检索结果-内蒙古大学图书馆

Proceedings of the 28th ACM SIGPLAN Annual Symposium on Principles and Practice of parallel Programming

作者： Fei Dai Yawen Chen Zhiyi Huang Haibo Zhang Fangfang Zhang University of Otago Dunedin New Zealand Qilu University of Technology Jinan China

ISBN: (纸本)9798400700156

All-reduce is the crucial communication primitive to reduce model parameters in distributed Deep Neural Networks (DNN) training. Most existing all-reduce algorithms are designed for traditional electrical interconnect systems, which cannot meet the communication requirements for distributed training of large DNNs due to the low data bandwidth of the electrical interconnect systems. One of the promising alternatives for electrical interconnect is optical interconnect, which can provide high bandwidth, low transmission delay, and low power cost. We propose an efficient scheme called WRHT (Wavelength Reused Hierarchical Tree) for implementing all-reduce operation in optical interconnect systems. WRHT can take advantage of WDM (Wavelength Division Multiplexing) to reduce the communication time of distributed data-parallel DNN training. Simulations using real DNN models show that, compared to all-reduce algorithms in the electrical and optical network systems, our approach reduces communication time by 75.76% and 91.86%, respectively.

关键词： optical interconnects distributed DNN training WDM all-reduce

来源：评论

学校读者我要写书评

暂无评论

Joint Computation and Communication Analysis of Hard real-time Applications in Manycores

Joint Computation and Communication Analysis of Hard Real-Ti...

引用

Symposium on Integrated Circuits and systems Design (SBCCI)

作者： Anderson R. P. Domingues Lucas Damo Sergio Johann Filho Fernando Gehm Moraes School of Technology Pontifical Catholic University of Rio Grande do Sul – PUCRS Porto Alegre Brazil

ISBN: (数字)9798350391695

ISBN: (纸本)9798350391701

Manycores appeared as a scalable computing infrastructure to support massively parallel applications. Due to the advancements of on-chip technology, manycores present an attractive trade-off between power consumption, area, and fabrication cost. Due to the inherent distributed computing model of manycores, predicting the behavior of modern, mission-critical applications becomes challenging. On the one hand, the community introduced real-time (RT) networks-on-chips (NoC) to support the RT analysis of communication in manycores. On the other hand, implementing non-functional requirements (NFRs) adds complexity to the design, making RT-NoCs prohibitively in resource-constrained scenarios. This work presents a computation and communication approach to RT analysis in manycores. Our approach suits systems without RT-NoCs, enabling manycores equipped with simple low-cost NoCs to run RT applications. We show a proof-of-concept where we can reduce the frequency of the baseline manycore to fit the needs of the target application. Results demonstrate that our approach guarantees RT without modifying the baseline hardware.

关键词： Integrated circuits Fabrication Power demand Mission critical systems Predictive models real-time systems Hardware System-on-chip distributed computing System analysis and design

来源：评论

学校读者我要写书评

暂无评论

Throughput and Memory Optimization for parallel Implementations of Dataflow Networks Using Multi-Reader Buffers 4

Throughput and Memory Optimization for Parallel Implementati...

引用

4th workshop on Next Generation real-time Embedded systems, NG-RES 2023

作者： Letras, Martin Falk, Joachim Teich, Jürgen Germany

ISBN: (纸本)9783959772686

In this paper, we introduce the concept of Multi-Reader Buffers (MRBs) for high throughput and memory-efficient implementation of dataflow applications. Our work is motivated by the huge amount of data that needs to be processed and typically accessed in a FIFO manner, particularly in image and video processing applications. Here, multi-cast, fork, and merge operator implementations known today produce huge memory overheads by storing and communicating copies of the same data. As a remedy, we first introduce MRBs as buffers preserving FIFO semantics for a finite number of readers of the same data while storing each data item only once. Second, we present an approach for memory minimization of data flow networks by replacing all multi-cast actors and connected FIFOs with MRBs. Third, we present a Design Space Exploration approach to selectively replace multi-cast actors with MRBs in order to explore memory, throughput, and processor resource allocation tradeoffs. Our results show that the explored Pareto fronts of our approach improve the solution quality over a reference by 78 % in average for six benchmark applications in terms of a hypervolume indicator. © Martin Letras, Joachim Falk, and Jürgen Teich.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Software-defined event-based camera and vision algorithms for digitizing highly dynamic manufacturing tasks

Software-defined event-based camera and vision algorithms fo...

