检索结果-内蒙古大学图书馆

international symposium on parallel architectures, algorithms and Networks (ISPAN)

作者： T. Touyama S. Horiguchi Department of Electrical and Electronics Engineering Nippon Institute of Technology Miyashiro Saitama Japan Japan Advanced Institute of Science and Technology Graduate School of Information Science Tatsunokuchi Ishikawa Japan

Massively parallel computers consisting of a large number of processing elements have been developed and expected as high performance computers in advanced science and technology. Practical parallel computation model has been required to analyze parallel algorithms on massively parallel computers. We present a practical parallel computation model LogPQ taking account of communication queues into the LogP model. The LogPQ model has three queues for each communication line, and four supplement parameters in addition to the LogP model. This paper addresses the performance of parallel matrix multiplication using the LogPQ model. The parallel performances on the parallel machine CM-5 are compared between the LogP and LogPQ model. It is seen that the LogPQ model expects the execution times more accurately than the LogP model.

关键词： Concurrent computing Computational modeling High performance computing parallel algorithms Delay Message passing Algorithm design and analysis parallel machines parallel programming Bandwidth

来源：评论

学校读者我要写书评

暂无评论

Automatic software generation in the OORHS distributed computing framework

Automatic software generation in the OORHS distributed compu...

引用

international symposium on parallel architectures, algorithms and Networks (ISPAN)

作者： D.S. Yeung A.K.Y. Wong Department of Computing Hong Kong Polytechnic University Hong Kong China

Software generation in the OORHS (object-oriented reciprocative hypercomputing system) is user-transparent. It addresses the issue of ease of use by minimizing the number of steps leading to a programming solution. The OORHS requires from the user only a high level APPL program, which is, in effect, a specification. For every APPL program, the system automatically performs all the necessary distributed computing steps. The precompiler, based on the object-oriented paradigm, instantiates the encapsulated program objects embedded in an APPL program. These program objects are distributed at the source level. They are compiled and then executed at the allocated sites. This unique approach, known as local compilation, eliminates the need to store the compilers used by other machines locally. It enhances the compatibility between the compiled program and the host processor. The precompiler generates a program objects dictionary for every APPL program. The contents in the dictionary facilitates program visualization.

关键词： Distributed computing Dictionaries Software libraries Load management parallel programming Visualization Embryo Fault tolerance Global communication parallel processing

来源：评论

学校读者我要写书评

暂无评论

BSP pro: Java-based BSP performance profiling system

Proceedings of the International Symposium on Parallel Archi...

引用

Proceedings of the international symposium on parallel architectures, algorithms and Networks, I-SPAN 1999年 54-59页

作者： Zheng, Weiqun Khan, Shamim Xie, Hong Murdoch Univ Perth

This paper introduces a new parallel performance profiling system for the Bulk Synchronous parallel (BSP) model. The profiling system, called BSP Pro, consists of a performance profiling tool, BSP Profiler, and a performance visualisation tool, BSP Visualiser. The aim of BSP Pro is to assist in the analysis and improvement of BSP program performance by minimising load imbalance among processes. BSP Pro is different from other systems, such as the profiling tools within the Oxford BSP toolset, in terms of both its features and its implementation. It uses BSP Profiler to trace and generate more comprehensive profiling information resulting from BSP program executions. The profiling information is then visualised and shown as performance profiling graphs using BSP Visualiser. The visualising component of BSP Pro is fully developed in Java and utilises Java graphics to expose and highlight process load imbalance in both computation and interprocess communication.

关键词： Java programming language

来源：评论

学校读者我要写书评

暂无评论

Mapping YOLOv4-Tiny on FPGA-Based DNN Accelerator by Using Dynamic Fixed-Point Method

Mapping YOLOv4-Tiny on FPGA-Based DNN Accelerator by Using D...

引用

international symposium on parallel architectures, algorithms and programming (PAAP)

作者： Peng Li Cheng Che State Key Laboratory of Information Security Institute of Information Engineering Chinese Academy of Sciences Beijing China School of Cyber Security University of Chinese Academy of Sciences Beijing China

ISBN: (纸本)9781665496407

In the past few decades, with the large-scale application of deep learning technology, the neural network inference speed problem is becoming more and more severe, especially in some mobile and embedded real-time processing systems. To improve the inference speed, some software optimization techniques have been proposed by previous researchers. However, these methods only focused on the model self and still cannot meet the rapid development of large and complex deep learning models. Fortunately, deep neural network (DNN) accelerators based on FPGA SoC has opened a promising opportunity for the real-time inference. In this paper, we proposed a novel 16-bit dynamic fixed-point number quantization method to map the object detection network YOLOv4-tiny into FPGA-based heterogeneous deep learning accelerators. We evaluate this model in Xilinx Zynq-7020 SoC on the ZedBoard platform. Experiment on coco dataset shows that this approach can improve the model inference speed to 4× than pure CPU platform, and the accuracy loss is negligible with only 3-5%. This mapping method can be competitive with or even superior to other state-of-the-art accelerating method in deep learning fields.

