Speedup Ratio Research Articles

3D magnetotelluric modeling using high-order tetrahedral Nédélec elements on massively parallel computing platforms

We present a routine for 3D magnetotelluric (MT) modeling based upon high-order edge finite element method (HEFEM), tailored and unstructured tetrahedral meshes, and high-performance computing (HPC). This implementation extends the PETGEM modeller capabilities, initially developed for active-source electromagnetic methods in frequency-domain. We assess the accuracy, robustness, and performance of the code using a set of reference models developed by the MT community in well-known reported workshops. The scale and geological properties of these 3D MT setups are challenging, making them ideal for addressing a rigorous validation. Our numerical assessment proves that this new algorithm can produce the expected solutions for arbitrarily 3D MT models. Also, our extensive experimental results reveal four main insights: (1) high-order discretizations in conjunction with tailored meshes can offer excellent accuracy; (2) a rigorous mesh design based on the skin-depth principle can be beneficial for the solution of the 3D MT problem in terms of numerical accuracy and run-time; (3) high-order polynomial basis functions achieve better speed-up and parallel efficiency ratios than low-order polynomial basis functions on cutting-edge HPC platforms; (4) a triple helix approach based on HEFEM, tailored meshes, and HPC can be extremely competitive for the solution of realistic and complex 3D MT models and geophysical electromagnetics in general.

Molecular Docking for Ligand-Receptor Binding Process Based on Heterogeneous Computing

Molecular docking aims to predict possible drug candidates for many diseases, and it is computationally intensive. Particularly, in simulating the ligand-receptor binding process, the binding pocket of the receptor is divided into subcubes, and when the ligand is docked into all cubes, there are many molecular docking tasks, which are extremely time-consuming. In this study, we propose a heterogeneous parallel scheme of molecular docking for the binding process of ligand to receptor to accelerate simulating. The parallel scheme includes two layers of parallelism, a coarse-grained layer of parallelism implemented in the message-passing interface (MPI) and a fine-grained layer of parallelism focused on the graphics processing unit (GPU). At the coarse-grain layer of parallelism, a docking task inside one lattice is assigned to one unique MPI process, and a grouped master-slave mode is used to allocate and schedule the tasks. Meanwhile, at the fine-gained layer of parallelism, GPU accelerators undertake the computationally intensive computing of scoring functions and related conformation spatial transformations in a single docking task. The results of the experiments for the ligand-receptor binding process show that on a multicore server with GPUs the parallel program has achieved a speedup ratio as high as 45 times in flexible docking and as high as 54.5 times in semiflexible docking, and on a distributed memory system, the docking time for flexible docking and that for semiflexible docking gradually decrease as the number of nodes used in the parallel program gradually increases. The scalability of the parallel program is also verified in multiple nodes on a distributed memory system and is approximately linear.

Implementation of OpenMP Parallelization of Rate-Dependent Ceramic Peridynamic Model

A rate-dependent peridynamic ceramic model, considering the brittle tensile response, compressive plastic softening and strain-rate dependence, can accurately represent the dynamic response and crack propagation of ceramic materials. However, it also considers the strain-rate dependence and damage accumulation caused by compressive plastic softening during the compression stage, requiring more computational resources for the bond force evaluation and damage evolution. Herein, the OpenMP parallel optimization of the rate-dependent peridynamic ceramic model is investigated. Also, the modules that compute the interactions between material points and update damage index are vectorized and parallelized. Moreover, the numerical examples are carried out to simulate the dynamic response and fracture of the ceramic plate under normal impact. Furthermore, the speed-up ratio and computational efficiency by multi-threads are evaluated and discussed to demonstrate the reliability of parallelized programs. The results reveal that the total wall clock time has been significantly reduced after optimization, showing the promise of parallelization process in terms of accuracy and stability.

