Search

In this paper, we present the development of a new version of the BrkgaCuda, called BrkgaCuda 2.0, to support the design and execution of Biased Random-Key Genetic Algorithms (BRKGA) on CUDA/GPU-enabled computing platforms, employing new techniques to accelerate the execution. We compare the performance of our implementation against the standard CPU implementation and a previous GPU implementation. In the same spirit as the standard implementation, all central aspects of the BRKGA logic are dealt with in our framework, and little effort is required to reuse the framework on another problem. The user can also implement the decoder on the CPU in C++ or GPU in CUDA. Moreover, the BrkgaCuda 2.0 provides a decoder that receives a permutation created by sorting the indices of the chromosomes using the genes as keys. To evaluate our framework, we use a total of 54 instances of the Traveling Salesman Problem, the Set Cover Problem, and the Capacitated Vehicle Routing Problem, using two decoders in the last one. Our focus on this work is to improve the performance within the BRKGA logic. We show that the BrkgaCuda 2.0 is the fastest among similar frameworks while still providing solutions of similar quality. [ABSTRACT FROM AUTHOR]

This paper introduces a parallel Topology Optimization (TO) platform capable of optimizing designs for multiple objectives, whilst subject to multiple constraints, in the open source Fierro Finite Element code. The TO methodology uses a continuous material interpolation scheme, which avoids checkerboard designs without additional filters and constraints as seen with piece-wise constant material interpolation schemes. Additionally, analytic Hessian-vector products are used, with the relevant adjoint calculation shown. The code utilizes OpenMPI to make parallel the major computational segments of TO: global equation assembly, global equation solution, and the non-linear optimization of the design. The algorithm leverages several software packages: ELEMENTS (grants FE basis functions), MATAR (grants efficient multidimensional dense and sparse matrix storage), Zoltan2 (Mesh Decomposition Algorithm), MueLu (parallel multi-grid solver for the global equilibrium equations), and ROL (non-linear optimization). It is found that the Fierro TO platform optimizes problems with mechanical and thermal objectives subject to multiple constraints: mass, several moment of inertia targets, and constraints related to load bearing regions. Additionally, the performance of several ROL algorithms using analytic Hessian-vector products is compared with the Method of Moving Asymptotes, approximate Hessian, for topology optimization. [ABSTRACT FROM AUTHOR]

Conventional visual search tasks do not address attention directly and their core manipulation of 'set size' – the number of displayed items – introduces stimulus confounds that hinder interpretation. However, alternative approaches have not been widely adopted, perhaps reflecting their complexity, assumptions, or indirect attention-sampling. Here, a new procedure, the ATtention Location And Size ('ATLAS') task used probe displays to track attention's location, breadth, and guidance during search. Though most probe displays comprised six items, participants reported only the single item they judged themselves to have perceived most clearly – indexing the attention 'peak'. By sampling peaks across variable 'choice sets', the size and position of the attention window during search was profiled. These indices appeared to distinguish narrow- from broad attention, signalled attention to pairs of items where it arose and tracked evolving attention-guidance over time. ATLAS is designed to discriminate five key search modes: serial-unguided, sequential-guided, unguided attention to 'clumps' with local guidance, and broad parallel-attention with or without guidance. This initial investigation used only an example set of highly regular stimuli, but its broader potential should be investigated. [ABSTRACT FROM AUTHOR]

Stream surfaces can reveal complex flow behaviors that are not available on streamlines, but their construction algorithm is also more sophisticated. A common strategy is to sample a series of streamlines from a seeding curve and stitch the neighboring streamlines to form surfaces. To maintain a uniform density of sample points, extra streamlines may be inserted when the flow diverges. This leads to high computation cost with sequential insertion operations, hindering the use of stream surfaces in interactive scenarios and large-scale learning. In this paper, we proposed two parallel strategies to construct stream surfaces on GPUs. Both strategies generate stream surfaces in multiple passes, where each pass consists of a streamline tracing phase and a ribbon construction phase. The first strategy aims at increasing throughput, by hiding the long generation time of an individual surface with computation from other surfaces. The second strategy aims at reducing latency, by generating redundant streamlines for a single surface. We evaluate the performance of the two strategies and compare them with a CPU baseline using ten datasets exhibiting various characteristics. [ABSTRACT FROM AUTHOR]

In this research, we describe the modulus-based synchronous multisplitting iteration method for solving a class of horizontal nonlinear complementarity problems by utilizing the matrix multisplitting technique. We derive several convergence criteria to guarantee the suggested method's convergence, and we provide numerical examples to demonstrate our method's efficacy. [ABSTRACT FROM AUTHOR]

Multi-effect parallel feed desalination system (MED-PF) has emerged as an efficient technology for large-scale seawater desalination. Enhancing the system's performance ratio and incorporation of a thermal vapor compressor (TVC) have proven to be effective in previous studies. Notably, these studies have suggested that the optimal placement for the TVC is within the middle section. The current study explores the impact of a new component, a brine preheater, on the performance of multi-effect desalination system. The hot rejected brine is used to heat the feed seawater for systems with various number of effects. Results demonstrate that utilizing the rejected brine for feed preheating significantly improves the system's performance ratio among other important parameters. Comparative analysis also shows that utilization of heat associated with the rejected brine for feed preheating leads to a reduction in the cost of water produced in the range of 4.55–9.63%. For example, water production cost of 1.87 $/m3 for a 8-effect MED-PF system without brine preheating reduces to 1.69 $/m3 upon preheating the feed without an additional, external energy source. Nevertheless, this approach results in an improvement in the system's performance ratio (PR) in the range of 7.46% to 13.9%. The effect on the total heat transfer surface area (sA) and specific flow rate of cooling water is also taken into consideration. [ABSTRACT FROM AUTHOR]

In this paper, a relaxed two-step modulus-based matrix synchronous multisplitting iteration method for solving the linear complementarity problems is constructed. The convergence conditions of the proposed method are analyzed with the convergence range of the relaxation parameters. Some parallel numerical experiments under OpenACC framework are given to show that the proposed method can accelerate the convergence rate of the existing method significantly. [ABSTRACT FROM AUTHOR]

