检索结果-内蒙古大学图书馆

***: MPI-Based Asynchronous Task Execution for Python

IEEE TRANSACTIONS ON PARALLEL AND distributed SYSTEMS 2023年第2期34卷 611-622页

作者： Rogowski, Marcin Aseeri, Samar Keyes, David Dalcin, Lisandro King Abdullah Univ Sci & Technol KAUST Extreme Comp Res Ctr ECRC Thuwal 239556900 Saudi Arabia King Abdullah Univ Sci & Technol KAUST Comp Sci Program Thuwal 239556900 Saudi Arabia

We present ***, a lightweight, asynchronous task execution framework targeting the Python programming language and using the Message Passing Interface (MPI) for interprocess communication. *** follows the interface of the *** package from the Python standard library and can be used as its drop-in replacement, while allowing applications to scale over multiple compute nodes. We discuss the design, implementation, and feature set of *** and compare its performance to other solutions on both shared and distributed memory architectures. On a shared-memory system, we show *** to consistently outperform Python's *** with speedup ratios between 1.4X and 3.7X in throughput (tasks per second) and between 1.9X and 2.9X in bandwidth. On a Cray XC40 system, we compare *** to Dask - a well-known Python parallel computing package. Although we note more varied results, we show *** to outperform Dask in most scenarios.

关键词： MPI Python parallelism master-worker parallel programming models distributed computing high performance computing task execution multiprocessing

来源：评论

学校读者我要写书评

暂无评论

Novel Industry Architectures for Connectivity Solutions in the Smart Distribution Grids

引用

IEEE ACCESS 2023年 11卷 68093-68112页

作者： Borenius, Seppo Kekolahti, Pekka Hammainen, Heikki Lehtonen, Matti Mahonen, Petri Aalto Univ Sch Elect Engn Espoo 02150 Finland

The electric energy system is undergoing a major change due to the increasing requirements of dynamic performance. In distribution grids, this evolution will necessitate expanded automation, which in turn will require enhanced connectivity solutions. Strongly evolving communications technologies and architectures, particularly mobile communications as well as cloud and edge computing, will provide new opportunities and alternatives for connectivity solutions. This paper contributes by identifying potential technical and industry architectures for the connectivity solutions required to manage distribution grids in the early 2030s. The study utilizes a senior expert panel and a Delphi survey. Industry architectures are modelled as value networks. The paper uses the Finnish distribution grids as a case example. Regarding technical architectures, the results reveal skepticism concerning those emerging 5G mobile network features that target industrial applications and about the need for extensive distributed computing in the proximity of consumers and prosumers. The most probable industry architectures are found to be those that enable the Distribution System Operators (DSOs) to maintain direct control of critical technical components, or that enable Communications Service Providers (CSPs) to handle the operations of both communications solutions and distributed computing. CSPs are seen as well positioned for this task due to their existing networking and computing infrastructure. However, this may also involve business risks for both DSOs and CSPs.

关键词： Computer architecture Business Surveys 5G mobile communication Cloud computing Information and communication technology Cloud computing security distributed computing Edge computing Mobile communication Power grids Smart grids Overlay networks communications distributed computing edge computing industry architectures mobile communications overlay networks power grids smart grids value networks

来源：评论

学校读者我要写书评

暂无评论

Configure Your Federation: Hierarchical Attention-enhanced Meta-Learning Network for Personalized Federated Learning

引用

ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY 2023年第4期14卷 1-24页

作者： Gao, Yujia Wang, Pengfei Liu, Liang Zhang, Chi Ma, Huadong Beijing Univ Posts & Telecommun 10 Xitucheng Rd Beijing 100876 Peoples R China

Federated learning, as a distributed machine learning framework, enables clients to conduct model training without transmitting their data to the server, which is used to solve the dilemma of data silos and data privacy. It can work well on clients having similar data characteristics and distribution. However, it has some limitations where the dataset of clients may be different in distribution, quantity, and concept in many application scenarios. Personalized federated learning is a new federated learning paradigm that aims to guarantee client personalized models' effectiveness when collaborating with the cloud server. Intuitively, providing further facilitated collaborations for the clients with similar data characteristics and distribution can benefit personalized model building. However, due to the invisibility of client data, it is challenging to extract client characteristics and define collaborative relationships among them from a fine-grained view. Moreover, a reasonable collaborative training approach needs to be designed for a distributed server-client framework. In this article, we design a Hierarchical Attention-enhanced Meta-learning Network (HAM) to address this issue. The main advantage of HAM is that it utilizes the meta-learning approach of taking model parameters as features and learns to learn an extra model for each client to analyze similarities according to their local dataset automatically. According to its two-layers framework, HAM can reasonably achieve a tradeoff between clients' personality and commonality and provides a hybrid model with useful information from all clients. Considering there are two networks (HAM and base network) that need to learn for each client during the federated training process, we then provide an alternative learning approach to train them in an end-to-end fashion. To further clarify the approach, we describe the personalized federated learning settings framework as FedHAM where the HAM network is distributed dep

