检索结果-内蒙古大学图书馆

10th International Conference on Information Systems and Computing Technology, ISCTech 2022

作者： Liang, Xiangbin Zhao, Baokang Peng, Wei Wang, Tianshu College of Computer National University of Defense Technology Changsha China

ISBN: (纸本)9798350332933

Quick UDP Internet Connection (QUIC) is a network transmission protocol proposed by Google which has lower latency, higher flexibility and stronger security compared with TCP. Multipath QUIC (MPQUIC) further enhance the user experience and adapt the large-scale network traffic by using multiple paths. Different from Multipath TCP (MPTCP), MPQUIC scheduler can perform finer granularity stream scheduling. In heterogeneous paths, MPQUIC not only has to consider inter-stream HoL blocking, but also encounters severe intra-stream HoL blocking within individual stream;in addition, MPQUIC scheduler needs to control scheduling rate to ensure that stream data is not backed up in sub-path queue, but also the bandwidth of sub-path is fully utilized. However, existing scheduling algorithms cannot solve the problems above. Therefore, we propose multi-stream hierarchical scheduling algorithm(MS-HS), which speeds up the transmission of high-priority streams by controlling data allocation rate and giving each stream two priority levels. We evaluate the benefits of MS-HS under heterogeneous paths by comparing it with the classical scheduler of MPQUIC. Our evaluation shows that our scheduler can reduce up to 31.4% of the high-priority stream completion time, while MS-HS achieves higher throughput than the original best scheduler. © 2022 IEEE.

关键词： scheduling algorithms

来源：评论

学校读者我要写书评

暂无评论

DISTMM: Accelerating Distributed Multimodal Model Training 21

DISTMM: Accelerating Distributed Multimodal Model Training

引用

21st USENIX Symposium on Networked Systems Design and Implementation, NSDI 2024

作者： Huang, Jun Zhang, Zhen Zheng, Shuai Qin, Feng Wang, Yida The Ohio State University United States Amazon Web Services United States Boson AI Inc. United States

ISBN: (纸本)9781939133397

Multimodal model training takes multiple types of inputs to process with differently structured submodules, and aggregates outcomes from the submodules to learn the relationship among various types of inputs, e.g., correlating text to image for text-to-image generation. The differences of submodule architectures as well as their inputs lead to heterogeneity in terms of computation efficiency. Failing to account for such heterogeneity, existing distributed training systems treat all submodules as a monolithic entity and thus have sub-optimal performance. Moreover, the outcome aggregation phase introduces cross-sample dependencies with contrasting positive and negative sample pairs (i.e., contrastive loss). Such dependencies make the existing pipeline parallelism scheduling algorithms not applicable for multimodal training with contrastive loss. To address the limitations of existing solutions, we propose DISTMM. For a given multimodal model, DISTMM exploits the heterogeneity among submodules, applying different distributed parallelism strategies for each submodule, e.g., using Tensor Parallelism for a computation-intensive submodule, and Data Parallelism for a submodule with a small number of parameters. DISTMM balances the computation of parallelized submodules to reduce the computing resource idle time of waiting for the slowest submodule. DISTMM further optimizes the locality of submodules by leveraging the heterogeneous bandwidth of interconnections among accelerators. To address the limitation of existing pipeline execution schedules, we propose a new pipeline execution primitive, called batch-sync instruction, and a corresponding schedule, called DISTMM-Pipe. We build a prototype of DISTMM and evaluate it with existing solutions on models with various sizes ranging from 1.1 billion to 26 billion parameters and observe 1.32-3.27× speedup over Megatron-LM. © 2024 Proceedings of the 21st USENIX Symposium on Networked Systems Design and Implementation, NSDI 2024. A

关键词： scheduling algorithms

来源：评论

学校读者我要写书评

暂无评论

Research on real-time scheduling optimization technology of power system based on deep learning

引用

Applied Mathematics and Nonlinear Sciences 2024年第1期9卷

作者： Lu, Min Jiang, Yicheng Wang, Jin Zhu, Jianping State Grid Zhejiang Electric Power Co. Ltd Zhejiang Hangzhou310012 China NARI-TECH Nanjing Control Systems Ltd. Jiangsu Nanjing211106 China

In the context of the increasingly severe world climate form, how to rationally arrange and dispatch energy has become an urgent need. This paper proposes a deep learning-based power system scheduling model based on the concept of perfect scheduling, using GRU to learn scheduling data. A different training set is constructed to train the model according to the load characteristics at different moments, and the model is updated in real time based on the data at the current moment. The analysis of the algorithms reveals that the scheduling error rate of this model ranges from - 3% to 2%, and the average RMSE of the scheduling scheme is 2.72, placing it in close proximity to the optimal scheduling strategy. Due to a 6.5% reduction in scheduling cost compared to the average cost of the two analyzed algorithms, the average time reduction is 76.3%. The scheduling optimization model proposed in this paper exhibits excellent performance. © 2024 Min Lu et al., published by Sciendo.

