Multicomputer Research Articles

Multiresolution Preconditioners for Solving Realistic Multi-Scale Complex Problems

In this work, the hierarchic multiresolution (MR) preconditioner is combined with the multilevel fast multipole algorithm-fast Fourier transform (MLFMA-FFT) and efficiently parallelized in multicore computers for computing electromagnetic scattering and radiation from complex problems exhibiting deep multi-scale features. The problem is formulated using the thin-dielectric-sheet (TDS) approximation for thin dielectric materials and the electric and combined field integral equations (EFIE/CFIE) for conducting objects. The parallel MLFMA-FFT is tailored to accommodate the MR hierarchical functions, which provide vast improvement of the matrix system conditioning by accurately handling multi-scale mesh features in different levels of detail. The higher (coarser) level hierarchical functions are treated by an algebraic incomplete LU decomposition preconditioner, which has been efficiently embedded into the parallel framework to further accelerate the solution. Numerical examples are presented to demonstrate the precision and efficiency of the proposed approach for the solution of realistic multi-scale scattering and radiation problems.

Оптимизация выполнения целочисленных арифметических операций на основе нейронных сетей

Nowadays, the main methods of improving the performance of arithmetic operations on fixed and floating numbers are performed by multicore computer systems (CS) are the improvement of the microarchitecture of systems and algorithms implemented in compilers and libraries. Separately, it is necessary to highlight the complex tasks of implementing operations with long numbers that exceed the size of a machine word. This is due to the limited size of the processor registers and the data bus. As a result, the implementation of traditional operations, such as addition, multiplication, division, become more difficult to implement in modern computer systems. To solve this problem, there are currently several algorithms. However, the available algorithms can be quite slow, since they are sequential and implement bitwise operations on the bits of a number.In order to improve the performance of this class of calculations, a model based on a neural network is proposed. The purpose of this model is to reduce the execution time and CPU usage. A neural network model for performing arithmetic operations has been developed. A method of generating a data set based on graph traversal is proposed and tested experimentally. Algorithms for performing arithmetic operations based on a pretrained neural network are constructed.

Prediction of multicore CPU performance through parallel data mining on public datasets

In the present scenario, high-performance computing needs more attention towards multicore computing. While designing the CPU, we need to consider hardware for processing speed, cache bandwidth, minimum memory requirements, etc. So for the selection of the best combination of CPU data mining tools may play an important role. In this era, data mining attracts more interest on parallel computing to enhance the performance of multicores. This paper demonstrates a parallel strategy similar to the traditional parallel programming paradigm to improve the performance of multicores by using a data mining approach. We have selected one approach EM with Gaussian and analyze the impact of its parallel execution on selected multicore clusters obtained through data mining. We have also evaluated our finding with a virtual environment of having 32 different families of CPUs, showing a speedup of up to ≈1.02x. This paper considers data clusters, cache mapping techniques and ranking of MPI programming techniques.

A virtual procedure for real-time monitoring of intervisibility between conflicting agents at intersections using point cloud and trajectory data

A new procedure is developed for effectively monitoring intervisibility between conflicting agents in real-time at intersections. Dense light detection and ranging (Lidar) point clouds and time-stamped trajectory data are used to model static intersection environment and emulate traffic participants’ dynamic motion states, respectively. The proposed procedure reads trajectory data in sequence according to their timestamps. An agent-based approach that enables the application of multi-core parallel computing is applied to estimate conflict points and identify conflicting agents in pairs. Then, a linear elevation array is created from the Lidar point cloud data, based on which the elevation of each trajectory point is obtained in a real-time manner. Meanwhile, three-dimensional bounding cuboids are generated at each path point to represent digital twins of agents. Once a pair of conflicting agents are identified, a hybrid approach is triggered to examine whether the agents’ line-of-sights are occluded by either static (e.g., tree trunk) or dynamic obstacles (e.g., cars). Accordingly, virtual warning signals can be generated. The effectiveness of the procedure is demonstrated through controlled experiments. The procedure was also tested in two virtual scenarios. The mean processing time at each frame is less than 0.1 s as achieved with limited computational power. With implementation of parallel computing technique, processing time is not sensitive to number of agents within the intersection. In addition, by enabling the outputs of virtual warning signals, spatial distribution of conflict points, and individual conflict-related time series data, the procedure shall help provide substantial insights into intersection safety.