Data-Aware Adaptive Pruning Model Compression Algorithm Based on a Group Attention Mechanism and Reinforcement Learning

The success of convolutional neural networks (CNNs) benefits from the stacking of convolutional layers, which improves the model’s receptive field for image data but also causes a decrease in inference speed. To improve the inference speed of large convolutional network models without sacrificing performance indicators too much, a data-aware adaptive pruning algorithm is proposed. The algorithm consists of two parts, namely, a channel pruning method based on the attention mechanism and a data-aware pruning policy based on reinforcement learning. Experimental results on the CIFAR-100 dataset show that the performance of the proposed pruning algorithm is reduced by only 2.05%, 1.93% and 5.66% after pruning the VGG19, ResNet56 and EfficientNet networks, respectively, but the speedup ratios are 3.63, 3.35, and 1.14, respectively, and the comprehensive pruning performance is the best. In addition, the generalization ability of the reconstruction model is evaluated on the ImageNet dataset and FGVC Aircraft dataset, and the performance of the proposed algorithm is the best, which shows that the proposed algorithm learns data-related information in the pruning process, that is, it is a data-aware algorithm.

Random Partition Based Adaptive Distributed Kernelized SVM for Big Data

In this paper, we present a distributed classification technique for big data by efficiently using distributed storage architecture and data processing units of a cluster. While handling such large data, the existing approaches consider specific data partitioning techniques which demand complete data be processed before partitioning. This leads to an excessive overhead of high computation and data communication. The proposed method does not require any pre-structured data partitioning technique and is also adaptive to big data mining tools. We hypothesize that an effective aggregation of the information generated from data partitions by subprocesses of the complete learning process can lead to accurate prediction results while reducing the overall time complexity. We build three SVM based classifiers, namely one phase voting SVM (1PVSVM), two phase voting SVM (2PVSVM), and similarity based SVM (SIMSVM). Each of these classifiers utilizes the support vectors as the local information to construct the synthesized learner for efficiently reducing the training time and ensuring minimal communication between processing units. In this context, an extensive empirical analysis demonstrates the effectiveness of our classifiers when compared to other existing approaches on several benchmark datasets. However, among existing methods and three of our proposed (1PVSVM, 2PVSIM, and SIMSVM) methods, SIMSVM is the most efficient. Considering MNIST dataset, SIMSVM achieves an average speedup ratio of 0.78 and minimum scalability of 73% when the data size is scaled up to 10 times. It also retains high accuracy (99%) similar to centralized approaches.

Research on heterogeneous acceleration platform based on FPGA

In the context of today’s artificial intelligence, the volume of data is exploding. Although scaling distributed clusters horizontally to cope with the increasing demands on computing power for massive data processing is feasible. But the unlimited addition of nodes will lead to bloated cluster size. Most of the transistors in CPUs are used to build cache memory and control units, which are not efficient for computing operations of massive data processing. Currently, academia uses hardware devices such as GPU (Graphics Processing Unit), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array) to accelerate deep learning, image processing, which require massive computational operations. The paper first discussed the advantages and technical requirements of FPGA acceleration based on the characteristics of the Spark cluster. Then the paper proposed the design of the FPGA-CPU heterogeneous acceleration platform, and introduced the base-two-FFT algorithm. Finally, the paper present and compared the computation time of the base-two-FFT algorithm before and after the acceleration. The results show that the heterogeneous cluster has a speedup ratio of about 1.79 times compared to the CPU cluster.

Research on Parallel Support Vector Machine Based on Spark Big Data Platform

With the development of cloud computing and distributed cluster technology, the concept of big data has been expanded and extended in terms of capacity and value, and machine learning technology has also received unprecedented attention in recent years. Traditional machine learning algorithms cannot solve the problem of effective parallelization, so a parallelization support vector machine based on Spark big data platform is proposed. Firstly, the big data platform is designed with Lambda architecture, which is divided into three layers: Batch Layer, Serving Layer, and Speed Layer. Secondly, in order to improve the training efficiency of support vector machines on large-scale data, when merging two support vector machines, the “special points” other than support vectors are considered, that is, the points where the nonsupport vectors in one subset violate the training results of the other subset, and a cross-validation merging algorithm is proposed. Then, a parallelized support vector machine based on cross-validation is proposed, and the parallelization process of the support vector machine is realized on the Spark platform. Finally, experiments on different datasets verify the effectiveness and stability of the proposed method. Experimental results show that the proposed parallelized support vector machine has outstanding performance in speed-up ratio, training time, and prediction accuracy.