In this paper, the influences of stray capacitances are explored for various parallel arrangements of primary windings (PWs) in input-series transformer-integration (ISTI) flyback converters. On this basis, selecting the principle of the basic parallel PWs arrangements is clarified in the design process. First, the influences caused by the energy changes of stray capacitances are analyzed for PWs in ISTI converter. Based on this, the arrangements of single PWs and various parallel PWs are presented. Second, the energy influences of both the inner and inter-winding capacitances are analyzed and compared for two basic parallel PWs in an ISTI converter. In addition, selecting the principle of these PWs arrangements is summarized. Finally, four integrated transformers with various PW arrangements are designed, and experimental comparisons are implemented on a 1000 Vdc/48 W ISTI flyback converter. Through the obtained results, the validity of theoretical analysis is verified. [ABSTRACT FROM AUTHOR]

The disaster information collection mission should be executed after the disaster occurs to provide details for the decision-makers. During the execution of the information collection mission, some disruptions may occur and prevent the resource used for information collection from completing the mission as planned. It is difficult for decision-makers to make reactive resource scheduling plan that optimize the mission's execution time, quality, and cost at the same time under such circumstances. This article focuses on designing the reactive decision support algorithm for the disaster information collection resource scheduling, which aims to provide multi high-quality scheduling plans for decision-makers to choose. The problem studied in this article is modeled as an extension of Resource-Constrained Project Scheduling Problem (RCPSP). First, the basic problem formulation for a normal schedule and two disruption recovery models are presented. Second, a novel framework of a parallel pareto local search based on decomposition is designed to repair the schedule within the time limit. Third, two solution acceptance criteria based on constraint handling and negative correlation are specially designed to maintain high-quality population with diversity. The experiments show that the proposed method outperforms the other competitors with respect to Inverted Generational Distance, Spacing, and Hypervolume, which means that the proposed method can help decision-makers to make better decisions. [ABSTRACT FROM AUTHOR]

In malleable job scheduling, jobs can be executed simultaneously on multiple machines with the processing time depending on the number of allocated machines. In this setting, jobs are required to be executed non-preemptively and in unison, in the sense that they occupy, during their execution, the same time interval over all the machines of the allocated set. In this work, we study generalizations of malleable job scheduling inspired by standard scheduling on unrelated machines. Specifically, we introduce a general model of malleable job scheduling, where each machine has a (possibly different) speed for each job, and the processing time of a job j on a set of allocated machines S depends on the total speed of S with respect to j. For machines with unrelated speeds, we show that the optimal makespan cannot be approximated within a factor less than e e - 1 , unless P = N P . On the positive side, we present polynomial-time algorithms with approximation ratios 2 e e - 1 for machines with unrelated speeds, 3 for machines with uniform speeds, and 7/3 for restricted assignments on identical machines. Our algorithms are based on deterministic LP rounding. They result in sparse schedules, in the sense that each machine shares at most one job with other machines. We also prove lower bounds on the integrality gap of 1 + φ for unrelated speeds (φ is the golden ratio) and 2 for uniform speeds and restricted assignments. To indicate the generality of our approach, we show that it also yields constant factor approximation algorithms for a variant where we determine the effective speed of a set of allocated machines based on the L p norm of their speeds. [ABSTRACT FROM AUTHOR]

In this paper, for solving large sparse vertical linear complementarity problems, the modulus-based synchronous multisplitting iteration method is established. Convergence theorems of the proposed method are given, which are shown to generalize and improve the existing results of the serial method. Numerical tests by OpenACC technique are given to show the efficiency of the proposed method. [ABSTRACT FROM AUTHOR]

To resolve the issue of the curse of dimensionality in continuous large-scale optimization problems, the cooperative coevolution divide-and-conquer framework was proposed by dividing the problem into several subcomponents either randomly or based on the interaction between variables, each of which can be optimized separately using metaheuristic suboptimizers. The goal of researchers is to optimize the performance of algorithms in terms of both quality of solution and computational speed, seeing that large-scale optimization can be a computationally expensive process. This work proposes a parallel implementation to the cooperative coevolution framework for solving large-scale global optimization problems using the Graphics Processing Unit (GPU) and CUDA platform. A distributed variant of the cooperative coevolution framework is outlined to expose a degree of parallelism. Features of the GPU parallel technology and CUDA platform such as shared and global memories are used to optimize the subcomponents of the problem in parallel, speeding up the optimization process while attempting to maintain comparable search quality to works in the literature. The CEC 2010 large-scale global optimization benchmark functions are used for conducting experiments and comparing results in terms of improvements in search quality and search efficiency. Results of proposed parallel implementation show that a speedup of up to x13.01 is possible on large-scale global optimization benchmarks using the GPUs. [ABSTRACT FROM AUTHOR]

Sample rate conversion is an essential scheme used in almost every radio design. Supporting sampling rates higher than the clock rates require parallel processing. In this paper, we propose an algorithm for a sample rate converter (SRC) with multiple parallel output phases so that the conversion ratio can be a fixed rational number. Due to the structure of the proposed algorithm, it is suitable for embedded platforms which are restricted by their clock frequency but require very high sample rates. A dual phase output variant of the proposed algorithm is simulated with a 400 MHz input signal to perform a 15/8 conversion. The test and verification of the SRC algorithm is presented with the aid of a design example. A VLSI architecture of the dual phase output SRC is implemented on a Virtex-7 field-programmable gate array (FPGA) and results are presented. [ABSTRACT FROM AUTHOR]

