检索结果-内蒙古大学图书馆

Integrated and Fungible scheduling of Deep Learning Workloads Using Multi-Agent Reinforcement Learning

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS 2025年第3期36卷 391-406页

作者： Li, Jialun Xiao, Danyang Yang, Diying Mo, Xuan Wu, Weigang Guangdong Polytech Normal Univ Sch Comp Sci Guangzhou 510640 Peoples R China Sun Yat Sen Univ Sch Comp Sci & Engn Guangzhou 510275 Peoples R China

GPU clusters have been widely used to co-locate various deep learning (DL) workloads in a multi-tenant way. Although such resource sharing can significantly reduce training cost, resource contention and interference among co-located workloads make task scheduling very complex and challenging. To simplify the scheduling problem, existing algorithms usually divide the procedure of scheduling into two sub-tasks, i.e., task placement and resource allocation, and allocate resources according to pre-defined and fixed resource demands. However, such a paradigm significantly constrains the selection of potential scheduling solutions. In this article, we present MAIFS, a novel multi-agent reinforcement learning based scheduling algorithm that handles task placement and resource allocation integratedly, and allows fungible resource allocation based on resource sensitivity of DL workloads. The core of MAIFS lies in two mechanisms. The multi-agent attention mechanism is designed to learn and share inter-related resource state features observed from different agents, which enables agents to explore fungible resource allocation solutions. The dynamic coordination graph mechanism is designed for coordinating interactive task placement decisions of agents during integrated scheduling, so as to mitigate potential task conflicts. Simulated experiments using two large scale production DL workload traces and physical deployment experiments based on a Kubernetes based GPU cluster show that MAIFS can outperform state-of-the-art scheduling algorithms by up to 44% in terms of makespan and 46% in terms of job completion time (JCT).

关键词： Resource management scheduling algorithms Graphics processing units Dynamic scheduling Sensitivity Random access memory Job shop scheduling Heuristic algorithms Training Servers GPU cluster task scheduling deep learning workload MARL MLaaS

来源：评论

学校读者我要写书评

暂无评论

Cost and Makespan-Aware Task scheduling With Deep Reinforcement Learning in Multicloud Environments

引用

IEEE TRANSACTIONS ON COMPUTATIONAL SOCIAL SYSTEMS 2025年

作者： Tang, Xuhao Liu, Fagui Wang, Bin Zhang, Mengjiao Jiang, Jun Tang, Quan Wu, Qingbo Chen, C. L. Philip South China Univ Technol Sch Comp Sci & Engn Guangzhou 510006 Peoples R China Pengcheng Lab Dept New Networks Shenzhen 518066 Peoples R China Natl Univ Def Technol Coll Comp Changsha 410073 Peoples R China

The multicloud environments (MCE) represent a novel paradigm encompassing multiple infrastructure as a service (IaaS) providers, enabling users to tailor and optimize cloud services according to their specific requirements. This approach effectively addresses the limitations of a single cloud environment (SCE) regarding technical constraints, geographical coverage deficiencies, and cost-effectiveness concerns while catering to the increasingly diverse and expanding user demands. In MCE, users must employ appropriate strategies to efficiently allocate diverse tasks across multiple cloud service providers (CSPs) by leveraging the best available resources. Traditional scheduling algorithms are inadequate for addressing the complexities of such MCE. This study introduces a framework for the task scheduling procedure in MCE, treating independent task scheduling as a Markov decision process (MDP). We propose a novel agent environment framework that is designed based on the distinctive characteristics of MCE and enables independent task scheduling. Furthermore, we propose a task scheduling algorithm for MCE based on deep reinforcement learning (DRL) to optimize cost and makespan according to diverse user requirements. The simulation experiments are conducted using both simulated datasets and real-world datasets, demonstrating that our proposed algorithm surpasses the other five algorithms in terms of cost minimization and makespan optimization.