An Advanced Dynamic Visualization Method of Global Ocean Tide with Multi-core Processor Based on OpenGL

Abstract. In terms of ocean tide visualization, to meet the requirement of both display range and operational efficiency, an advanced method is proposed, in which the tide height is rapidly computed with global tide model EOT10a, and dynamically displayed by OpenGL. Aiming at the large amounts of calculation of global tide height, the feature of multicore processor is integrated into the method. The experiment shows that, compared to a single-core processor, when using a 6-core processor, the speedup ratio is about 5.4, parallel efficiency reaches 90%, and 880 000 tide heights can be calculated per second. Eventually, the result would be output as a tide height graph by OpenGL. This method could be a useful tool for marine cartography due to the large display range and the high efficiency.

Parallel assignment of flow directions over flat surfaces in massive digital elevation models

The assignment of flow directions in flat regions needs to be treated with special algorithms to obtain hydrologically correct flow patterns. Based on the sequential algorithm and the three-step parallel framework, we propose a parallel algorithm for assigning the hydrologically correct flow directions in flat regions. The proposed parallel algorithm assigns pre-divided tiles to multiple consumer processes, which construct local graphs that encode the geodesic distance information among tile border flat cells and higher terrain or lower terrain. Based on the local graphs in all tiles, the producer process constructs the global graph and computes the global geodesic distances of tile border flat cells from higher terrain and lower terrain. The consumer processes then compute the final geodesic distances in each tile and assign the hydrologically correct flow directions. Four experiments are conducted to evaluate the performance of our proposed algorithm. The speed-up ratios are greater than 3 when the number of consumer processes is greater than 9. When the number of consumer processes reaches 20, the average strong scaling efficiency is around 40%. The weak scaling efficiency is lower than 20% when the number of consumer processes is greater than 5. Compared with the sequential algorithm that is simpler to implement, the proposed parallel algorithm requires less memory on consumer processes than the sequential algorithm and can process massive digital elevation models that cannot be successfully processed using the sequential algorithm. It fills the void in the processing pipeline of automatic drainage network extraction based on the three-step parallel framework and enables parallel implementation of the entire processing pipeline possible, which should substantially improve the attractiveness of the three-step parallel framework.

A large scale parallel fluid‐structure interaction computing platform for simulating structural responses to a detonation shock

AbstractDue to the intrinsic nature of multi‐physics, it is prohibitively complex to design and implement a simulation software platform for study of structural responses to a detonation shock. In this article, a partitioned fluid‐structure interaction computing platform is designed for parallel simulating structural responses to a detonation shock. The detonation and wave propagation are modeled in an open‐source multi‐component solver based on OpenFOAM and blastFoam, and the structural responses are simulated through the finite element library deal.II. To capture the interaction dynamics between the fluid and the structure, both solvers are adapted to preCICE. For improving the parallel performance of the computing platform, the inter‐solver data is exchanged by peer‐to‐peer communications and the intermediate server in conventional multi‐physics software is eliminated. Furthermore, the coupled solver with detonation support has been deployed on a computing cluster after considering the distributed data storage and load‐balancing between solvers. The 3D numerical result of structural responses to a detonation shock is presented and analyzed. On 256 processor cores, the speedup ratio of the simulations for a detonation shock reach 178.0 with 5.1 million of mesh cells and the parallel efficiency achieve 69.5%. The results demonstrate good potential of massively parallel simulations. Overall, a general‐purpose fluid‐structure interaction software platform with detonation support is proposed by integrating open source codes. And this work has important practical significance for engineering application in fields of construction blasting, mining, and so forth.