Extreme learning machine (ELM) has received considerable attention due to its rapid learning speed and powerful fitting capabilities. One of its important variants, the elastic-net ELM (Enet-ELM), was recently proposed to improve its sparsity and stability of simulations simultaneously. However, entering the era of big data, the explosive growth of data volume and dimensions poses a huge challenge to Enet-ELM. On the other hand, the alternating direction method of multipliers (ADMM) is a powerful iterative algorithm for solving large-scale optimization problems by splitting a large problem into a set of executable sub-problems. But its performance is highly restricted by its astringency and convergence rates. In this paper, we therefore develop a novel Enet-ELM algorithm based on the over-relaxed ADMM, termed over-relaxed Enet-ELM (OE-ELM), which accelerates model training by applying the results of the previous iteration to the next iteration. Besides, we also propose a parallel version of OE-ELM (POE-ELM) to implement parallel and distributed computation, which is trained by the consensus over relaxation ADMM algorithm. Finally, the convergence analysis conducted on the two proposed algorithms proves the effectiveness of model training, and extensive experiments on classification and regression datasets demonstrate their competitiveness in accuracy and convergence rate. [ABSTRACT FROM AUTHOR]

Some efficient top-k algorithms, i.e., Fagin's Algorithm, threshold algorithm (TA), and best position algorithm (BPA), can be used to answer k nearest neighbor (kNN) queries. However, extending the existing algorithms without further changes to the algorithms themselves would not be efficient since there are the different characteristics between the kNN queries and top-k queries. For example, the kNN queries are more distance-sensitive rather than the position of data points. Second, it is necessary to add some novel parallel heuristics and pruning policies for the kNN queries. Third, there are still many redundant random accesses among FA, TA, and BPA. In this paper, we address aforementioned these problems and take these algorithms to answer parallel kNN (PkNN) queries in spatial databases. We integrate the advantages of the B + -tree and Open MP programming and propose three efficient parallel kNN query algorithms, namely distance priority-based PkNN, optimized PkNN, and partition-based PkNN. Our performance evaluation shows that our proposed algorithms achieve significant improvement in comparison with existing algorithms, i.e., BPA and BPA2. In addition, our approaches are also capable of returning kNN results incrementally which greatly shorten the query response time and enhance user experience. [ABSTRACT FROM AUTHOR]

Given a large data graph, trimming techniques can reduce the search space by removing vertices without outgoing edges. One application is to speed up the parallel decomposition of graphs into strongly connected components (SCC decomposition), which is a fundamental step for analyzing graphs. We observe that graph trimming is essentially a kind of arc-consistency problem, and AC-3, AC-4, and AC-6 are the most relevant arc-consistency algorithms for application to graph trimming. The existing parallel graph trimming methods require worst-case O (n m) time and worst-case O (n) space for graphs with n vertices and m edges. We call these parallel AC-3-based as they are much like the AC-3 algorithm. In this work, we propose AC-4-based and AC-6-based trimming methods. That is, AC-4-based trimming has an improved worst-case time of O (n + m) but requires worst-case space of O (n + m) ; compared with AC-4-based trimming, AC-6-based has the same worst-case time of O (n + m) but an improved worst-case space of O (n) . We parallelize the AC-4-based and AC-6-based algorithms to be suitable for shared-memory multi-core machines. The algorithms are designed to minimize synchronization overhead. For these algorithms, we also prove the correctness and analyze time complexities with the work-depth model. In experiments, we compare these three parallel trimming algorithms over a variety of real and synthetic graphs on a multi-core machine, where each core corresponds to a worker. Specifically, for the maximum number of traversed edges per worker by using 16 workers, AC-3-based traverses up to 58.3 and 36.5 times more edges than AC-6-based trimming and AC-4-based trimming, respectively. That is, AC-6-based trimming traverses much fewer edges than other methods, which is meaningful especially for implicit graphs. In particular, for the practical running time, AC-6-based trimming achieves high speedups over graphs with a large portion of trimmable vertices. [ABSTRACT FROM AUTHOR]

Given a user-specified minimum degree threshold γ , a γ -quasi-clique is a subgraph where each vertex connects to at least γ fraction of the other vertices. Quasi-clique is a natural definition for dense structures, so finding large and hence statistically significant quasi-cliques is useful in applications such as community detection in social networks and discovering significant biomolecule structures and pathways. However, mining maximal quasi-cliques is notoriously expensive, and even a recent algorithm for mining large maximal quasi-cliques is flawed and can lead to a lot of repeated searches. This paper proposes a parallel solution for mining maximal quasi-cliques that is able to fully utilize CPU cores. Our solution utilizes divide and conquer to decompose the workloads into independent tasks for parallel mining, and we addressed the problem of (i) drastic load imbalance among different tasks and (ii) difficulty in predicting the task running time and the time growth with task-subgraph size, by (a) using a timeout-based task decomposition strategy, and by (b) utilizing a priority task queue to schedule long-running tasks earlier for mining and decomposition to avoid stragglers. Unlike our conference version in PVLDB 2020 where the solution was built on a distributed graph mining framework called G-thinker, this paper targets a single-machine multi-core environment which is more accessible to an average end user. A general framework called T-thinker is developed to facilitate the programming of parallel programs for algorithms that adopt divide and conquer, including but not limited to our quasi-clique mining algorithm. Additionally, we consider the problem of directly mining large quasi-cliques from dense parts of a graph, where we identify the repeated search issue of a recent method and address it using a carefully designed concurrent trie data structure. Extensive experiments verify that our parallel solution scales well with the number of CPU cores, achieving 26.68 × runtime speedup when mining a graph with 3.77M vertices and 16.5M edges with 32 mining threads. Additionally, mining large quasi-cliques from dense parts can provide an additional speedup of up to 89.46 × . [ABSTRACT FROM AUTHOR]

As a result of its advantages in the mid-temperature range, the proton ceramic fuel cell (PCFC) has gotten much attention compared to other fuel cells. In PCFC, whereby the vapor is generated at the cathode side, a flow field with high efficiency in water removal should be in its repertoire. A new inter-parallel flow field was designed and compared to the traditional flow fields: the serpentine, the parallel, and interdigitated flow fields to check its capacity to be a potential flow field for PCFC. The one-cell stack 3D model with the inter-parallel flow field was simulated to analyze the working details within the stack. The design harnessed the advantages of its three unique flow paths in solving the balance problem of pressure drop and effective oxygen transport and water removal. Results show that the design is a potential flow field for the cathode side of the PCFC. [ABSTRACT FROM AUTHOR]