关键词： Costs Cloud computing scheduling algorithms Optimization Heuristic algorithms Decision making Dynamic scheduling Complexity theory Deep reinforcement learning Resource management Cost deep reinforcement learning (DRL) makespan multicloud task scheduling

来源：评论

学校读者我要写书评

暂无评论

Critical-Task-Driven Multi-Objective Evolutionary Algorithm for scheduling Large-Scale Workflows in Cloud Computing

引用

IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTATIONAL INTELLIGENCE 2025年

作者： Liu, Xiaolu Yao, Feng Xing, Lining Chen, Huangke Zhao, Wei Zheng, Long Natl Univ Def Technol Coll Syst Engn Changsha 410073 Peoples R China Xidian Univ Key Lab Collaborat Intelligence Syst Minist Educ Xian 710071 Peoples R China

Simultaneously optimizing energy consumption and makespan is essential for deploying workflows on cloud platforms. Meanwhile, it poses various challenges, such as multiple types of decision variables and large-scale decision variables. Until now, some studies have adopted multi-objective evolutionary optimization techniques to handle these challenges. But, most relevant studies regard the optimization problem as a black box and do not fully explore the domain knowledge to strengthen the search efficiency. To further answer the above challenges, this study tailors a critical-task-driven evolutionary algorithm, namely CTMOEA, to optimize both energy consumption and makespan by evolving two types of decision variables: 1) mappings from tasks to resources;and 2) task runtime. In particular, an adaptive incentive mechanism is designed to distinguish decision variables with higher criticality and allocate more computing resources to evolve them, thus handling large-scale decision variables in a targeted manner. Moreover, a critical task remapping mechanism is designed to adaptively remap critical tasks onto the same resources as their successor tasks to eliminate data transfer time between them, thus pursuing a simultaneous reduction in energy consumption and makespan. At last, based on 16 real-world workflow traces, we validate the superior overall performance of the proposed CTMOEA by significantly outperforming five baselines on 13 test instances. Also, we conduct an ablation analysis to demonstrate the contributions of the three problem-specific mechanisms to overall performance.

关键词： Cloud computing Energy consumption Optimization Costs scheduling algorithms Data centers Runtime Data transfer Data models Computational modeling workflow scheduling evolutionary computation multi-objective optimization large-scale energy efficient

来源：评论

学校读者我要写书评

暂无评论

Packet scheduling in Multi-Flow Carrier Aggregation With QoS Provisioning: Cross-Layer Design and Performance Analysis

引用

IEEE TRANSACTIONS ON MOBILE COMPUTING 2025年第2期24卷 507-524页

作者： Alorainy, Abdulaziz King Abdulaziz City Sci & Technol KACST Future Generat Connect & Wireless Sensors Res Ins Riyadh 12354 Saudi Arabia

Multi-flow carrier aggregation (CA) is a key solution for improving data rates in heterogeneous networks (HetNets). Efficient data splitting is important to enhance the performance of multi-flow CA. In this paper, packet scheduling in multi-flow CA is investigated via a cross-layer design approach with comprehensive quality of service (QoS) provisioning. In particular, an approximation of the average queuing delay (AQD) in terms of the average buffers' length and the average packet arrivals is presented. Then, using the presented AQD approximation, various packet scheduling schemes are developed in this paper. First, a low-complexity buffer and link aware heuristic packet scheduling scheme is proposed. Next, in order to reduce backhaul signalling overhead, a hybrid packet scheduling scheme that combines heuristic packet scheduling and random packet scheduling is considered. Then, an optimal packet scheduling scheme, albeit of high computational complexity, is proposed as a benchmark for other packet scheduling schemes. Also, the performances of the proposed packet scheduling schemes are analyzed and various data link layer performance parameters such as AQD, packet loss probability (PLP) and statistical queuing delay bound (SDB) are taken into consideration. Moreover, the complexity and the signalling overhead of the proposed packet scheduling schemes are investigated. Presented numerical results demonstrate that the heuristic packet scheduling scheme can achieve a performance that is similar to the performance of the optimal packet scheduling scheme while overcoming complexity and feasibility issues of the optimal packet scheduling scheme. Also, the hybrid packet scheduling scheme reduces signalling overhead while reaping some of the benefits of using buffer and link aware packet scheduling.