Succinct parallel Lempel–Ziv factorization on a multicore computer

Succinct parallel Lempel–Ziv factorization on a multicore computer

Pros and cons of accelerated computing on T-phase propagation simulation with a 3D parabolic equation model

Submarine seismic activities can generate low frequency (<40 Hz) acoustic waves, known as T-phases, that propagate in the ocean with low attenuation for thousands of km. In this regard, high-performance computing is essential to simulate the propagation from an earthquake source to a receiver. This work discusses the pros and cons of accelerated computing on the simulation of T-phase propagation using a 3D geodesic Cartesian parabolic equation (PE) model. The approximately 15 000 km path of the T-phases generated in the Kermadec Trench (Pacific Ocean) to the Ascension Island (Atlantic Ocean) is simulated. This simulation requires an approximate 11 × 2000 × 15000 km3 (depth × cross-range × range) simulation domain to evaluate reflections and diffractions correctly. In general, this type of 3DPE long-range marching solution simulation in GPU is advantageous and up to 20 times faster than in sequential computation, while eight times faster than in multi-core computation using 28 cores. However, when high-resolution cross-range discretization (<50 m) is used, the GPU memory limits the computation. In these cases, the multi-core computation can become a better option to perform simulations. Numerical results revealed the importance of accelerated computing to understand 3D effects on T-phases passing the Drake Passage.

HIGH-PERFORMANCE MODELING, IDENTIFICATION AND ANALYSIS OF HETEROGENEOUS ABNORMAL NEUROLOGICAL MOVEMENT’S PARAMETERS BASED ON COGNITIVE NEURO FEEDBACK-INFLUENCES

The technique is based on a hybrid model of the neuro-system (Brain cortex nodes and tremor-object), which describes on the basis of wave signal propagation the state and behavior of tremor-objects T, namely the segmental de-scription of 3D elements of trajectories of anormal neurological movements of the studied tremor-objects (limb of the hand) taking into account the matrix of cogni-tive influences of groups of cortex neuro-nodes. The rapid analytical solution of the model as a vector function that describes the 3D elements of the trajectories at each movements segment are constructed using the hybrid integral Fourier’s transformations and hybrid spectral function. The main element of the solution is the adaptive infuences matrix that determines the state parameters of the action of certain groups of brain neuro- cortex. Models and methods of multivariable identification are being developed to investigate their neuro-feedback, which suggest high-speed parallel computations on multicore computers. This model-ing technology consider as a scientific basis for designing inelidgence information systems of the quality medical diagnostic i of critical neurological diseases.Key words: Computer simulation, Software system, High-performance compu-ting, Tremor diseases, Modeling of objects and processes, Multi-parameter iden-tification.

On the Efficiency of Algorithms with Multi-level Parallelism

The paper investigates the efficiency of algorithms for solving computational mathematics problems that use a multilevel model of parallel computing on heterogeneous computer systems. A methodology for estimating the acceleration of algorithms for computers using a multilevel model of parallel computing is proposed. As an example, the parallel algorithm of the iteration method on a subspace for solving the generalized algebraic problem of eigenvalues of symmetric positive definite matrices of sparse structure is considered. For the presented algorithms, estimates of acceleration coefficients and efficiency were obtained on computers of hybrid architecture using graphics accelerators, on multi-core computers with shared memory and multi-node computers of MIMD-architecture.