引用

2024 Low-Cost Digital Solutions for Industrial Automation, LoDiSA 2024

作者： Walker, Emily Turner, Christopher J. Oyekan, John Department of Automatic Control & Systems Engineering The University of Sheffield SheffieldS1 3JD United Kingdom Surrey Business School The University of Surrey Surrey GuildfordGU2 7XH United Kingdom Real Time & Distributed Systems Group Department of Computer Science University of York Heslington YorkYO10 5GH United Kingdom

ISBN: (纸本)9781837242672

Event-based cameras are inspired by the way the retina in the eye process information. Rather than capturing image frames at a constant rate, event cameras generate data asynchronously, based on changes in pixel value between frames. As a result, these cameras have a high spatiotemporal accuracy, low power constraints and reduced artefacts from motion blur. Hence, they represent favourable properties for digitizing highly dynamic manufacturing tasks such as machine vibrations and rapid human assembly actions. However, the costs of event-based cameras are prohibitively high for small to medium sized enterprises (SMEs). In this research, we present the development of a rig that combines low-cost visual sensors with low-cost algorithms to perform action recognition for highly dynamic manual manufacturing tasks. We applied our algorithms to achieve an action classification system via a novel threshold-based approach and in parallel use a neural network to perform classification of objects in the video feed. The results achieved demonstrated that our approach was able to successfully identify various rapid manual tasks being performed on identified objects. In addition to SMEs, our approach would be suited for providing low-cost digitization on legacy machines. © This is an open access article published by the IET under the Creative Commons Attribution License (http://***/licenses/by/3.0/).

关键词： Machine vibrations

来源：评论

学校读者我要写书评

暂无评论

BestNeighbor: efficient evaluation of kNN queries on large time series databases

引用

KNOWLEDGE AND INFORMATION systems 2021年第2期63卷 349-378页

作者： Levchenko, Oleksandra Kolev, Boyan Yagoubi, Djamel-Edine Akbarinia, Reza Masseglia, Florent Palpanas, Themis Shasha, Dennis Valduriez, Patrick Univ Montpellier Inria CNRS LIRMM Montpellier France Univ Paris Paris France NYU Dept Comp Sci New York NY USA

This paper presents parallel solutions (developed based on two state-of-the-art algorithms iSAX and sketch) for evaluating k nearest neighbor queries on large databases of time series, compares them based on various measures of quality and time performance, and offers a tool that uses the characteristics of application data to determine which algorithm to choose for that application and how to set the parameters for that algorithm. Specifically, our experiments show that: (i) iSAX and its derivatives perform best in both time and quality when the time series can be characterized by a few low-frequency Fourier Coefficients, a regime where the iSAX pruning approach works well. (ii) iSAX performs significantly less well when high-frequency Fourier Coefficients have much of the energy of the time series. (iii) A random projection approach based on sketches by contrast is more or less independent of the frequency power spectrum. The experiments show the close relationship between pruning ratio and time for exact iSAX as well as between pruning ratio and the quality of approximate iSAX. Our toolkit analyzes typical time series of an application (i) to determine optimal segment sizes for iSAX and (ii) when to use parallel Sketches instead of iSAX. Our algorithms have been implemented using Spark, evaluated over a cluster of nodes, and have been applied to both real and synthetic data. The results apply to any databases of numerical sequences, whether or not they relate to time.