Study on Full-scale Measurements of Wind Field Characteristics of Coastal Complex Mountainous Terrains

Based on full-scale wind field measurements of coastal complex mountainous terrains, data of fluctuating wind velocities at the height of 30m for four sites, including mountaintop and hillsides, are obtained. The wind load characteristics of mean wind velocities, wind directions, turbulence intensities and speed-up ratios of wind velocities are comprehensively examined. Results show that the maximum mean wind velocity at the mountaintop site is 12.4 m/s. The probability density distributions of mean wind velocities for the four measurement sites can be well represented by the Weibull probability model. The predominant wind directions are around the northeast and southwest. The longitudinal, lateral and vertical turbulence intensities decrease with the increase of mean wind velocities. The turbulence intensities for the mountaintop site are as many as 0.13,0.12 and 0.089 under maximum wind velocities, respectively for the previously mentioned three directions. The speed-up ratios of wind velocities between mountaintop and hillside sites are reduced, as the wind velocities increase. However, in cases of intensive wind with mean wind velocities larger than 8 m/s, the speed-up effects of wind velocities can also appear. The maximum speed-up ratio can reach 1.17.

A study on the characteristics of roof wind field by wind tunnel test

More and more solar-powered devices are installed on roofs, and wind load is its structural design control load. Therefore, roof wind field condition is considered to be the most important prerequisite in studies of wind load on structures. In this study, wind tunnel testing was carried out to observe the roof wind field characteristics of a multistorey building. The change rule and characteristics of the multistorey building roof wind field, under different wind directions were studied, and the roof wind field and ground wind field were then compared. The analysis showed that the closer to the roof, the greater the roof wind field effect on the building. When the distance from the roof Z ≥ 9 m, it appears that changes in roof wind speed and turbulence intensity are not affected by the building. When the height from roof Z ≤ 5 m, a turbulence intensity enlargement ratio Δ S I of 0.1–0.3 is recommended. For practical application, a speed-up ratio of the mean wind speed Δ S V of 0.9 is recommended. • The wind tunnel test is carried out to study the characteristics of roof wind field. • The wind speed and turbulence intensity are not affected by the building when Z ≥ 9 m. • When Z ≤ 5 m, the enlargement ratio Δ S I is recommended to be 0.1–0.3. • The speed-up ratio of the mean wind speed on the roof Δ S V is recommended to be 0.9. • The conclusions can provide references for structural wind resistance design of roof equipment.

Euclidean distance stratified random sampling based clustering model for big data mining

Big data mining is related to large-scale data analysis and faces computational cost-related challenges due to the exponential growth of digital technologies. Classical data mining algorithms suffer from computational deficiency, memory utilization, resource optimization, scale-up, and speed-up related challenges in big data mining. Sampling is one of the most effective data reduction techniques that reduces the computational cost, improves scalability and computational speed with high efficiency for any data mining algorithm in single and multiple machine execution environments. This study suggested a Euclidean distance-based stratum method for stratum creation and a stratified random sampling-based big data mining model using the K-Means clustering (SSK-Means) algorithm in a single machine execution environment. The performance of the SSK-Means algorithm has achieved better cluster quality, speed-up, scale-up, and memory utilization against the random sampling-based K-Means and classical K-Means algorithms using silhouette coefficient, Davies Bouldin index, Calinski Harabasz index, execution time, and speedup ratio internal measures.

PSATop-[formula omitted]: Approximate range top-[formula omitted] computation on big data

Approximate top-k query returns a list of k tuples that have approximate largest scores with respect to the user given query. However, existing algorithms cannot effectively process the approximate top-k queries on big data, because they either restrict the class of ranking functions, or fail to take selection conditions into consideration. In this paper, a novel algorithm PSATop-k, which combines partitioning and sampling techniques, is proposed to answer approximate range top-k query efficiently. PSATop-k is suitable for queries with selection conditions and arbitrary ranking functions. PSATop-k first determines the sampling size that meets the accuracy requirement, then draws sufficient random tuples to return result by accessing a subset of the partitioned data. The experimental results on the real-life and synthetic datasets demonstrate that PSATop-k performs much better than the existing algorithms. Specially, as result set size varies from 10 to 80, PSATop-k runs 12.22 to 14.30 times faster than TA-based and Coreset-based methods on average. The speedup ratios are 11.43 to 23.34 and 55.86 to 649.06 in the experiments of error bound and tuple number, respectively.

