Search NASA⌕ Search

SEARCH · Search NASA

Results for “Parallel algorithms”

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 1,495 records · Page 83

Multigrid with Overlapping Patches

Solving boundary value problems with optimal efficiency requires adaptivity and multilevel techniques. Previously, an implementation of the AFACx algorithm is presented that is based on rectangular Cartesian grids. This implementation does not allow for the over]ap of grids that lie on the same level of refinement. We investigate the case in which these grids overlap. A standard technique for overlapping grids is the Schwarz algorithm. Some ways of using the Schwarz algorithm in a standard multigrid scheme are presented. Also, a problem that arises in some situations with non-aligned, overlapping grids is described. This situation comes up in a natural way when the Schwarz algorithm is used as a relaxation scheme within a multilevel algorithm. We identify the reason for the bad convergence and show that by more sophisticated interpolation the difficulties can be overcome. Then we present a multiplicative Schwarz algorithm for a large number of grids that has a high potential for parallelization. Finally we give some numerical results for the FACx algorithm with overlapping grids on each refinement level. The implementation of the described codes uses C++ and the array class libraries A++ and P++. Using the A++/P++ programming environment, it was possible to move from a serial code to a parallel code within a few days.

Berndt, Markus↗

A spectral scheme for Kohn–Sham density functional theory of clusters

Starting from the observation that one of the most successful methods for solving the Kohn–Sham equations for periodic systems – the plane-wave method – is a spectral method based on eigenfunction expansion, we formulate a spectral method designed towards solving the Kohn–Sham equations for clusters. This allows for efficient calculation of the electronic structure of clusters (and molecules) with high accuracy and systematic convergence properties without the need for any artificial periodicity. The basis functions in this method form a complete orthonormal set and are expressible in terms of spherical harmonics and spherical Bessel functions. Computation of the occupied eigenstates of the discretized Kohn–Sham Hamiltonian is carried out using a combination of preconditioned block eigensolvers and Chebyshev polynomial filter accelerated subspace iterations. Several algorithmic and computational aspects of the method, including computation of the electrostatics terms and parallelization are discussed. We have implemented these methods and algorithms into an efficient and reliable package called ClusterES (Cluster Electronic Structure). A variety of benchmark calculations employing local and non-local pseudopotentials are carried out using our package and the results are compared to the literature. Convergence properties of the basis set are discussed through numerical examples. Computations involving large systems that contain thousands of electrons are demonstrated to highlight the efficacy of our methodology. The use of our method to study clusters with arbitrary point group symmetries is briefly discussed.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Motion detection in astronomical and ice floe images

Two approaches are presented for establishing correspondence between small areas in pairs of successive images for motion detection. The first one, based on local correlation, is used on a pair of successive Voyager images of the Jupiter which differ mainly in locally variable translations. This algorithm is implemented on a sequential machine (VAX 780) as well as the Massively Parallel Processor (MPP). In the case of the sequential algorithm, the pixel correspondence or match is computed on a sparse grid of points using nonoverlapping windows (typically 11 x 11) by local correlations over a predetermined search area. The displacement of the corresponding pixels in the two images is called the disparities to cubic surfaces. The disparities at points where the error between the computed values and the surface values exceeds a particular threshold are replaced by the surface values. A bilinear interpolation is then used to estimate disparities at all other pixels between the grid points. When this algorithm was applied at the red spot in the Jupiter image, the rotating velocity field of the storm was determined. The second method of motion detection is applicable to pairs of images in which corresponding areas can experience considerable translation as well as rotation.