A non-integer step ratio parallel computing method for hybrid finite-element and discrete-element on multi-core computer

Commercial finite-element and discrete-element software usually adopts a unified computational time step and limit the number of our computer cores. In the face of large-scale with refined scale computing models, this will undoubtedly increase the time cost. The multi time step method of integer step ratio is a method to deal with this kind of problem. A non-integer ratio multi-time step computing method based on domain decomposition and subcycling method is proposed and applied to the finite-element and discrete-element engineering problems. The whole model is divided into finite-element method (FEM) subdomains and discrete-element method (DEM) subdomains according to the load balance of CPU. A coupling region between adjacent domains consists of multiple boundary nodes and particles based on the non-integer step ratio. Boundary data in coupling region are transferred through MPI communication protocol. This method can make full use of the advantages of multi-core computer and improve the calculation efficiency of engineering software.

Python Parallel Processing and Multiprocessing: A Rivew

Parallel and multiprocessing algorithms break down significant numerical problems into smaller subtasks, reducing the total computing time on multiprocessor and multicore computers. Parallel programming is well supported in proven programming languages such as C and Python, which are well suited to “heavy-duty” computational tasks. Historically, Python has been regarded as a strong supporter of parallel programming due to the global interpreter lock (GIL). However, times have changed. Parallel programming in Python is supported by the creation of a diverse set of libraries and packages. This review focused on Python libraries that support parallel processing and multiprocessing, intending to accelerate computation in various fields, including multimedia, attack detection, supercomputers, and genetic algorithms. Furthermore, we discussed some Python libraries that can be used for this purpose.

Fresa: A Plant Propagation Algorithm for Black-Box Single and Multiple Objective Optimization

Fresa implements a nature inspired plant propagation algorithm for the solution of single and multiple objective optimization problems. The method is population based and evolutionary. Treating the objective function as a black box, the implementation is able to solve problems exhibiting behaviour that is challenging for mathematical programming methods. Fresa is easily adapted to new problems which may benefit from bespoke representations of solutions by taking advantage of the dynamic typing and multiple dispatch capabilities of the Julia language. Further, the support for threads in Julia enables an efficient implementation on multi-core computers.

Parallel Execution on HeterogeneousMultiprocessors from Algorithm Models Based on Petri Nets

Current supercomputers are composed by nodes containing a combination of general purpose computing units (CPUs) and specific mathematical coprocessors. In this way, GPGPUs or Xeon Phi cards are attached to the nodes to improve its performance. Both types of processors, CPUs and coprocessors, have many differences, like their architecture, the clock rate of the processors and the operation of the related memory. These are the main factors that conform an heterogeneous multiprocessor. A parallel program that wants to achieve the sum of the performance of both types of processors, must consider not only the complexity of the parallel algorithm, but also the differences in the architecture of processors, increasing its complexity. As a contribution on this problem, this paper presents a model of parallel execution based on Petri Nets, called PN-PEM, that is used not only to model a parallel algorithm, but also to execute it directly on a computer with heterogeneous multiprocessors. An asynchronous execution of tasks and a dynamic scheduler are the main characteristics that allow execute a parallel program on this type of parallel computer. Tests done on a multicore computer with two Xeon Phi cards reach the aggregate performance of both type of processors, confirming the quality of the model used.

Parallel Scheduling of Grid Jobs on Quadcore Systems using Grouping Methods

As Grid computing continues to make inroads into different spheres of our lives and multicore computers become ubiquitous, the need to leverage the gains of multicore computers for the scheduling of Grid jobs becomes a necessity. Most Grid schedulers remain sequential in nature and are inadequate in meeting up with the growing data and processing need of the Grid. Also, the leakage of Moore’s dividend continues as most computing platforms still depend on the underlying hardware for increased performance. Leveraging the Grid for the data challenge of the future requires a shift away from the traditional sequential method. This work extends the work of [1] on a quadcore system. A random method was used to group machines and the total processing power of machines in each group was computed, a size proportional to speed method is then used to estimates the size of jobs for allocation to machine groups. The MinMin scheduling algorithm was implemented within the groups to schedule a range of jobs while varying the number of groups and threads. The experiment was executed on a single processor system and on a quadcore system. Significant improvement was achieved using the group method on the quadcore system compared to the ordinary MinMin on the quadcore. We also find significant performance improvement with increasing groups. Thirdly, we find that the MinMin algorithm also gained marginally from the quadcore system meaning that it is also scalable.