关键词： distributed computing federated learning deep learning

来源：评论

学校读者我要写书评

暂无评论

Joint Online Optimization of Task Rescheduling and Data Redistribution

引用

JOURNAL OF INTERNET TECHNOLOGY 2023年第1期24卷 11-22页

作者： Song, Yao Xiao, Limin Wang, Liang Wei, Wei Wang, Jinquan Beihang Univ State Key Lab Software Dev Environm Beijing Peoples R China Beihang Univ Sch Comp Sci & Engn Beijing Peoples R China Xian Univ Technol Sch Comp Sci & Engn Xian Peoples R China

Wide-area distributed computing environment is the main platform for storing large amounts of data and conducting wide-area computing. Tasks and data are jointly scheduled among multiple computing platforms to improve system efficiency. However, large network latency and limited bandwidth in wide-area networks may cause a large delay in scheduling information and data migration, which brings low task execution efficiency and a long time waiting for data. Traditional works mainly focus on allocating tasks based on data locality or distributing data replications, but optimizing task allocation or data placement alone is insufficient from a global perspective. To mitigate the impact of large network latency and limited bandwidth on system performance, joint online optimization of task rescheduling and data redistribu-tion is proposed in this study. The task allocation and data placement can be adjusted collaboratively during the system running process through the task stealing and backfilling mechanism and the data replication placement mechanism. The experimental results indicate that compared with the state-of-the-art method, the proposed method improves the system throughput and computing resource utilization by 20.67% and 20.26% respectively, and can significantly re-duce the global data migration costs.

关键词： distributed computing Joint scheduling Task rescheduling Data redistribution

来源：评论

学校读者我要写书评

暂无评论

Standardization of Extended Reality (XR) over 5G and 5G-Advanced 3GPP New Radio

引用

IEEE NETWORK 2023年第4期37卷 22-28页

作者： Gapeyenko, Margarita Petrov, Vitaly Paris, Stefano Marcano, Andrea Pedersen, Klaus I. Nokia Stand Espoo Finland Nokia Stand Paris France Nokia Stand Odense Denmark Aalborg Univ Aalborg Denmark Northeastern Univ Boston MA 02115 USA

Extended Reality (XR) is one of the major innovations to be introduced in 5G/5G-Advanced communication systems. A combination of augmented reality, virtual reality, and mixed reality, supplemented by cloud gaming, revisits the way that humans interact with computers, networks, and each other. However, efficient support of XR services imposes new challenges for existing and future wireless networks. This article presents a tutorial on integrating support for the XR into the 3GPP New Radio (NR), summarizing a range of activities handled within various 3GPP Service and Systems Aspects (SA) and Radio Access Networks (RAN) groups. The article also delivers a system level simulation study evaluating the performance of different XR services over NR Release 17. The article concludes with a vision of further enhancements to better support XR in NR Release 18 and future NR releases also outlining open problems in this area.

关键词： Cloud computing Data centers Codecs Bandwidth Throughput Reliability engineering Encoding distributed computing

来源：评论

学校读者我要写书评

暂无评论

Forseti: Dynamic chunk-level reshaping for data processing on heterogeneous clusters

引用

JOURNAL OF PARALLEL AND distributed computing 2023年 171卷 14-23页

作者： Alamro, Sultan Lan, Tian Subramaniam, Suresh Qassim Univ Coll Engn Dept Elect Engn Buraydah 51452 Saudi Arabia George Washington Univ Dept Elect & Comp Engn Washington DC 20052 USA