Manohar, M.↗

Multi-core and Many-core Shared-memory Parallel Raycasting Volume Rendering Optimization and Tuning

Given the computing industry trend of increasing processing capacity by adding more cores to a chip, the focus of this work is tuning the performance of a staple visualization algorithm, raycasting volume rendering, for shared-memory parallelism on multi-core CPUs and many-core GPUs. Our approach is to vary tunable algorithmic settings, along with known algorithmic optimizations and two different memory layouts, and measure performance in terms of absolute runtime and L2 memory cache misses. Our results indicate there is a wide variation in runtime performance on all platforms, as much as 254% for the tunable parameters we test on multi-core CPUs and 265% on many-core GPUs, and the optimal configurations vary across platforms, often in a non-obvious way. For example, our results indicate the optimal configurations on the GPU occur at a crossover point between those that maintain good cache utilization and those that saturate computational throughput. This result is likely to be extremely difficult to predict with an empirical performance model for this particular algorithm because it has an unstructured memory access pattern that varies locally for individual rays and globally for the selected viewpoint. Our results also show that optimal parameters on modern architectures are markedly different from those in previous studies run on older architectures. And, given the dramatic performance variation across platforms for both optimal algorithm settings and performance results, there is a clear benefit for production visualization and analysis codes to adopt a strategy for performance optimization through auto-tuning. These benefits will likely become more pronounced in the future as the number of cores per chip and the cost of moving data through the memory hierarchy both increase.

97 MATHEMATICS AND COMPUTING↗

A Pervasive Parallel Processing Framework for Data Visualization and Analysis at Extreme Scale

Most of today’s visualization libraries and applications are based off of what is known today as the visualization pipeline. In the visualization pipeline model, algorithms are encapsulated as “filtering” components with inputs and outputs. These components can be combined by connecting the outputs of one filter to the inputs of another filter. The visualization pipeline model is popular because it provides a convenient abstraction that allows users to combine algorithms in powerful ways. Unfortunately, the visualization pipeline cannot run effectively on exascale computers. Experts agree that the exascale machine will comprise processors that contain many cores. Furthermore, physical limitations will prevent data movement in and out of the chip (that is, between main memory and the processing cores) from keeping pace with improvements in overall compute performance. To use these processors to their fullest capability, it is essential to carefully consider memory access. This is where the visualization pipeline fails. Each filtering component in the visualization library is expected to take a data set in its entirety, perform some computation across all of the elements, and output the complete results. The process of iterating over all elements must be repeated in each filter, which is one of the worst possible ways to traverse memory when trying to maximize the number of executions per memory access. This project investigates a new type of visualization framework that exhibits a pervasive parallelism necessary to run on exascale machines. Our framework achieves this by defining algorithms in terms of functors, which are localized, stateless operations. Functors can be composited in much the same way as filters in the visualization pipeline. But, functors’ design allows them to be concurrently running on massive amounts of lightweight threads. Only with such fine-grained parallelism can we hope to fill the billions of threads we expect will be necessary for efficient computation on an exascale computer. This project concludes with a functional prototype containing pervasively parallel algorithms that perform demonstratively well on many-core processors. These algorithms are fundamental for performing data analysis and visualization at extreme scale.

97 MATHEMATICS AND COMPUTING↗

SLOAVx: Scalable LOgarithmic AlltoallV Algorithm for Hierarchical Multicore Systems

Scientific applications use collective communication operations in Message Passing Interface (MPI) for global synchronization and data exchanges. Alltoall and AlltoallV are two important collective operations. They are used by MPI jobs to exchange messages among all of MPI processes. AlltoallV is a generalization of Alltoall, supporting messages of varying sizes. However, the existing MPI AlltoallV implementation has linear complexity, i.e., each process has to send messages to all other processes in the job. Such linear complexity can result in sub optimal scalability of MPI applications when they are deployed on millions of cores. To address above challenge, in this paper, we introduce a new Scalable LOgarithmic AlltoallV algorithm, named SLOAV, for MPI AlltoallV collective operation. SLOAV aims to achieve global exchange of small messages of different sizes in a logarithmic number of rounds. Furthermore, given the prevalence of multicore systems with shared memory, we design a hierarchical AlltoallV algorithm based on SLOAV by leveraging the advantages of shared memory, which is referred to as SLOAVx. Compared to SLOAV, SLOAVx significantly reduces the inter-node communication, thus improving the entire system performance and mitigating the impact of message latency. We have implemented and embedded both algorithms in Open MPI. Our evaluation on large-scale computer systems shows that for the 8-byte and 1024-process MPI Alltoallv operation, the SLOAV can reduce the latency by as much as 86.4%, when compared to the state-of-the-art, and SLOAVx can further optimize the SLOAV by up to 83.1% in terms of message latency on multicore systems. In addition, experiments with NAS Parallel Benchmark (NPB) demonstrate that our algorithms are very effective for real-world applications.