Ability to perform fast analysis on massive public blockchain transaction data is needed in various applications such as tracing fraudulent financial transactions. The blockchain data is continuously growing and is organized as a sequence of blocks containing transactions. This organization, however, cannot be used for parallel graph algorithms which need efficient distributed graph data structures. Using message passing libraries (MPI), we develop a scalable cluster-based system that constructs a distributed transaction graph in parallel and implement various transaction analysis algorithms. We report performance results from our system operating on roughly 5 years of 10.2 million block Ethereum Mainnet blockchain data. We report timings obtained from tests involving distributed transaction graph construction, partitioning, page ranking of addresses, degree distribution, token transaction counting, connected components finding and our new parallel blacklisted address trace forest computation algorithm on a 16 node economical cluster set up on the Amazon cloud. Our system is able to construct a distributed graph of 766 million transactions in 218 s and compute the forest of blacklisted address traces in 32 s. [ABSTRACT FROM AUTHOR]

The complexity of intra prediction in high-efficiency video coding (HEVC) is increased due to the addition of five variable sized prediction units (PUs) and 35 directional predictions. In this work, we propose an efficient parallel-pipelined architecture that can process 8 samples in parallel for every clock cycle. The functional units needed to predict the PU samples work in a pipelined fashion. With this balanced combination of parallel-pipelined structure, we are able to achieve higher throughput with limited hardware resources than existing literature works. The samples are processed row-wise, so that they can be directly transform coded, thus eliminating the need for an intermediate memory buffer of 8 K between the two modules. A compact reconfigurable reference buffer of size 0.8 KB is incorporated to reduce the read-write latency associated with reference samples' fetching. A dedicated module for arithmetic operations is used in the intra engine that ensures the reuse of multipliers to increase the hardware efficiency. The architecture so designed supports all the PU sizes and directional modes. The proposed design is tested and implemented on a field-programmable gate array (FPGA) platform operating at 150 MHz frequency to achieve 8 samples throughput with a hardware cost of 16.2 K Look-Up Tables (LUTs) and 5.7 K registers to support HD 4 K real-time video encoding applications. [ABSTRACT FROM AUTHOR]

Big data analysis requires the speedup of parallel computing. However, in the virtualized systems, the power of parallel computing is not fully exploited due to the limit of current VMM schedulers. Xen, one of the most popular virtualization platforms, has been widely used by industry to host parallel job. In practice, the virtualized systems are expected to accommodate both parallel jobs and serial jobs, and resource contention between virtual machines results in severe performance degradation of the parallel jobs. Moreover, the physical resource is vastly wasted during the communication process due to the ineffective scheduling of parallel jobs. Unfortunately, the existing schedulers of Xen are initially targeting at serial jobs, which are not capable of correctly scheduling the parallel jobs. This paper presents vChecker, an application-level co-scheduler which mitigates the performance degradation of the parallel job and optimizes the utilization of the hardware resource. Our co-scheduler takes number of available CPU cores in one hand, and satisfies need of the parallel jobs in other hand, which helps the credit scheduler of Xen to appropriately schedule the parallel job. As our co-scheduler is implemented at application level, no modifications on the hypervisor is required. The experimental result shows that the vChecker optimizes the performance of the parallel job in Xen and enhances the utilization of the system. [ABSTRACT FROM AUTHOR]

Prototyping, whether physical, virtual, or computational, is an important step in the engineering design process. Iterative prototyping strategies are commonly taught in engineering curricula and implemented in industry, but there may be other ways to approach the prototyping process. Engineers often use physical prototypes to learn about their designs, communicate ideas, and validate effectiveness. In this study, the effects of iterative and parallel prototyping strategies are compared through a design competition with a heavy focus on gaining knowledge from the physical models. Design success, engineering design self-efficacy, and solution space exploration are considered to evaluate the different effects of these two prototyping strategies. Results suggest that a parallel prototyping strategy yields greater design success, increased confidence and reduced anxiety when conducting engineering design, and greater exploration of the solution space. In addition, participants seem largely unaware of these benefits based a post-prototyping survey. This work shows the value of parallel prototyping, which has implications for how prototyping is taught to engineering novices and how engineering designers in industry should approach the prototyping process. This study also provides strong evidence for a need to study the benefits and drawbacks of a parallel prototypin approach in more complex situations. [ABSTRACT FROM AUTHOR]

A frequent pattern is a substructure that appears in a database with frequency (aka. support) no less than a user-specified threshold, while a closed pattern is one that has no super-pattern that has the same support. Here, a substructure can refer to different structural forms, such as itemsets, subsequences, subtrees, and subgraphs, and mining such substructures is important in many real applications such as product recommendation and feature extraction. Currently, there lacks a general programming framework that can be easily customized to mine different types of patterns, and existing parallel and distributed solutions are IO-bound rendering CPU cores underutilized. Since mining frequent and/or closed patterns are NP-hard, it is important to fully utilize the available CPU cores. This paper presents such a general-purpose framework called PrefixFPM. The framework is based on the idea of prefix projection which allows a divide-and-conquer mining paradigm. PrefixFPM exposes a unified programming interface to users who can readily customize it to mine their desired patterns. We have adapted the state-of-the-art serial algorithms for mining patterns including subsequences, subtrees, and subgraphs on top of PrefixFPM, and extensive experiments demonstrate an excellent speedup ratio of PrefixFPM with the number of CPU cores. [ABSTRACT FROM AUTHOR]

Modern scientific research challenges require new technologies, integrated tools, reusable and complex experiments in distributed computing infrastructures. But above all, computing power for efficient data processing and analyzing. Containers technologies have emerged as a new paradigm to address such intensive scientific applications problems. Their easy deployment in a reasonable amount of time and the few required computational resource make them more suitable. Containers are considered light virtualization solutions. They enable performance isolation and flexible deployment of complex, parallel, and high-performance systems. Moreover, they gained popularity to modernize and migrate scientific applications in computing infrastructure management. Additionally, they reduce computational time processing. In this paper, we first give an overview of virtualization and containerization technologies. We discuss the taxonomies of containerization technologies of the literature, and then we provide a new one that covers and completes those proposed in the literature. We identify the most important application domains of containerization and their technological progress. Furthermore, we discuss the performance metrics used in most containerization techniques. Finally, we point out research gaps in the related aspects of containerization technology that require more research. [ABSTRACT FROM AUTHOR]