Data-intensive computing frameworks typically split job workload into fixed-size chunks, allowing them to be processed as parallel tasks on distributed machines. Ideally, when the machines are homogeneous and have identical speed, chunks of equal size would finish processing at the same time. However, such determinism in processing time cannot be guaranteed in practice. Diverging processing times can result from various sources such as system dynamics, machine heterogeneity, and variable network conditions. Such variation, together with dynamics and uncertainty during task processing, can lead to significant performance degradation at job level, due to long tails in job completion time resulted from residual chunk workload and stragglers. In this paper, we propose Forseti, a novel processing scheme that is able to reshape data chunk size on the fly with respect to heterogeneous machines and a dynamic execution environment. Forseti mitigates residual workload and stragglers to achieve significant improvement in performance. We note that Forseti is a fully online scheme and does not require any a priori knowledge of the machine configuration nor job statistics. Instead, it infers such information and adjusts data chunk sizes at runtime, making the solution robust even in environments with high volatility. In its implementation, Forseti also exploits a virtual machine reuse feature to avoid task start-up and initialization cost associated with launching new tasks. We prototype Forseti on a real-world cluster and evaluate its performance using several realistic benchmarks. The results show that Forseti outperforms a number of baselines, including default Hadoop by up to 68% and SkewTune by up to 50% in terms of average job completion time. (c) 2022 Elsevier Inc. All rights reserved.

关键词： distributed computing Scheduling

来源：评论

学校读者我要写书评

暂无评论

Joint Task Assignment, Power Allocation and Node Grouping for Cooperative computing in NOMA-mmWave Mobile Edge computing

引用

IEEE ACCESS 2023年 11卷 93664-93678页

作者： Khazali, Azadeh Bozorgchenani, Arash Tarchi, Daniele Shayesteh, Mahrokh G. Kalbkhani, Hashem Urmia Univ Dept Elect & Comp Engn Orumiyeh *** Iran Univ Leeds Sch Comp Leeds LS2 9JT W Yorkshire England Univ Bologna Dept Elect Elect & Informat Engn Guglielmo Marcon I-40126 Bologna Italy Sharif Univ Technol Elect Engn Dept Wireless Res Lab ACRI Tehran *** Iran Urmia Univ Technol Fac Elect Engn Orumiyeh *** Iran

In this paper, we investigate the cooperation of idle computation resources of nearby mobile devices in mobile edge computing (MEC) systems, in which each mobile device jointly offloads computation tasks to a MEC node and a nearby mobile device by employing non-orthogonal multiple access (NOMA) in a millimeter-wave (mmWave) heterogeneous network. In this setup, the nearby device acts as a helper by performing local computation and offloading data simultaneously to the MEC system. We formulate an optimization problem for joint taSk assignmenT, poweR allOcation and Node Grouping (STRONG) aiming to minimize the energy consumption of devices (i.e., user and helper devices). To tackle this problem, we present a two-step solution. First, we adopt a low complexity search-based algorithm for both helper and MEC server selection. Next, considering the non-convex nature of the energy minimization problem, we develop algorithms that provide sub-optimal solutions for power allocation to the helper and MEC server, as well as the offloading task ratios between them. Numerical results are provided to validate the effectiveness of our proposed algorithms. The results not only validate the efficiency of our approach but also demonstrate the superiority of our cooperative NOMA-based MEC scenario compared to methods without cooperation and other cooperation-based scenarios.

关键词： Mobile edge computing distributed computing NOMA millimeter wave communication energy efficiency

来源：评论

学校读者我要写书评

暂无评论

Optimizing throughput of Seq2Seq model training on the IPU platform for AI-accelerated CFD simulations

引用

FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE 2023年第1期147卷 149-162页

作者： Rosciszewski, Pawel Krzywaniak, Adam Iserte, Sergio Rojek, Krzysztof Gepner, Pawel Barcelona Supercomp Ctr Barcelona Spain Univ Jaume I UJI Castellon de La Plana Spain Czestochowa Tech Univ Inst Comp & Informat Sci Czestochowa Poland Warsaw Univ Technol Fac Mech & Ind Engn Narbutta 85 PL-02524 Warsaw Poland Gdansk Univ Technol Fac Elect Telecommun & Informat Dept Comp Architecture Narutowicza 11-12 PL-80233 Gdansk Poland