Computer Science↗

Parallel Memory-Independent Communication Bounds for SYRK

In this paper, we focus on the parallel communication cost of multiplying a matrix with its transpose, known as a symmetric rank-k update (SYRK). SYRK requires half the computation of general matrix multiplication because of the symmetry of the output matrix. Recent work (Beaumont et al., SPAA '22) has demonstrated that the sequential I/O complexity of SYRK is also a constant factor smaller than that of general matrix multiplication. Inspired by this progress, we establish memory-independent parallel communication lower bounds for SYRK with smaller constants than general matrix multiplication, and we show that these constants are tight by presenting communication-optimal algorithms. The crux of the lower bound proof relies on extending a key geometric inequality to symmetric computations and analytically solving a constrained nonlinear optimization problem. Here, the optimal algorithms use a triangular blocking scheme for parallel distribution of the symmetric output matrix and corresponding computation.

Communication costs↗

Microprocessor arrays for large scale computation

An important new direction in computer architecture centers around the achievement of very high computational power (capacity, speed and reliability) through the use of tens of thousands of microprocessors, micromemories, and switch modules, all interconnected into a large homogeneous network using one of certain advanced connection schemes. When surrounded and supported by conventional computers and memories, such a machine holds potential for out-performing both conventional and array-based computers of the mid-1980's by one to two orders of magnitude, at least for particular classes of applications amenable to high parallelism, such as aerodynamic simulation. The homogeneous feature of this machine concept also implies size extendability, fault tolerance, and improved flexibility to handle a variety of algorithms of interest. Current work is addressing the design of technologically efficient interconnection configurations and the development of new computation algorithms that are especially efficient for highly parallel computation.

Kautz, W. H.↗

A note on parallel and pipeline computation of fast unitary transforms

The parallel and pipeline organization of fast unitary transform algorithms such as the Fast Fourier Transform are discussed. The efficiency is pointed out of a combined parallel-pipeline processor of a transform such as the Haar transform in which 2 to the n minus 1 power hardware butterflies generate a transform of order 2 to the n power every computation cycle.

Fino, B. J.↗

Boundary element analysis on vector and parallel computers

Boundary element analysis (BEA) can be characterized as a numerical technique that generally shifts the computational burden in the analysis toward numerical integration and the solution of nonsymmetric and either dense or blocked sparse systems of algebraic equations. Researchers have explored the concept that the fundamental characteristics of BEA can be exploited to generate effective implementations on vector and parallel computers. In this paper, the results of some of these investigations are discussed. The performance of overall algorithms for BEA on vector supercomputers, massively data parallel single instruction multiple data (SIMD), and relatively fine grained distributed memory multiple instruction multiple data (MIMD) computer systems is described. Some general trends and conclusions are discussed, along with indications of future developments that may prove fruitful in this regard.

Kane, J. H.↗

The Simplified Aircraft-Based Paired Approach With the ALAS Alerting Algorithm

This paper presents the results of an investigation of a proposed concept for closely spaced parallel runways called the Simplified Aircraft-based Paired Approach (SAPA). This procedure depends upon a new alerting algorithm called the Adjacent Landing Alerting System (ALAS). This study used both low fidelity and high fidelity simulations to validate the SAPA procedure and test the performance of the new alerting algorithm. The low fidelity simulation enabled a determination of minimum approach distance for the worst case over millions of scenarios. The high fidelity simulation enabled an accurate determination of timings and minimum approach distance in the presence of realistic trajectories, communication latencies, and total system error for 108 test cases. The SAPA procedure and the ALAS alerting algorithm were applied to the 750-ft parallel spacing (e.g., SFO 28L/28R) approach problem. With the SAPA procedure as defined in this paper, this study concludes that a 750-ft application does not appear to be feasible, but preliminary results for 1000-ft parallel runways look promising.