Streams are temporally ordered, rapid changing, ample in volume, and infinite in nature. It is nearly impossible to store the entire data stream due to its large volume and high velocity. In this work, the principle of parallelism is employed to accelerate stream data computing. GPU-based high-performance computing (HPC) framework is proposed for accelerated processing of big-data streams using the in-memory data structure. We have implemented three parallel algorithms to prove the viability of the framework. The contributions of Mille Cheval are: (1) the viability of streaming on accelerators to increase throughput, (2) carefully chosen hash algorithms to achieve low collision rate and high randomness, and (3) memory sketches for approximation. The objective is to leverage the power of a single node using in-memory computing and hybrid computing. HPC does not always require high-end hardware but well-designed algorithms. Achievements of Mille Cheval are: (1) relative error is 1.32 when error rate and overestimate rate are chosen as 0.001 and (2) the host memory space requirement is just 63 MB for 1 terabyte of data. The proposed algorithms are pragmatic. It is evident from experimental results that the framework demonstrates 10X speed-up as compared with CPU implementations and 3X speed-up as compared with GPU implementations. [ABSTRACT FROM AUTHOR]

Distributed computing environments execute great amount of various computing jobs that can fail or break for some reason. The analysis of the error messages describing the reasons of failures has become one of the most crucial parts of the existing monitoring systems. This analysis is complicated by the presence of a large number of messages, especially in the case of the retrospective analysis. ClusterLogs framework was developed as a modular and flexible tool for the clustering of error messages of computing jobs in distributed computing infrastructures. The general purpose of this tool is to simplify the error messages analysis by grouping together messages that share similar failure reasons and textual patterns. Proposed clustering method includes a set of sequential data processing stages and provides various clustering options: deterministic similarity-based clustering and unsupervised multiple machine learning methods with preliminary vectorization of error messages using the word embedding technique. The performance tests had revealed the most time consuming stages. In this paper we describe the parallelilzing method for these stages and demonstrate how it has allowed the increased performance of the whole clustering pipeline. The performance tests were executed on the HPC system Polus. [ABSTRACT FROM AUTHOR]

A robotic manipulator is a networked assemblage of kinematic pairs, mobile platform(s) and/or tool(s) that generate motion and forces at its end-effector. A manipulator's theoretical dynamic, kinematic and workspace capabilities are governed by the design parameters (mass, geometry, dimensions, etc.) of its kinematic pairs and its architecture (number of limbs, degrees of freedom, actuation ability, etc.). By adjusting or changing the manipulator's parameters (both architectural and kinematic pair-related), either from a design standpoint or physically, the robot undergoes what the authors broadly refer to as a 'reassembling transformation'. By effecting reassembling transformations on a robot, we may generate desired changes in the workspace, kinematic and dynamic capabilities for a specific task manoeuvre. This work presents a novel methodology that marries quaternion rotors, the Lagrangian formulation, geometric algebra and associated mathematical apparatus, to tackle the problem of analysing and modelling robotic manipulators or mechanisms that undergo such 'reassembling transformations'. Upon presenting a quaternion-based methodology for position, kinematics and Lagrangian dynamics, reassembling transformation types are introduced and detailed. We apply reassembling transformations to a Delta manipulator, in two separate case studies, each with unique target changes to the dynamic capabilities of the post transformed robot in its chosen task manoeuvres. The framework presented in this paper is generalised and its methods can be used to analyse and transform any parallel manipulator and also extended to the study of non-linear systems of equations for modelling multi-body dynamics. [ABSTRACT FROM AUTHOR]

In this paper, a parallel mesh-free finite pointset method (FPM) for the 3D variable coefficient transient heat conduction problem (I-FPM-3D) on regular/irregular region is proposed by coupling several techniques as follows. The partial differential equation with the high-order derivatives is first decomposed into several first-order equations to improve the numerical stability, reduce the computational complexity and easily enforce the Neumann boundary condition. Then, each first-order derivative is solved by the FPM repeatedly. Moreover, the MPI parallel technique is introduced to enhance the computational efficiency, and an appropriate Wendland kernel function is employed which has higher accuracy than the Gaussian function. 3D transient heat conduction problems with different boundary conditions, including the case in a complex cylindrical domain, are investigated and compared with the analytical solutions to illustrate the flexibility and the accuracy of the parallel I-FPM-3D. The convergence and the computational efficiency of the parallel I-FPM-3D are also analyzed. Finally, the temperature field in a 3D functional gradient material is predicted by the parallel I-FPM-3D and compared with the other numerical results. The numerical results indicate that the temperature variation process in the functional gradient materials can be visualized accurately. [ABSTRACT FROM AUTHOR]

Fuzzy C-means clustering integration algorithm is a method to improve clustering quality by using integration ideas, but as the amount of data increases, its time complexity increases. A parallel FCM clustering integration algorithm based on MapReduce is proposed. The algorithm uses a random initial clustering centre to obtain differentiated cluster members. By establishing an overlapping matrix between clusters, the clustering labels are unified to find logical equivalence clusters. The cluster members share the classification information of the data objects by voting to obtain the final clustering result. The experimental results show that the parallel FCM clustering integration algorithm has good performance, and has high speedup and good scalability. [ABSTRACT FROM AUTHOR]

Wireless sensor network (WSN) can effectively help us monitor the surrounding environment and prevent the occurrence of some natural disasters earlier, but we can only get the information of the surrounding environment correctly if we know the locations of nodes. How to know the exact positions of nodes is a strict challenge in WSN. Intelligent computing algorithms have been developed in recent years. They easily solve complex optimization problems, especially for those that cannot be modeled mathematically. This paper proposes a novel algorithm, named parallel whale optimization algorithm (PWOA). It contains two information exchange strategies between groups, and it significantly enhances global search ability and population diversity of the original whale optimization algorithm (WOA). Also, the algorithm is adopted to optimize the localization of WSN. Twenty-three mathematical optimization functions are accustomed to verifying the efficiency and effectiveness of the novel approach. Compared with some existing intelligent computing algorithms, the proposed PWOA may reach better results. [ABSTRACT FROM AUTHOR]