Intelligence Processing Units (IPU) have proven useful for many AI applications. In this paper, we evaluate them within the emerging field of AI for simulation, where traditional numerical simulations are supported by artificial intelligence approaches. We focus specifically on a program for training machine learning models supporting a computational fluid dynamics application. We use custom TensorFlow provided by the Poplar Software Development Kit to adapt the program for the IPU-POD16 platform and investigate its ease of use and performance scalability. Training a model on data from OpenFOAM simulations allows us to get accurate simulation state predictions in test time. We describe how to optimize multi-threading runtime options and utilize the popdist library to overcome a performance bottleneck in feeding training data to the IPU on the host side. Due to communication overheads, using data parallelism to utilize two IPUs instead of one does not improve the throughput. However, once the intra-IPU costs have been paid, the hardware capabilities for inter-IPU communication allow for good scalability. Increasing the number of IPUs from two to 16 improves the throughput from 560.8 to 2805.8 samples/s. Additionally, the experimental results show that reducing the precision of input data storage from FP32 to FP16 allows to improve training throughput by 12%, while tuning selected runtime variables, by up to 6.3%. (c) 2023 Published by Elsevier B.V.

关键词： Intelligence Processing Unit Computational fluid dynamics Deep learning Time -series prediction distributed computing High-performance computing

来源：评论

学校读者我要写书评

暂无评论

The Basic Core of a Parallel Machines Scheduling Game

引用

M&SOM-MANUFACTURING & SERVICE OPERATIONS MANAGEMENT 2023年第6期25卷 2233-2248页

作者： Alon, Tzvi Anily, Shoshana Hebrew Univ Jerusalem Dept Stat IL-9190501 Jerusalem Israel Hebrew Univ Jerusalem Federmann Ctr Study Rat IL-9190501 Jerusalem Israel Tel Aviv Univ Coller Sch Management IL-6997801 Tel Aviv Israel

Problem definition: We consider the parallel machine scheduling (PMS) under job-splitting game defined by a set of manufacturers where each holds uniform parallel machines and each is committed to produce some jobs submitted to her by her clients while bearing the cost of the sum of completion times of her jobs on her machines. An efficient algorithm for this scheduling problem is well known. We consider the corresponding cooperative game, where the manufacturers are players that want to join forces. We show that collaboration is profitable. Yet, the stability of the cooperation depends on the cost allocation scheme;we focus on the core of the game. Methodology/results: We prove that the PMS game is totally balanced and its core is infinitely large, by developing a sophisticated methodology of linear complexity that finds a line segment in its symmetric core. We call this segment the basic core of the game. Managerial implications: This PMS game has the potential for various applications both in traditional industry and in distributed computing systems in the hi-tech industry. The formation of a partnership among entrepreneurs, companies, or manufacturers necessitates not only a plan for joining forces toward the achievement of the ultimate goals, but also an acceptable agreement regarding the cost allocation among the partners. Core allocations guarantee the stability of the partnership as no subset of players can gain by defecting from the grand coalition.

关键词： cooperative games core parallel machines scheduling job splitting distributed computing

来源：评论

学校读者我要写书评

暂无评论

WFLTree: A Spanning Tree Construction for Federated Learning in Wireless Networks

引用

JOURNAL OF CIRCUITS SYSTEMS AND COMPUTERS 2023年第13期32卷

作者： Li, Shuo Zheng, Yanwei Zou, Yifei Shaanxi Normal Univ Sch Math & Stat Xian 710062 Shaanxi Peoples R China Changji Univ Sch Math & Data Sci Changji 831100 Xinjiang Peoples R China Shandong Univ Sch Comp Sci & Technol Qingdao 266237 Shandong Peoples R China

Nowadays, more and more federated learning algorithms have been implemented in edge computing, to provide various customized services for mobile users, which has strongly supported the rapid development of edge intelligence. However, most of them are designed relying on the reliable device-to-device communications, which is not a realistic assumption in the wireless environment. This paper considers a realistic aggregation problem for federated learning in a single-hop wireless network, in which the parameters of machine learning models are aggregated from the learning agents to a parameter server via a wireless channel with physical interference constraint. Assuming that all the learning agents and the parameter server are within a distance gamma from each other, we show that it is possible to construct a spanning tree to connect all the learning agents to the parameter server for federated learning within O(log gamma) time steps. After the spanning tree is constructed, it only takes O(log gamma) time steps to aggregate all the training parameters from the learning agents to the parameter server. Thus, the server can update its machine learning model once according to the aggregated results. Theoretical analyses and numerical simulations are conducted to show the performance of our algorithm.

关键词： Federated learning edge intelligence wireless networks distributed computing

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：