Perry, Raleigh B.↗

Faster approximate subgraph counts with privacy

One of the most common problems studied in the context of differential privacy for graph data is counting the number of non-induced embeddings of a subgraph in a given graph. These counts have very high global sensitivity. Therefore, adding noise based on powerful alternative techniques, such as smooth sensitivity and higher-order local sensitivity have been shown to give significantly better accuracy. However, all these alternatives to global sensitivity become computationally very expensive, and to date efficient polynomial time algorithms are known only for few selected subgraphs, such as triangles, k-triangles, and k-stars. In this paper, we show that good approximations to these sensitivity metrics can be still used to get private algorithms. Using this approach, we much faster algorithms for privately counting the number of triangles in real-world social networks, which can be easily parallelized. We also give a private polynomial time algorithm for counting any constant size subgraph using less noise than the global sensitivity; we show this can be improved significantly for counting paths in special classes of graphs

Nguyen, Dung↗

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↗

Computation of Earth Science Products on Spaceborne Platforms

Spaceborne sensors like NASA's Hyperion hyperspectral imager generate huge data volumes, and several near-term trends indicate that data volumes will only increase. Next-generation hyperspectral missions, such as NASA's Hyperspectral Infrared Imager (HyspIRI), will operate at higher duty cycles and higher data rates, and their users will expect products to be generated from the data in near real time [1]. Barring a sudden advance in satellite downlink capacity, these trends point to a need to process data and generate products onboard the spacecraft. Rather than downlink an entire hyperspectral image cube, onboard processing enables satellites to downlink partial or completed scientific data products, which are often one to two orders of magnitude smaller than the original image. In addition, a satellite with onboard data processing resources and direct broadcast transmission equipment could send data products directly to first responders, research scientists or other users on the ground. Next-generation space-capable data processors will have a combination of reconfigurable gate arrays, digital signal processors and general-purpose CPUs. Correctly programmed and configured, these resources are sufficient to run sophisticated data analysis programs, including hyperspectral image processing algorithms that commonly run on desktop computers [2]. This paper describes how we implemented one such program, the HSEG hierarchical image segmentation algorithm, software commonly used on desktop and parallel processors, on a hardware platform designed to mimic a next-generation space-capable data processor [3]. We also describe our approach to porting the algorithm to and optimizing it for the new platform, and determine the expected performance gains enabled by our design. This extended abstract will describe the HSEG algorithm and hardware platform in greater detail, provide an analysis of the key function within the algorithm that required hardware acceleration, and describe our implementation of that function in hardware.

Fisher, Kevin↗

Energy Proportionality and Performance in Data Parallel Computing Clusters

Energy consumption in datacenters has recently become a major concern due to the rising operational costs andscalability issues. Recent solutions to this problem propose the principle of energy proportionality, i.e., the amount of energy consumedby the server nodes must be proportional to the amount of work performed. For data parallelism and fault tolerancepurposes, most common file systems used in MapReduce-type clusters maintain a set of replicas for each data block. A coveringset is a group of nodes that together contain at least one replica of the data blocks needed for performing computing tasks. In thiswork, we develop and analyze algorithms to maintain energy proportionality by discovering a covering set that minimizesenergy consumption while placing the remaining nodes in lowpower standby mode. Our algorithms can also discover coveringsets in heterogeneous computing environments. In order to allow more data parallelism, we generalize our algorithms so that itcan discover k-covering sets, i.e., a set of nodes that contain at least k replicas of the data blocks. Our experimental results showthat we can achieve substantial energy saving without significant performance loss in diverse cluster configurations and workingenvironments.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Scalable and fast heterogeneous molecular simulation with predictive parallelization schemes