Video processing applications often need high computing capacity but have performance and power constraints, especially in portable devices. General purpose processors can no longer meet the requirements. This paper presents a parallel reconfigurable computing architecture consisting of reconfigurable processing units connected by an area-efficient routing. The hierarchical configuration contexts can cut the implementation overhead and the energy dissipation spent on fast reconfiguration. The proposed architecture targets multiple-standard video processing. The design is able to give high performance comparable to the fixed-function ASIC through deep pipelining and a large amount of computing parallelism. The experimental results show the proposed architecture has great performance and practicability. [ABSTRACT FROM AUTHOR]

This study proposes a set of new robust parallel hybrid metaheuristic algorithms based on artificial bee colony (ABC) and teaching learning-based optimization (TLBO) for the multi-dimensional numerical problems. The best practices of ABC and TLBO are implemented to provide robust algorithms on a distributed memory computation environment using MPI libraries. Island parallel versions of the proposed hybrid algorithm are observed to obtain much better results than those of sequential versions. Parallel pseudorandom number generators are used to provide diverse solution candidates to prevent stagnation into local optima. The performances of the proposed hybrid algorithms are compared with eight different metaheuristics algorithms of particle swarm optimization, differential evolution variants, ABC variants and evolutionary algorithm. The empirical results show that the new hybrid parallel algorithms are scalable and the best performing algorithms when compared to the state-of-the-art metaheuristics. [ABSTRACT FROM AUTHOR]

Superconducting fault current limiters are highly effective solutions to reduce and control excessive fault current which is inevitable in electric power systems. Resistive type superconducting fault current limiter (RSFCL) is used in parallel by multiple superconducting units, but there is a problem of uneven current distribution in parallel, which seriously affects its reliability and safety. In order to explore the problem of current sharing in parallel and find out the law of current distribution, a series of experiments are designed to verify it. Experimental research has found that whether in the steady state or the transient state, the more the superconducting units are connected in parallel, the smaller the current difference is, but it does not mean that the current sharing is better. This paper introduces a new evaluation model of current sharing, which evaluates the current sharing effect of different numbers of parallel situations in various states, and finds the law of current sharing, which provides a reference for application research of the subsequent. [ABSTRACT FROM AUTHOR]

General sparse matrix-sparse matrix (SpGEMM) multiplication is one of the basic kernels in a great many applications. Several works focus on various optimizations for SpGEMM. To fully exploit the powerful computing capability of the Sunway TaihuLight supercomputer for SpGEMM, this paper designs the partitioning method and parallelization of CSR-based SpGEMM to make it well match to the Sunway architecture. In addition, this paper optimizes the partitioning method based on the distribution of the floating-point calculations of the CSR-based SpGEMM to achieve the load balance and performance improvement on the Sunway. We, respectively, analyze the performance, including the memory footprint and the execution time, of the parallel CSR-based SpGEMM and the optimized CSR-based SpGEMM on the Sunway. The experimental results show that the optimized CSR-based SpGEMM outperforms over the parallel CSR-based SpGEMM and has good scalability on the Sunway. [ABSTRACT FROM AUTHOR]

In this paper, a parallel SSD (Single Shot MultiBox Detector) fusion network based on inverted residual structure (IR-PSN) is proposed to solve the problems of the lack of extracted feature information and the unsatisfactory effect of small object detection by deep learning. Firstly, the Inverted Residual Structure (IR) is adopted into the SSD network to replace the pooling layer. The improved SSD network is called deep network of IR-PSN to extract high-level feature information of the image. Secondly, a shallow network based on the inverted residual structure is constructed to extract low-level feature information of the image. Finally, the shallow network is fused with the deep network to avoid the lack of small object feature information and improve the detection rate of small object. The experimental results show that the proposed method has satisfied results for small object detection under the premise of ensuring the accuracy rate P and recall rate R of the comprehensive object detection. [ABSTRACT FROM AUTHOR]

A novel parallelization method of genetic algorithm (GA) solution of the Traveling Salesman Problem (TSP) is presented. The proposed method can considerably accelerate the solution of the equivalent TSP of many complex vehicle routing problems (VRPs) in the cloud implementation of intelligent transportation systems. The solution provides routing information besides all the services required by the autonomous vehicles in vehicular clouds. GA is considered as an important class of evolutionary algorithms that can solve optimization problems in growing intelligent transport systems. But, to meet time criteria in time-constrained problems of intelligent transportation systems like routing and controlling the autonomous vehicles, a highly parallelizable GA is needed. The proposed method parallelizes the GA by designing three concurrent kernels, each of which running some dependent effective operators of GA. It can be straightforwardly adapted to run on many-core and multi-core processors. To best use the valuable resources of such processors in parallel execution of the GA, threads that run any of the triple kernels are synchronized by a low-cost switching mechanism. The proposed method was experimented for parallelizing a GA-based solution of TSP over multi-core and many-core systems. The results confirm the efficiency of the proposed method for parallelizing GAs on many-core as well as on multi-core systems. [ABSTRACT FROM AUTHOR]

In the last decade, parallel computing techniques become more popular and offer various architectures to perform diverse processes in an individual computing node or a collection of strongly linked nodes with homogeneous hardware. In this paper, we propose a distributed secure unequal cluster-based multipath routing protocol (USCDRP) for improving the energy efficiency and also to enhance the security as well as reliability. The presented USCDRP is the enhanced work of SCMRP, which provides the security to a particular extent. The presented method offers reliability, energy efficiency and maximum lifetime. Besides, an independent clustering method is provided to eliminate the orphan nodes and cluster maintenance strategy to enhance the network lifetime. A detailed experimental analysis ensured the superiority of the USCDRP over the compared methods concerning throughput, energy efficiency and network lifetime analysis. The simulation outcome indicated that all the nodes in WSN become dead node, and the network becomes inactive in the round number of 2021 by Sec-LEACH, 2412 by SCMRP and 3245 by USCDRP. [ABSTRACT FROM AUTHOR]

