are predicted to be produced but are disallowed by the current nepheline discriminator. This may help identify a path for gaining access to glass compositions that are limited by only the nepheline constraint. Dependencies on cooling rates should also be evaluated. A better understanding of these regions, as well as the impacts of B{sub 2}O{sub 3} and CaO on nepheline crystallization, may allow for relaxation or refinement of the nepheline constraint and subsequently allow for higher Al{sub 2}O{sub 3} concentrations.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Muster: Massively Scalable Clustering

Muster is a framework for scalable cluster analysis. It includes implementations of classic K-Medoids partitioning algorithms, as well as infrastructure for making these algorithms run scalably on very large systems. In particular, Muster contains algorithms such as CAPEK (described in reference 1) that are capable of clustering highly distributed data sets in-place on a hundred thousand or more processes.

Source record↗