A parallel algorithm for ridge-penalized estimation of the multivariate exponential family from data of mixed types

Computationally efficient evaluation of penalized estimators of multivariate exponential family distributions is sought. These distributions encompass among others Markov random fields with variates of mixed type (e.g., binary and continuous) as special case of interest. The model parameter is estimated by maximization of the pseudo-likelihood augmented with a convex penalty. The estimator is shown to be consistent. With a world of multi-core computers in mind, a computationally efficient parallel Newton–Raphson algorithm is presented for numerical evaluation of the estimator alongside conditions for its convergence. Parallelization comprises the division of the parameter vector into subvectors that are estimated simultaneously and subsequently aggregated to form an estimate of the original parameter. This approach may also enable efficient numerical evaluation of other high-dimensional estimators. The performance of the proposed estimator and algorithm are evaluated and compared in a simulation study. Finally, the presented methodology is applied to data of an integrative omics study.

Parallel finite element solver for multi-core computers with shared memory

A direct finite element solver is considered for symmetric sparse matrices arising from the application of the finite element method to problems of structural mechanics and solid mechanics. The solver is based on the block Cholesky method generalized to indefinite matrices. A distinctive feature of the proposed method from other known approaches is the original algorithm for parallelizing the factorization procedure, as well as assembling the matrix of a system of linear and linearized algebraic equations, based on the analysis of the adjacency graph for the finite elements of the design model. All the proposed parallelization algorithms use dynamic mapping of computational tasks to threads and are based on a general idea that uses a dependency vector that controls the execution of computational tasks in a parallel region. This approach is relatively simple and at the same time demonstrates high efficiency. It was developed for solving nonlinear static and dynamic problems of structural mechanics, requiring multiple assembly and factorization of a sparse matrix, but can be successfully applied in other areas of computational mathematics.

ANumerical Uncertainty in Parallel Processing Using Computational Fluid Dynamics as Example

Computational methods such as Computational Fluid Dynamics (CFD) traditionally yield a single output – a single number that is much like the result one would get if one were to perform a theoretical hand calculation. However, this paper will show that computation methods have inherent uncertainty which can also be reported statistically. In numerical computation, because many factors affect the data collected, the data can be quoted in terms of standard deviations (error bars) along with a mean value to make data comparison meaningful. In cases where two data sets are obscured by uncertainty, the two data sets are said to be indistinguishable. A sample CFD problem pertaining to external aerodynamics is copied and ran on 29 identical computers in a university computer lab. The expectation is that all 29 runs should return exactly the same result; unfortunately, in a few cases the result turns out to be different. This is attributed to the parallelization scheme which partitions the mesh to run in parallel on multiple cores of the computer. The distribution of the computational load is hardware-driven depending on the available resource of each computer at the time. Things, such as load-balancing among multiple Central Processing Unit (CPU) cores using Message Passing Interface (MPI) are transparent to the user. Software algorithm such as METIS or JOSTLE is used to automatically divide up the load between different processors. As such, the user has no control over the outcome of the CFD calculation even when the same problem is computed. Because of this, numerical uncertainty arises from parallel (multicore) computing. One way to resolve this issue is to compute problems using a single core, without mesh repartitioning. However, as this paper demonstrates even this is not straight forward. Keywords: numerical uncertainty, parallelization, load-balancing, automotive aerodynamics

Parallel Computation of Discrete Orthogonal Moment on Block Represented Images Using OpenMP

Herein, a parallel implementation of Discrete Orthogonal moments on block represented images is investigated. Moments and moment functions have been used widely as features for image analysis and pattern recognition tasks. The main disadvantage of all moment sets, is the high computational cost which is increased as higher-order moments are involved in the computations. In image block representation (IBR) the image is represented by homogeneous areas which are called blocks. The IBR allows moment computation with zero computational error for binary images, low computational error for gray images, low computational complexity, while can achieve high processing rates. The results from parallel implementation on a multicore computer using OpenMP, exhibit significant performance.