关键词： scheduling algorithms Quality of service Delays Complexity theory Macrocell networks Mobile computing Cross layer design Dual connectivity multi-flow carrier aggregation packet scheduling queuing delay

来源：评论

学校读者我要写书评

暂无评论

scheduling algorithms for Tree-based Convergecast Networks with a Random Number of Messages 14

Scheduling Algorithms for Tree-based Convergecast Networks w...

引用

XIV Int Symposium Problems Redundancy Inform Control Syst

作者： Bakin, Evgeny Evseev, Grigory St Petersburg State Univ Aerosp Instrumentat St Petersburg Russia

ISBN: (纸本)9781479940509

In the paper we consider tree-based L-level convergecast networks, in which leaf nodes correspond to data sources (DS), a root node - to a base station (BS), and other nodes - to intermediate repeaters (IR). At a startup phase each DS generates a random number of messages for transmission to a BS. For such a model a lower bound for an average duration of collision-free convergecast is given. Also we propose a collision-free messages transmission scheduling algorithms for which the lower bound is achieved at L -> infinity.

关键词： trees (mathematics) wireless sensor networks BS DS IR base station collision-free messages transmission scheduling algorithms data sources intermediate repeaters tree-based convergecast networks Control systems Decision support systems Redundancy Repeaters scheduling algorithms Sensors Wireless sensor networks

来源：评论

学校读者我要写书评

暂无评论

Response Time Analysis and Optimal Priority Assignment for Global Non-Preemptive Fixed-Priority Rigid Gang scheduling

引用

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS 2025年第3期36卷 455-470页

作者： Sun, Binqi Kloda, Tomasz Chen, Jiyang Lu, Cen Caccamo, Marco Tech Univ Munich D-85748 Munich Germany Univ Toulouse AAS CNRS INSA F-31000 Toulouse France

Non-preemptive rigid gang scheduling combines the efficiency of parallel execution with the reduced overhead of non-preemptive scheduling. This approach is particularly advantageous for parallel hardware accelerators, such as Google's Edge Tensor Processing Unit (TPU), which is widely used for deep neural network (DNN) inference on embedded systems. This paper studies sporadic global non-preemptive fixed-priority (NP-FP) rigid gang scheduling, which is well-suited for DNN applications in Edge TPU pipelines. Each gang task spawns a fixed number of threads that must execute concurrently across distinct processing units. We introduce the first carry-in limitation technique specifically designed for gang task response time analysis, addressing the unique challenges posed by intra-task parallelism. This technique is formulated as a generalized knapsack problem, and we develop both a linear programming relaxation and a dynamic programming approach to solve it under different time complexities. Additionally, we propose the first optimal priority assignment policy for NP-FP gang schedulability tests. Our proposed schedulability analysis and optimal priority assignment policy are evaluated through extensive experiments, including both synthetic task sets and a case study using DNN benchmarks on commercial off-the-shelf Edge TPU accelerators. The results demonstrate that the proposed approaches effectively enhance the state-of-the-art global NP-FP gang schedulability tests, achieving improvements of up to 57.9% for synthetic task sets and 76.7% for Edge TPU benchmarks. Furthermore, we conduct an ablations study to examine the impact of different algorithmic components in the proposed technique, providing valuable insights for future research.