Recently, many evolutionary algorithms (EAs) have been used to solve clustering problem. However, compared to K-means which is a simple and fast clustering algorithm, these EA-based clustering algorithms take too much computation time. In addition, the parameters of most EAs are fixed or dynamical adjustment by a simple method on different datasets, and it will cause that the performance of these algorithms is good on some datasets but bad on others. In order to overcome these disadvantages, a novel parallel object-tracking behavior algorithm (POTBA) based on dynamics is proposed in this paper. The proposed algorithm consists of three different models which are parallel object-tracking model, parameters self-learning model and energy model, respectively. First, the parallel object-tracking model is designed to accelerate the computation speed and avoid local minima. Second, the parameters of POTBA are self-adjusted by the parameters self-learning model. Third, the energy model is introduced to depict energy changes of POTBA during the evolutionary process. The correctness and convergence properties of POTBA are analyzed theoretically. Moreover, the effectiveness and parallelism of POTBA are evaluated through several standard datasets, and the experimental results demonstrate that POTBA exhibits superior overall performance than five other state-of-the-art algorithms. In the aspect of search performance, the results of POTBA are better than other comparison algorithms on most used datasets. In the aspect of time performance, the time overhead of POTBA is significantly reduced through parallel computing. When the number of processors increases to 32, the computation time of POTBA is less or close to K-means which is the fastest comparison algorithm. [ABSTRACT FROM AUTHOR]

Evaluation of a robotic (serial or parallel) manipulator’s static and kinematics performance requires the solution of closed and complicated non-closed form systems of equations for both the forward and inverse problem types. In this work, a robotic system or manipulator is represented as a network whose motion generating kinematic pairs are represented as the network’s inter-connected nodes. Within this network theoretic context, we develop a formulation employing higher dimensional multivectors defined in Clifford Algebra that approximates the computational outcomes of such complicated systems of equations, for both inverse and forward problem types. The statics and kinematics performance of these mechanical networks (serial and parallel robots) are rapidly evaluated without the need to solve their respective traditional systems of equations using what we call ‘screw hypervolumes’, which capture the algebro-geometric structures generated by a robot’s physical design parameters and end-effector position in terms of kinematics and statics. These dimensionless hypervolume functions enable the specialist to produce a very computationally efficient function/index that approximates the kinematic and static output spaces of robots. We find that this formulation inadvertently presents a curious mathematical analogue to these systems of equations. A 6R serial robot, the Delta robot and the 3-RRR manipulator are analysed using the hypervolume methods for the case studies. This approach can be adapted and applied to other multi-agent physical systems displaying non-commutative interactions as well. [ABSTRACT FROM AUTHOR]

In this work, for solving large sparse vertical linear complementarity problems, a relaxation two-step modulus-based synchronous multisplitting method without auxiliary variable is constructed. The convergence conditions are presented for the proposed method, where the range of the relaxation parameters is obtained. By OpenACC framework, the proposed method is shown to be more efficient than the existing method with high speedups with numerical tests. [ABSTRACT FROM AUTHOR]

The Uintah computational framework is used for the parallel solution of partial differential equations on adaptive mesh refinement grids using modern supercomputers. Uintah is structured with an application layer and a separate runtime system. Uintah is based on a distributed directed acyclic graph of computational tasks, with a task scheduler that efficiently schedules and executes these tasks on both CPU cores and on-node accelerators. The runtime system identifies task dependencies, creates a task graph prior to the execution of these tasks, automatically generates MPI message tags, and automatically performs halo transfers for simulation variables. Automating halo transfers in a heterogeneous environment poses significant challenges when tasks compute within a few milliseconds, as runtime overhead affects wall time execution, or when simulation variables require large halos spanning most or all of the computational domain, as task dependencies become expensive to process. These challenges are magnified at production scale when application developers require each compute node perform thousands of different halo transfers among thousands simulation variables. The principal contribution of this work is to (1) identify and address inefficiencies that arise when mapping tasks onto the GPU in the presence of automated halo transfers, (2) implement new schemes to reduce runtime system overhead, (3) minimize application developer involvement with the runtime, and (4) show overhead reduction results from these improvements. [ABSTRACT FROM AUTHOR]

As an important machine learning method, reinforcement learning plays a more and more important role in practical application. In recent years, many scholars have studied parallel reinforcement learning algorithm, and achieved remarkable results in many applications. However, when using existing parallel reinforcement learning to solve problems, due to the limited search scope of agents, it often fails to reduce the running episodes of algorithms. At the same time, the traditional model-free reinforcement learning algorithm does not necessarily converge to the optimal solution, which may lead to some waste of resources in practical applications. In view of these problems, we apply Particle swarm optimization (PSO) algorithm to asynchronous reinforcement learning algorithm to search for the optimal solution. First, we propose a new asynchronous variant of PSO algorithm. Then we apply it into asynchronous reinforcement learning algorithm, and proposed a new asynchronous reinforcement learning algorithm named Sarsa algorithm based on backward Q-learning and asynchronous particle swarm optimization (APSO-BQSA). Finally, we verify the effectiveness of the asynchronous PSO and APSO-BQSA algorithm proposed in this paper through experiments. [ABSTRACT FROM AUTHOR]

Particle swarm optimization (PSO) usually requires a large number of fitness evaluations to obtain a sufficiently good solution, which poses an obstacle for applying PSO to computationally expensive problems. In this paper, a multiple surrogates based PSO (MSPSO) framework is proposed, which consists of an inner loop optimization and an outer one. In the outer loop optimization, a PSO algorithm is used in both the optimization mode and the sampling one. In the inner loop optimization, a multiple surrogate based parallel optimization strategy is designed. Furthermore, the search history and the possible solutions from the outer loop optimization are provided for the inner one, and the result of the inner loop optimization is employed to guide the search of the outer one. To verify the performance of the proposed approach, a number of numerical experiments are conducted by using ten benchmark test functions and three time series regression modeling problems. The results indicate that the proposed framework is capable of converging to a good solution for the low-dimensional, non-convex and multimodal problems. [ABSTRACT FROM AUTHOR]