Multiscale and inhomogeneous molecular systems are challenging topics in the field of molecular simulation. In particular, modeling biological systems in the context of multiscale simulations and exploring material properties are driving a permanent development of new simulation methods and optimization algorithms. In computational terms, those methods require parallelization schemes that make a productive use of computational resources for each simulation and from its genesis. Here, we introduce the heterogeneous domain decomposition approach, which is a combination of an heterogeneity-sensitive spatial domain decomposition with an a priori rearrangement of subdomain walls. Within this approach and paper, the theoretical modeling and scaling laws for the force computation time are proposed and studied as a function of the number of particles and the spatial resolution ratio. We also show the new approach capabilities, by comparing it to both static domain decomposition algorithms and dynamic load-balancing schemes. Specifically, two representative molecular systems have been simulated and compared to the heterogeneous domain decomposition proposed in this work. Finally, these two systems comprise an adaptive resolution simulation of a biomolecule solvated in water and a phase-separated binary Lennard-Jones fluid.

36 MATERIALS SCIENCE↗

Multi-core and many-core shared-memory parallel raycasting volume rendering optimization and tuning

Given the computing industry trend of increasing processing capacity by adding more cores to a chip, the focus of this work is tuning the performance of a staple visualization algorithm, raycasting volume rendering, for shared-memory parallelism on multi-core CPUs and many-core GPUs. Our approach is to vary tunable algorithmic settings, along with known algorithmic optimizations and two different memory layouts, and measure performance in terms of absolute runtime and L2 memory cache misses. Our results indicate there is a wide variation in runtime performance on all platforms, as much as 254% for the tunable parameters we test on multi-core CPUs and 265% on many-core GPUs, and the optimal configurations vary across platforms, often in a non-obvious way. For example, our results indicate the optimal configurations on the GPU occur at a crossover point between those that maintain good cache utilization and those that saturate computational throughput. This result is likely to be extremely difficult to predict with an empirical performance model for this particular algorithm because it has an unstructured memory access pattern that varies locally for individual rays and globally for the selected viewpoint. Our results also show that optimal parameters on modern architectures are markedly different from those in previous studies run on older architectures. In addition, given the dramatic performance variation across platforms for both optimal algorithm settings and performance results, there is a clear benefit for production visualization and analysis codes to adopt a strategy for performance optimization through auto-tuning. These benefits will likely become more pronounced in the future as the number of cores per chip and the cost of moving data through the memory hierarchy both increase.

97 MATHEMATICS AND COMPUTING↗

Implementation of a fully-balanced periodic tridiagonal solver on a parallel distributed memory architecture

While parallel computers offer significant computational performance, it is generally necessary to evaluate several programming strategies. Two programming strategies for a fairly common problem - a periodic tridiagonal solver - are developed and evaluated. Simple model calculations as well as timing results are presented to evaluate the various strategies. The particular tridiagonal solver evaluated is used in many computational fluid dynamic simulation codes. The feature that makes this algorithm unique is that these simulation codes usually require simultaneous solutions for multiple right-hand-sides (RHS) of the system of equations. Each RHS solutions is independent and thus can be computed in parallel. Thus a Gaussian elimination type algorithm can be used in a parallel computation and the more complicated approaches such as cyclic reduction are not required. The two strategies are a transpose strategy and a distributed solver strategy. For the transpose strategy, the data is moved so that a subset of all the RHS problems is solved on each of the several processors. This usually requires significant data movement between processor memories across a network. The second strategy attempts to have the algorithm allow the data across processor boundaries in a chained manner. This usually requires significantly less data movement. An approach to accomplish this second strategy in a near-perfect load-balanced manner is developed. In addition, an algorithm will be shown to directly transform a sequential Gaussian elimination type algorithm into the parallel chained, load-balanced algorithm.

Eidson, T. M.↗