Automated picking of amyloid fibrils from cryo-EM images for helical reconstruction with RELION

Cryogenic electron microscopy (cryo-EM) is an important tool for determining the molecular structure of proteins and protein assemblies, including helical assemblies such as amyloid fibrils. In reconstruction of amyloid fibril structures from cryo-EM images, an important early step is the selection of fibril locations. This fibril picking step is typically done by hand, a tedious process when thousands of images need to be analyzed. Here we present a computer program called FibrilFinder that identifies the locations and directions of fibril segments in cryo-EM images, by using the properties that the fibrils should be linear objects and have widths within a specified range. The program outputs the fibril locations in text files compatible with the RELION density reconstruction program. After RELION is used to extract the particle image boxes contained in the fibril segments identified by FibrilFinder, a second program called FibrilFixer removes boxes that contain more than one fibril, for instance because two fibrils cross each other. As concrete and realistic examples, we describe the application of the two programs to cryo-EM images of two different amyloid fibrils, namely 40-residue amyloid-β fibrils derived from human brain tissue by seeded growth and fibrils formed by the C-terminal half of the low-complexity domain of the RNA-binding protein FUS. Both examples of amyloid fibrils can be picked from cryo-EM images using the same set of FibrilFinder and FibrilFixer parameters, showing that this software does not require re-optimization for each sample. A set of 1337 cryo-EM images was analyzed in 17 min with one multi-core computer. The new fibril picking software should enable the rapid analysis and comparison of more helical structures using cryo-EM, and perhaps serve as part of the greater automation of the entire structure determination process.

Asynchronous Parallel, Sparse Approximated SVRG for High-Dimensional Machine Learning

With the increasing of the data size and the development of multi-core computers, asynchronous parallel stochastic optimization algorithms such as KroMagnon have gained significant attention. In this paper, we propose a new Sparse approximation and asynchronous parallel Stochastic Variance Reduced Gradient (SSVRG) method for sparse and high-dimensional machine learning problems. Unlike standard SVRG and its asynchronous parallel variant, KroMagnon, the snapshot point of SSVRG is set to the average of all the iterates in the previous epoch, which allows it to take much larger learning rates and also makes it more robust to the choice of learning rates. In particular, we use the sparse approximation of the popular SVRG estimator to perform completely sparse updates at all iterations. Therefore, SSVRG has a much lower per-iteration computational cost than its dense counterpart, SVRG++, and is very friendly to asynchronous parallel implementation. Moreover, we provide the convergence guarantees of SSVRG for both strongly convex and non-strongly convex problems, while existing asynchronous algorithms (e.g., KroMagnon and ASAGA) only have convergence guarantees for strongly convex problems. Finally, we extend SSVRG to non-smooth and asynchronous parallel settings. Numerical experimental results demonstrate that SSVRG converges significantly faster than the state-of-the-art asynchronous parallel methods, e.g., KroMagnon, and is usually more than three orders of magnitude faster than SVRG++.

Parallel spreadsheet evaluation and dynamic cycle detection

SummaryIt was estimated there would be 72 million users using spreadsheets monthly in 2017 some of which build complex financial, scientific, and mathematical models. Most of these end‐users are not trained IT professionals but domain experts. In the age of multicore computing and ever‐increasing amounts of data, how can end‐users access this powerful, parallel hardware to accelerate spreadsheet computation? Some existing solutions are usually not fully automatic and require a level of interaction from end‐users to facilitate parallel execution. Ideally, an end‐user tool would transparently exploit the underlying hardware and automatically discover available parallelism in the spreadsheet without any required interaction. This article presents an algorithm for automatic parallel evaluation of the cells in a spreadsheet and dynamic, parallel cycle detection. It is implemented in the Funcalc research spreadsheet application which supports higher‐order functions in the spreadsheet paradigm. Altogether, this promises a powerful and expressive platform for end‐user development. Our results on a 48 logical core machine show a maximum 7–14× speed‐up on a set of benchmark spreadsheets and a maximum 21–32× speed‐up on a set of synthetic spreadsheets with predefined topologies.