Recently, energy efficiency has gained attention from researchers interested in optimizing computing resources. Solving real-world problems using optimization techniques (such as metaheuristics) requires a large number of computing resources and time, consuming an enormous amount of energy. However, only a few and limited research efforts in studying the energy consumption of metaheuristics can be found in the existing literature. In particular, genetic algorithms (GAs) are being used so widely to solve a large range of problems in scientific and real-world problems, but hardly found explained in their internal consumption behavior. In the present article, we analyze the energy consumption behavior of such techniques to offer a useful set of findings to researchers in the mentioned domains. We expand our study to include several algorithms and different problems and target the components of the algorithms so that the results are still more appealing for researchers in arbitrary domains of application. Our experiments on the sequential GAs show the controlling role of the fitness operator on energy consumption and also reveal possible energy hot spots in GAs operations, such as mutation operator. Further, our distributed evaluations besides a statistical analysis of the results demonstrate that the communication scheme could highly affect the energy consumption of the parallel evaluations of the GAs. [ABSTRACT FROM AUTHOR]

The parallelization of particle swarm optimization (PSO) is an efficient way to improve the performance of PSO. The multiple population parallelization is one way to parallelize PSO, in which three parameters need to be manually set in advance. They are migration interval, migration rate, and migration direction, which decide when, how many and from which subpopulation to which subpopulation particles will be migrated, respectively. However, there are two shortcomings concerning manually setting these three parameters in advance. One is that good particles cannot be migrated in time since particles can only be migrated every a given interval and in a given direction in parallel PSO. The other is that a large number of unnecessary migrations will take place since a given rate of particles in each subpopulation will be migrated every a given interval in a given direction. Both may be bad for parallel PSO to find high-quality solutions as quickly as possible, and this will result in a huge communication cost. Inspired by the phenomenon of osmosis, this paper presents a multiple population parallel version of PSO based on osmosis. It can adaptively decide when, how many, and from which subpopulation to which subpopulation particles will be migrated. Its usefulness, especially for high-dimensional functions, is demonstrated by numerical experiments. [ABSTRACT FROM AUTHOR]

This paper considers a tollbooth system with two parallel heterogeneous servers and two vehicle types (say, cars and trucks), where one server collects tolls from vehicles of both types and the other only serves one type of them. With such characteristic, the system is referred to as "skill-based servers" for brevity in this paper. Meanwhile, vehicles are accommodated in a single common lane and get served on the "global First-Come-First-Served" basis. In fact, such a system and its variants are commonly encountered and their performance measures are of great significance to managers. We first develop a Quasi-Birth-Death process to explicitly model this tollbooth system. Then by applying spectral expansion technique, we derive the stationary probabilities for the computation of the system's performance measures, such as mean queue size and the idle probability of each server. The Laplace–Stieltjes transforms of an arbitrary vehicle's sojourn time is derived as well. Finally, numerical results are presented to show the impact of parameters' selection upon performance measures, and they have intriguing managerial implications. The results also reveal that the tollbooth system with skill-based servers is much more efficient compared to the system with dedicated servers, which has been studied by Mélange et al. (Comput Oper Res 71:23–33, 2016), especially in rush hours. [ABSTRACT FROM AUTHOR]

In this work we give a characterization of pseudo-parallel surfaces in S c n × R and H c n × R , extending an analogous result by Asperti-Lobos-Mercuri for the pseudo-parallel case in space forms. Moreover, when n = 3 , we prove that any pseudo-parallel surface has flat normal bundle. We also give examples of pseudo-parallel surfaces which are neither semi-parallel nor pseudo-parallel surfaces in a slice. Finally, when n ≥ 4 we give examples of pseudo-parallel surfaces with non vanishing normal curvature. [ABSTRACT FROM AUTHOR]

We first briefly report on the status and recent achievements of the ELPA-AEO (Eigen value Solvers for Petaflop Applications—Algorithmic Extensions and Optimizations) and ESSEX II (Equipping Sparse Solvers for Exascale) projects. In both collaboratory efforts, scientists from the application areas, mathematicians, and computer scientists work together to develop and make available efficient highly parallel methods for the solution of eigenvalue problems. Then we focus on a topic addressed in both projects, the use of mixed precision computations to enhance efficiency. We give a more detailed description of our approaches for benefiting from either lower or higher precision in three selected contexts and of the results thus obtained. [ABSTRACT FROM AUTHOR]

In the automata theoretic approach to explicit state LTL model checking, the synchronized product of the model and an automaton that represents the negated formula is checked for emptiness. In practice, a (transition-based generalized) Büchi automaton (TGBA) is used for this procedure. This paper investigates whether using a more general form of acceptance, namely a transition-based generalized Rabin automaton (TGRA), improves the model checking procedure. TGRAs can have significantly fewer states than TGBAs; however, the corresponding emptiness checking procedure is more involved. With recent advances in probabilistic model checking and LTL to TGRA translators, it is only natural to ask whether checking a TGRA directly is more advantageous in practice. We designed a multi-core TGRA checking algorithm and performed experiments on a subset of the models and formulas from the 2015 Model Checking Contest and generated LTL formulas for models from the BEEM database. While we found little to no improvement by checking TGRAs directly, we show how various aspects of a TGRA's structure influences the model checking performance. In this paper, we also introduce a Fin-less acceptance condition, which is a disjunction of TGBAs. We show how to convert TGRAs into automata with Fin-less acceptance and show how a TGBA emptiness procedure can be extended to check Fin-less automata. [ABSTRACT FROM AUTHOR]

While we cannot efficiently emulate quantum algorithms on classical architectures, we can move the weight of complexity from time to hardware resources. This paper describes a proposition of a universal and scalable quantum computer emulator, in which the FPGA hardware emulates the behavior of a real quantum system, capable of running quantum algorithms while maintaining their natural time complexity. The article also shows the proposed quantum emulator architecture, exposing a standard programming interface, and working results of an implementation of an exemplary quantum algorithm. [ABSTRACT FROM AUTHOR]