关键词： Optimal scheduling Artificial neural networks scheduling Time factors Pipeline processing scheduling algorithms Real-time systems Interference Benchmark testing Sun Gang scheduling response time analysis optimal priority assignment Edge TPU real-time system

来源：评论

学校读者我要写书评

暂无评论

Earth Observation Satellite Downlink scheduling With Satellite-Ground Optical Communication Links

引用

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS 2025年第2期61卷 2281-2294页

作者： Li, Hai Li, Yongjun Liu, Yuanhao Deng, Boyu Li, Yu Li, Xin Zhao, Shanghong Air Force Engn Univ Coll Informat & Nav Xian 710077 Peoples R China Chinese Acad Mil Sci Beijing 100071 Peoples R China Natl Univ Def Technol Coll Informat & Commun Wuhan 430000 Peoples R China

Consistent growth in the number of high-resolution Earth observation satellites (EOSs) poses a significant challenge to the downlink scheduling problem for observation data. Satellite-to-ground optical communication is considered promising for downloading massive observation data due to its large communication capacity. To improve the downlink transmission capacity of the EOS observation system, this article focuses on the EOS downlink scheduling problem with satellite-ground optical communication links (EDSPO). First, we establish a mathematical model considering many practical constraints for EDSPO to maximize the profit of downloaded observation data. To tackle this complex scheduling problem, a downlink scheduling algorithm combining with the adaptive large neighborhood search algorithm and the tabu search algorithm (ALNS_TS) is then proposed. In ALNS_TS, a heuristic solution initialization is developed, five removal operators and five repair operators are designed based on the heuristic characteristics of EDSPO, and a task insert algorithm is proposed for repair operators. Moreover, ALNS_TS introduces the tabu value for each downlink transmission task to avoid repeated searches in the neighborhood of the same solution. Finally, extensive experiments are carried out to verify the effectiveness of the ALNS_TS and the superiority of the satellite-ground optical communication link. Simulation results show that the ALNS_TS performs better in profits and the number of completed tasks compared to the existing downlink scheduling algorithms. Besides, the satellite-ground optical communication link has been proven effective in improving the downlink transmission capacity of the EOS observation system.

关键词： Downlink Earth Observing System Optical fiber communication Microwave antennas Satellites Transmitting antennas scheduling algorithms Adaptive large neighborhood search (ALNS) Earth observation satellites (EOSs) downlink scheduling problem satellite-ground optical communication link tabu search (TS)

来源：评论

学校读者我要写书评

暂无评论

Towards Secure and Efficient Data scheduling for Vehicular Social Networks

引用

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY 2025年第1期74卷 1551-1569页

作者： Xia, Youhua Zhang, Tiehua Jin, Jiong He, Ying Yu, F. Richard Guangdong Lab Artificial Intelligence & Digital Ec Shenzhen 518107 Peoples R China Tongji Univ Coll Elect & Informat Engn Shanghai 200070 Peoples R China Swinburne Univ Technol Sch Sci Comp & Engn Technol Melbourne Vic 3122 Australia Shenzhen Univ Coll Comp Sci & Software Engn Shenzhen 518060 Peoples R China Carleton Univ Ottawa ON K1S 5B6 Canada

Efficient data transmission scheduling within vehicular environments poses a significant challenge due to the high mobility of such networks. Contemporary research predominantly centers on crafting cooperative scheduling algorithms tailored for vehicular networks. Notwithstanding, the intricacies of orchestrating scheduling in vehicular social networks both effectively and efficiently remain a formidable task. This paper introduces an innovative learning-based algorithm for scheduling data transmission that prioritizes efficiency and security within vehicular social networks. The algorithm first uses a specifically constructed neural network to enhance data processing capabilities. After this, it incorporates a Q-learning paradigm during the data transmission phase to optimize the information exchange, the privacy of which is safeguarded by differential privacy through the communication process. Comparative experiments demonstrate the superior performance of the proposed Q-learning enhanced scheduling algorithm relative to existing state-of-the-art scheduling algorithms in the context of vehicular social networks.