A Scalable Moving Boundary Treatment in the Lattice Boltzmann Method

Based on the lattice Boltzmann Method (LBM), a new moving boundary treatment is proposed by studying the influence of the moving boundary in a low-speed incompressible fluid. Simulating flow past a static cylinder, flow past a moving cylinder, and flow in a curved moving pipe shows that the LBM has high reliability and stability with moving boundary conditions. In order to reduce parallel computing time and take full advantage of the characteristics of the model, the solid grids are allocated according to the number of computing cores. Two parallel speedup ratios were tested, keeping the overall task volume unchanged to increase the number of cores and keeping the single-core task volume unchanged. In the first speedup ratio, the efficiency of the flow field calculation at 112 cores reached more than 80%. The second speedup ratio indicated that the proportion of communication in the program was tiny and suitable for large-scale parallel computing. The study of the moving boundary problem can effectively help solve the fluid-structure coupling problem. Due to the particularity of the model, it can be used to simulate the fluid-structure coupling problem of blood vessel flow effectively.

Random sampling accelerator for attribute reduction

As one of the crucial topics in the development of rough set, attribute reduction has received extensive attentions because it is practical and interpretable for us to perform dimensional reduction or feature selection. Currently, to further improve the efficiency of searching reducts, many researchers have devoted themselves to designing various accelerative mechanisms. Among these existing results, it should be pointed out that the accelerators designed by reducing the scale of samples strongly depend on the distribution of data. To fill such a gap, an accelerator based on the random sampling is developed. The superiorities of our accelerator are: (1) randomly selecting samples without considering the distribution of data; (2) guidance-based evaluations and selections of attributes; (3) easily to be combined with other popular searching strategies. By comparing with 5 state-of-the-art accelerators over 24 UCI data sets: (1) our accelerator may significantly reduce the time consumption of deriving reducts and then the average speed-up ratio can exceed 10; (2) the reduct derived by our accelerator can offer competent performance in classification task.

Learning to select cuts for efficient mixed-integer programming

Cutting plane methods play a significant role in modern solvers for tackling mixed-integer programming (MIP) problems. Proper selection of cuts would remove infeasible solutions in the early stage, thus largely reducing the computational burden without hurting the solution accuracy. However, the major cut selection approaches heavily rely on heuristics, which strongly depend on the specific problem at hand and thus limit their generalization capability. In this paper, we propose a data-driven and generalizable cut selection approach, named Cut Ranking, in the settings of multiple instance learning. To measure the quality of the candidate cuts, a scoring function, which takes the instance-specific cut features as inputs, is trained and applied in cut ranking and selection. In order to evaluate our method, we conduct extensive experiments on both synthetic datasets and real-world datasets. Compared with commonly used heuristics for cut selection, the learning-based policy has shown to be more effective, and is capable of generalizing over multiple problems with different properties. Cut Ranking has been deployed in an industrial solver for large-scale MIPs. In the online A/B testing of the product planning problems with more than 107 variables and constraints daily, Cut Ranking has achieved the average speedup ratio of 12.42% over the production solver without any accuracy loss of solution.