关键词： Deep learning Quantization (signal) Neural networks programming Real-time systems Software parallel architectures

来源：评论

学校读者我要写书评

暂无评论

The MULTIPLUS/MULPLIX parallel processing environment

The MULTIPLUS/MULPLIX parallel processing environment

引用

international symposium on parallel architectures, algorithms and Networks (ISPAN)

作者： J.S. Aude A.M. Meslin C.M.P. Santos G. Bronstein L.F.M. Cordeiro M.O. Barros M. Joao S.C. Oliveira NCE/UFRJ Federal University of Rio de Janeiro Rio de Janeiro Rio de Janeiro Brazil NCE/UFRJ Federal Univ. of Rio de Janeiro Brazil

The MULTIPLUS project aims at the development of a modular parallel architecture suitable for the study of several aspects of parallelism in both true shared memory and virtual shared memory environments. The MULTIPLUS architecture is able to support up to 1024 Processing Elements based on SPARC microprocessors. The MULPLIX Unix-like operating system offers a suitable parallel programming environment for the MULTIPLUS architecture by providing facilities for the creation of threads, the allocation of private and shared memory space and the efficient use of synchronization primitives. After presenting the main features of the MULTIPLUS architecture and of the MULPLIX operating system, the paper describes in detail the design and the implementation of the three MULTIPLUS architecture basic hardware modules: the Processing Element, the Multistage Interconnection Network and the I/O Processor. In addition, the definition of the MULPLIX parallel programming primitives is discussed and their use is illustrated through an example. Finally, future directions in the development of the MULTIPLUS research project are commented.

关键词： parallel processing Computer architecture Operating systems parallel programming Multiprocessor interconnection networks Concurrent computing parallel architectures Microprocessors Yarn Hardware

来源：评论

学校读者我要写书评

暂无评论

Graph-based Multi-view Partial Multi-label Learning

Graph-based Multi-view Partial Multi-label Learning

引用

international symposium on parallel architectures, algorithms and programming (PAAP)

作者： Wei Liu Songhe Feng Hui Tian School of Computer and Information Technology Beijing Jiaotong University Beijing China China Mobile Xiongan Information Communication Technologies Ltd China Mobile System Integration Communication. Ltd Beijing China

ISBN: (纸本)9781665452199

In multi-view partial multi-label learning (MVPML) problem, each instance is described by several heterogeneous feature representations and associated with a set of candidate labels, which include both ground-truth and noisy labels. The key to learn from MVPML data lies in how to deal with multi-view data and how to select the ground-truth labels from candidate label set. In this paper, we propose a Graph-based Multi-view Partial Multi-label method, which integrates exploiting multi-view information, noisy label disambiguation and training predictor model into a whole framework. Specifically, we first exploit the consensus information across different views by learning the similarity graph of each view and fuses these similarity graphs into a unified graph. Secondly, we decompose the observed label set into a ground-truth label matrix and a noisy label matrix, where the noisy label matrix is assumed to be sparse. Then, we embed the learned unified similarity graph into the process of label disambiguation to obtain a more reliable ground-truth label matrix. Finally, the predictive model is learned by the ground-truth label matrix. Extensive experiments indicate that our proposed method can achieve superior or comparable performance against state-of-the-art methods.

关键词： Training Fuses Predictive models programming Prediction algorithms parallel architectures Sparse matrices

来源：评论

学校读者我要写书评

暂无评论

A Case Study of Container Behavior Analysis with Cloud Resource Profile

A Case Study of Container Behavior Analysis with Cloud Resou...

引用

international symposium on parallel architectures, algorithms and programming (PAAP)

作者： Dunbo Cai Ling Qian Dongqin Xu Zhiguo Huang Center for Technology Research & Innovation China Mobile (Suzhou) Software Technology Co. Ltd. Suzhou China

ISBN: (纸本)9781665496407

In the field of cloud computing, a container is a lightweight abstraction of an operating system and the software application running for a particular service. To select a machine to deploy the container, the scheduling component of a cloud computing system usually considers the behavior of the container for decision. Several recent studies have shown that the prediction of the future behavior of a container can provide helpful guidance for scheduling. In this paper, we exploit the Alibaba cloud scheduling trace data which was generated on the year 2018 and named cluster-trace-v2018, to analyse containers behavior. For cluster-trace-v2018, we report our method for the prepossessing, container behavior feature engineering, and behavior classification. Specifically, we developed a processing procedure to filter out incomplete records. We designed a method for building behavior feature vector for containers on 12-hours long and 24-hours long periods. And, based on these features, we utilize the K-Means method to cluster the containers into a specified number of clusters. Experimenters on varying length of container feature vector show that the containers fits well into 10 clusters.