关键词： Social networking (online) scheduling algorithms Optimal scheduling Data communication Collaboration Computer architecture Real-time systems Data scheduling multi-layer perceptron privacy-preserving Q-learning vehicular social networks

来源：评论

学校读者我要写书评

暂无评论

Optimization-based resource scheduling techniques in cloud computing environment: A review of scientific workflows and future directions

引用

COMPUTERS & ELECTRICAL ENGINEERING 2025年 123卷

作者： Kathole, Atul B. Vhatkar, Kapil Lonare, Savita Kshirsagar, Aniruddha P. Dr DY Patil Inst Technol Dept Comp Engn Pune 411018 Maharashtra India Karmaveer Bhaurao Patil Coll Engn Dept Comp Sci & Engn Satara 415001 Maharashtra India

Due to the rapid growth of heterogeneous nature and demand requests over cloud resources, resource scheduling has become the most prominent issue in cloud computing. Thus, the cloud has offered elasticity, uncertainty, and dynamism depending on the services to the users on the internet. Here, the inefficient scheduling approaches have faced difficulty over resources that are being underutilized and over-utilized and it led to the degradation of service performance. The major intention of scheduling is to distribute the tasks between the cloud, where the scheduling algorithm has neglected the issues of imbalance. The scheduling algorithm has optimized the key performance indicators parameters such as resource utilization, cost, energy consumption, availability, reliability, makespan time, response time, and so on. Concerning this, many conventional scheduling algorithms are proposed. This research has depicted a broad literature analysis of resource scheduling in a cloud environment to help researchers detect adequate scheduling algorithms. The current status of the cloud resource allocation strategies is split into several categories. The methodological analysis of resource scheduling is presented in this research work. The optimization and meta-heuristic algorithms used in this work are also discussed. The parameters utilized for estimating the performance of the system are also analyzed. In addition, the benefits and challenges of the resource scheduling algorithms are also discussed. Finally, the research gap analysis is given, which is helpful for researchers in the field of cloud computing to make further improvements in scheduling techniques.

关键词： Resource scheduling Cloud computing environment scheduling algorithms Parameters for optimization Performance measures Research gaps and future challenges

来源：评论

学校读者我要写书评

暂无评论

Qin: Unified Hierarchical Cluster-Node scheduling for Heterogeneous Datacenters

IEEE TRANSACTIONS ON SUSTAINABLE COMPUTING

引用

IEEE TRANSACTIONS ON SUSTAINABLE COMPUTING 2025年第1期10卷 39-56页

作者： Guan, Wenkai Ababei, Cristinel Univ Minnesota Div Sci & Math Morris MN 56267 USA Marquette Univ Dept Elect & Comp Engn Milwaukee WI 53233 USA

Energy efficiency is among the most important challenges for computing. There has been an increasing gap between the rate at which the performance of processors has been improving and the lower rate of improvement in energy efficiency. This paper answers the question of how to reduce energy usage in heterogeneous datacenters. It proposes a unified hierarchical scheduling using a D-Choices technique, which considers interference and heterogeneity. Heterogeneity comes from servers' continuous upgrades and the integrated high-performance "big" and energy-efficient "little" cores. This results in datacenters becoming more heterogeneous and traditional job scheduling algorithms become suboptimal. To this end, we present a two-level hierarchical scheduler for datacenters that exploits increased server heterogeneity. It combines in a unified approach cluster and node level scheduling algorithms, and it can consider specific optimization objectives including job completion time, energy usage, and energy-delay-product (EDP). Its novelty lies in the unified approach and in modeling interference and heterogeneity. Experiments on a research cluster found that the proposed approach outperforms state-of-the-art schedulers by around 10% in job completion time, 39% in energy usage, and 42% in EDP. This paper demonstrated a unified approach as a promising direction in optimizing energy and performance for heterogeneous datacenters.

关键词： Servers Optimization Collaboration Interference scheduling Motion pictures scheduling algorithms Collaborative filtering datacenters heterogeneity Kalman filtering scheduling

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：