CPU-accelerated explicit three-dimensional discontinuous deformation analysis with cloud computing

Prompt feedback based on a numerical simulation, such as discontinuous deformation analysis, plays an important role in digitalized geotechnical engineering. Cloud computing also prevails owing its flexible configuration and computational power. This study proposes an explicit three-dimensional discontinuous deformation analysis method using explicit contact forces that can circumvent the difficulty of not converging the open–close iteration and the inefficiency of classic implicit discontinuous deformation analysis. The serial code is ready to be parallelized by relocating the postjudgment of the contact status and using LDL decomposition to solve fixed-size linear equations. We propose herein the block-wise parallelization concept, which significantly simplifies the work of parallel computation. No modification under the block level is needed. In addition, one can realize parallel computing by adding just three lines of the new code with the help of OpenMP. A simple virtual cloud machine is used to run the program to validate the improved efficiency of the newly developed algorithms. A 3.5 speedup ratio is achieved with four CPU cores and eight processors.

GP-SWAT (v1.0): a two-level graph-based parallel simulation tool for the SWAT model

Abstract. High-fidelity and large-scale hydrological models are increasingly used to investigate the impacts of human activities and climate change on water availability and quality. However, the detailed representations of real-world systems and processes contained in these models inevitably lead to prohibitively high execution times, ranging from minutes to days. Such models become computationally prohibitive or even infeasible when large iterative model simulations are involved. In this study, we propose a generic two-level (i.e., watershed- and subbasin-level) model parallelization schema to reduce the run time of computationally expensive model applications through a combination of model spatial decomposition and the graph-parallel Pregel algorithm. Taking the Soil and Water Assessment Tool (SWAT) as an example, we implemented a generic tool named GP-SWAT, enabling watershed-level and subbasin-level model parallelization on a Spark computer cluster. We then evaluated GP-SWAT in two sets of experiments to demonstrate the ability of GP-SWAT to accelerate single and iterative model simulations and to run in different environments. In each test set, GP-SWAT was applied for the parallel simulation of four synthetic hydrological models with different input/output (I/O) burdens. The single-model parallelization results showed that GP-SWAT can obtain a 2.3–5.8-times speedup. For multiple simulations with subbasin-level parallelization, GP-SWAT yielded a remarkable speedup of 8.34–27.03 times. In both cases, the speedup ratios increased with an increasing computation burden. The experimental results indicate that GP-SWAT can effectively solve the high-computational-demand problems of the SWAT model. In addition, as a scalable and flexible tool, it can be run in diverse environments, from a commodity computer running the Microsoft Windows operating system to a Spark cluster consisting of a large number of computational nodes. Moreover, it is possible to apply this generic tool to other subbasin-based hydrological models or even acyclic models in other domains to alleviate I/O demands and to optimize model computational performance.

Swarm intelligence‐based secure high‐order optimal density selection for industrial internet‐of‐things (IIoT) data on cloud environment

SummaryIndustrial internet‐of‐things (IIoT) cloud computing gives users with the ease of outsourcing calculations and carrying the risk of privacy disclosure. To solve this issue, secure high‐order clustering algorithm by fast search (SHOCFS) technique is introduced recently. The major factors of SHOCFS algorithm are speed improvement and detecting optimal density peaks. The optimal selection of density points in the SHOCFS algorithm still becomes an important issue to be considered. The swallow swarm optimization (SSO) is introduced in the proposed work for choosing optimal density peaks in clustering algorithm. This paper motivates to design a new secure cloud service system for cloud in IIoT environment. The secure high‐order optimal density selection (SHODS3O‐CFS) approach is proposed in hybrid cloud environment. The major contribution of the work is to develop an optimization‐based secure clustering algorithm for finding the best density points in clustering algorithm. It may reduce the overlapping peak points in the cluster and increase the security of system in hybrid cloud. The experimental results of the proposed algorithm and existing methods are measured via clustering metrics such as accuracy, clustering center quality, and speed ratio. The metrics like clustering center quality, Rand index (RI), encryption time (ms), and speedup ratio have been used to measure the results of the proposed system and existing methods. From the results, it is concluded that the proposed work has achieved higher average RI of 93.4%, which is 29.68% and 17.4% higher when compared with the privacy preservation high‐order clustering algorithm by fast search (PPHOCFS) and SHOCFS methods.