关键词： Cloud computing Processor scheduling Buildings Pipelines Clustering algorithms Containers programming

来源：评论

学校读者我要写书评

暂无评论

3rd international symposium on High Performance Computing, ISHPC 2000

3rd International Symposium on High Performance Computing, I...

引用

3rd international symposium on High Performance Computing, ISHPC 2000

ISBN: (纸本)9783540411284

The proceedings contain 56 papers. The special focus in this conference is on Compilers, architectures, Evaluation, algorithms, Models and Applications. The topics include: An object-based programming and execution model for processor-in-memory arrays;limits of task-based parallelism in irregular applications;the case for speculative multithreading on SMT processors;skewed data partition and alignment techniques for compiling programs on distributed memory multicomputers;processor mechanisms for software shared memory;an evaluation of page aggregation technique on different DSM systems;partitioned parallel radix sort;transonic wing shape optimization based on evolutionary algorithms;on performance modeling for HPF applications with ASL;performance sensitivity of routing algorithms to failures in networks of workstations;decentralized load balancing in multi-node broadcast schemes for hypercubes;design and implementation of an efficient thread partitioning algorithm;a flexible routing scheme for networks of workstations;java bytecode optimization with advanced instruction folding mechanism;performance evaluation of a java based chat system;on the influence of the selection function on the performance of networks of workstations;combining in-transit buffers with optimized routing schemes to boost the performance of networks with source routing;a comparison of locality-based and recency-based replacement policies;at our management comparison with related split data cache schemes sensitive to data localities;global magneto-hydrodynamic simulations of differentially rotating accretion disk by astrophysical rotational plasma simulator;exploring multi-level parallelism in cellular automata networks and functional BSP programs on enumerated vectors.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Accelerating GNN Inference by Soft Channel Pruning

Accelerating GNN Inference by Soft Channel Pruning

引用

international symposium on parallel architectures, algorithms and programming (PAAP)

作者： Wenbo Zhang Jingwei Sun Guangzhong Sun University of Science and Technology of China

ISBN: (纸本)9781665452199

Graph Neural Networks (GNNs) are effective models for processing graph-structured data. With the continuous growth of graph data scale and the deepening of graph neural network layers, the heavy cost of GNN inference has greatly limited its application in real-time tasks. This paper focus on accelerating the performance of GNN inference. We first measures the execution time ratio of each stage in the inference process for commonly used GNN models, and analyzes the performance characteristics of different stages. We find out that the critical performance factor of GNN inference is the feature dimension, which is different to CNN and NLP models. Therefore, we propose a soft channel pruning method with a ladder pruning pattern. It reduces the calculation from unimportant graph node features and achieve performance acceleration. Meanwhile, it reserves inference accuracy of GNNs. According to experimental validation on graph datasets of different scales, our method can effectively reduce the inference latency and achieve 2×–7× speedup. Also, compared with existing pruning methods, higher inference accuracy can be obtained with comparable speedup ratio.

关键词： Costs Graphics processing units programming Graph neural networks Time measurement Real-time systems Inference algorithms

来源：评论

学校读者我要写书评

暂无评论

Radix R parallel FFT algorithms with a global interconnection networks and its evaluation

Radix R parallel FFT algorithms with a global interconnectio...

引用

international symposium on parallel architectures, algorithms and Networks (ISPAN)

作者： T. Taketa K. Tanno S. Horiguchi Department of Electrical and Information Engineering Yamagata University Japan School of Information Science Japan Advanced Institute of Science and Technology Japan

In order to efficiently compute Fast Fourier transform (FFT) various parallel algorithms and their implementation to multiprocessors and multicomputers have been developed. In general, the local interconnection network is more high speed than a global one, but its capability depends on network architecture. On the other hand, the global interconnection network is not so high speed, but it does not depends on network architecture. It provides a flexible communication interface to the programmer. In this paper, we discuss parallel radix R FFT algorithms on a multiprocessor or multicomputer system with a global interconnection network. We propose two algorithms a stage-by-stage method and a multi-stage method. We also estimate the communication time. Then we show that the communication time is very sensitive to and affected by data exchange strategy. Finally, we implement these algorithms on two commercial massively parallel computers (nCUBE/2 and CM5) and measure the communication time.

关键词： Multiprocessor interconnection networks Signal processing algorithms Concurrent computing Computer networks Computer architecture Clustering algorithms Fast Fourier transforms parallel algorithms programming profession Information science

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：