关键词： time series parallel indexing distributed querying Fourier coefficients Frequency analysis Filtering

来源：评论

学校读者我要写书评

暂无评论

distributed Dynamic Data Driven Simulations: Basic Idea and an Illustration Example

Distributed Dynamic Data Driven Simulations: Basic Idea and ...

引用

International workshop/Symposium on distributed Simulation and real-time Applications (DS-RT)

作者： Xu Xie Kai Xu College of Systems Engineering National University of Defense Technology Changsha China

The dynamic data driven simulation (DDDS) is a simulation paradigm where the simulation is continuously influenced by fresh data sampled from the real system for better analysis and prediction of the system under study. Traditional DDDS operates in the “computation away from data” mode, which would incur long response time, high network bandwidth requirement, and critical information loss. This paper proposes a novel distributed dynamic data driven simulation paradigm. In this paradigm, each participated simulation models a portion of a system, while all the participants collectively model the whole system. Additionally, each simulation assimilates data collected locally to produce local state estimations, which are aggregated somehow to generate global state estimation. The proposed simulation paradigm is supposed to have properties such as shorter response time, lower network bandwidth requirement, and less information loss. Finally, a case is studied to illustrate the effectiveness of the novel simulation paradigm.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Plaster: an Embedded FPGA-based Cluster Orchestrator for Accelerated distributed Algorithms

Plaster: an Embedded FPGA-based Cluster Orchestrator for Acc...

引用

35th IEEE International parallel and distributed Processing Symposium (IPDPS)

作者： Farinelli, Lorenzo De Vincenti, Daniele Valentino Damiani, Andrea Stornaiuolo, Luca Brondolin, Rolando Santambrogio, Marco D. Sciuto, Donatella Politecn Milan Dipartimento Elettron Informaz & Bioingn DEIB Milan Italy

ISBN: (纸本)9781665435772

The increasing use of real-time data-intensive applications and the growing interest in Heterogeneous Architectures have led to the need for increasingly complex embedded computing systems. An example of this is the research carried out by both the scientific community and companies toward embedded multi-FPGA systems for the implementation of the inference phase of Convolutional Neural Networks. In this paper, we focus on optimizing the management system of these embedded FPGA-based distributed systems. We extend the state-of-the-art FARD framework to data-intensive applications in an embedded scenario. Our orchestration and management infrastructure benefits from compiled language and is accessible to end-users by the means of Python APIs, which provides a simple way to interact with the cluster and design apps to run on the embedded nodes. The proposed prototype system consists of a PYNQ-based cluster of multiple FPGAs and has been evaluated by running an FPGA-based You Only Look Once (YOLO) image classification algorithm.

关键词： multi-FPGAs Cluster Orchestrator Embedded

来源：评论

学校读者我要写书评

暂无评论

A Multi-Level parallel Algorithm for Detection of Single Scatterers in SAR Tomography

A Multi-Level Parallel Algorithm for Detection of Single Sca...

引用

Euromicro Conference on parallel, distributed and Network-Based Processing

作者： Massimiliano Russo Mehwish Nisar Antonio Pauciullo Pasquale Imperatore Marco Lapegna Diego Romano Dept. of Mathematics and Applications University of Naples Federico II Naples Italy Institute for the Electromagnetic Sensing of the Environment (IREA) National Research Council Naples Italy Inst. for High Performance Computing and Networking National Research Council Naples Italy

ISBN: (数字)9798331524937

ISBN: (纸本)9798331524944

Synthetic Aperture Radar (SAR) tomography is an advanced technique for monitoring deformations of the Earth’s surface. However, the computational complexity of SAR tomography algorithms often restricts their application to large-scale datasets. To address this issue, we introduce a multi-level parallel implementation of a single scatterer detection algorithm specifically designed to exploit the capabilities of modern heterogeneous High-Performance Computing (HPC) systems. By efficiently distributing the computational workload at different levels across multiple processing units, our parallel approach significantly reduces processing time, facilitating the analysis of extensive SAR datasets. We assess the performance of our parallel implementation using real-world SAR data, showcasing its effectiveness in enhancing both the efficiency and scalability of SAR tomography. Our work contributes to advancing remote sensing techniques and offers valuable insights into the application of HPC for large-scale environmental monitoring.

关键词： High performance computing Scalability Graphics processing units Tomography Performance analysis parallel algorithms Synthetic aperture radar Surface treatment Remote sensing Optimization

来源：评论

学校读者我要写书评

暂无评论

Analysis and Validations of Modularized distributed TL-UPQC systems With Supervisory Remote Management System

引用

IEEE TRANSACTIONS ON SMART GRID 2021年第3期12卷 2638-2651页

作者： Abdalaal, Radwa Mohamed Ho, Carl Ngai Man Univ Manitoba Dept Elect & Comp Engn RIGA Lab Winnipeg MB R3T 5V6 Canada

parallel operation of distributed power converters is used to realize high rated power and low current ripples with low rated power devices. It is also being preferred since central compensators fail to identify power quality problems reflected on low voltage levels, especially with the new structure of microgrids. This article proposes a novel parallel operation between modularized distributed transformerless unified power quality conditioner (TL-UPQC) for low voltage distribution networks. The operation is intended to improve the grid voltage profile while considering the connected modules rated capacity. The proposed methodology is suitable for smart grid applications where multiple renewable energy sources are integrated with the network. An advanced control technique that allows monitoring, wireless communication and coordination between the connected modules is presented. The proposed control methodology allows modularity and operation flexibility. System characterization has been studied to evaluate the influences of communication delays on overall system stability. The system has been evaluated by controller hardware-in-the-loop testing methodology. The power stage is simulated in a real time digital simulator, while the control algorithm is developed in a digital signal processor. Experimental results including random behavior of a photovoltaic system are presented.

关键词： Reactive power Power quality Distribution networks Automatic voltage control Wireless communication Topology Active filters D-STATCOM hardware-in-the-loop hierarchical control parallel operation power quality UPQC smart grid voltage regulation wireless communication

来源：评论

学校读者我要写书评

暂无评论

Simple and Automatic distributed Machine Learning on Ray 21

Simple and Automatic Distributed Machine Learning on Ray

引用

27th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD)

作者： Zhang, Hao Li, Zhuohan Zheng, Lianmin Stoica, Ion Univ Calif Berkeley Berkeley CA 94720 USA

ISBN: (纸本)9781450383325

In recent years, the pace of innovations in the fields of machine learning (ML) has accelerated, researchers in SysML have created algorithms and systems that parallelize ML training over multiple devices or computational nodes. As ML models become more structurally complex, many systems have struggled to provide all-round performance on a variety of models. Particularly, ML scale-up is usually underestimated in terms of the amount of knowledge and time required to map from an appropriate distribution strategy to the model. Applying parallel training systems to complex models adds nontrivial development overheads in addition to model prototyping, and often results in lower-than-expected performance. This tutorial identifies research and practical pain points in parallel ML training, and discusses latest development of algorithms and systems on addressing these challenges in both usability and performance. In particular, this tutorial presents a new perspective of unifying seemingly different distributed ML training strategies. Based on it, introduces new techniques and system architectures to simplify and automate ML parallelization. This tutorial is built upon the authors' years' of research and industry experience, comprehensive literature survey, and several latest tutorials and papers published by the authors and peer researchers. The tutorial consists of four parts. The first part will present a landscape of distributed ML training techniques and systems, and highlight the major difficulties faced by real users when writing distributed ML code with big model or big data. The second part dives deep to explain the mainstream training strategies, guided with real use case. By developing a new and unified formulation to represent the seemingly different data- and model-parallel strategies, we describe a set of techniques and algorithms to achieve ML auto-parallelization, and compiler system architectures for autogenerating and exercising parallelization strategies

关键词： distributed deep learning distributed system parallelism

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：