关键词： scheduling algorithms

来源：评论

学校读者我要写书评

暂无评论

Graph-Based Proximal Policy Optimization Empowered Adaptive Task scheduling Leveraging Cloud-Edge Collaboration for Consumer Electronics

引用

IEEE Transactions on Consumer Electronics 2024年

作者： Li, Peisong Yi, Meng Iqbal, Muddesar Wang, Xinheng Xiao, Ziren Xi'an Jiaotong-Liverpool University School of Advanced Technology Suzhou215123 China Southeast University School of Computer Science and Engineering Key Laboratory of Computer Network and Information Integration Ministry of Education Nanjing211189 China Prince Sultan University Renewable Energy Laboratory Communications and Networks Engineering Department College of Engineering Riyadh11586 Saudi Arabia CIP Technology ltd United Kingdom

The rapid advancement of IoT, AI, and edge computing has led to a significant increase in consumer devices and computation tasks, with new electronics incorporating these technologies to enhance services like VR/AR and autonomous driving, requiring real-time processing for safety and efficiency. However, recent research has focused on optimizing IoT task scheduling and resource allocation through various methods, yet overlooks the dynamic nature of IoT environments, fails to adapt to changing device counts and movements, and often ignores task completion rate in favor of minimizing latency and energy cost. In this context, an adaptive task scheduling and resource allocation strategy is proposed for Edge IoT systems, based on the designed Graph-based Proximal Policy Optimization (GPPO) algorithm. Firstly, the GPPO algorithm enhances PPO for adaptive task scheduling in the fluctuating MEC scenarios, adjusting for the varying number of nearby edge servers. Secondly, it accounts for consumer mobility by opting for local task execution if the consumer risks moving outside the edge server's range, ensuring result reception. Thirdly, it prioritizes task completion rate to increase the number of tasks finished within their acceptable duration. Experimental results demonstrated that the proposed method outperforms traditional methods. © 1975-2011 IEEE.

关键词： scheduling algorithms

来源：评论

学校读者我要写书评

暂无评论

Meteorological data layout and task scheduling in a multi-cloud environment

引用

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE 2023年第PartB期126卷

作者： Hao, Yongsheng Cao, Jie Wang, Qi Ma, Tinghuai Wang, Qin Zhang, Xin Nanjing Univ Informat Sci & Technol Sch Math & Stat Nanjing 210044 Peoples R China Nanjing Univ Informat Sci & Technol Network Ctr Nanjing 210044 Peoples R China Xuzhou Univ Technol Xuzhou 221018 Peoples R China Dalian Maritime Univ Informat Sci & Technol Coll Dalian 116026 Peoples R China Nanjing Univ Informat Sci & Technol Sch Comp Sci Nanjing 210044 Peoples R China Wenzhou Med Univ Sch Ophthalmol & Optometry Sch Biomed Engn Wenzhou 325035 Peoples R China

The meteorological cloud mainly provides computing ability and meteorological datasets for meteorological model tasks. If the location of the required dataset and the execution location of the task are different, this will consume a large amount of time and bandwidth to transfer the data for the task. Meteorological data layout allocates meteorological datasets to various clouds. Because meteorological datasets are required by multiple meteorological model tasks and multiple times, the data layout is very important in the meteorological clouds. This paper focuses on how to layout out the meteorological datasets based on the association (internal meteorological datasets and between meteorological model tasks) and schedule resources for meteorological model tasks in the meteorological cloud. First, to find the association in the meteorological datasets and meteorological models, we use Apriori algorithm to mine frequent itemsets between datasets used by different meteorological models, and then we use the result to help layout meteorological data. After that, we present a heuristic algorithm for scheduling meteorological tasks. Finally, simulation comparison shows that the meteorological data layout method has a lowest value in the number of involved clouds for every task, the average size of transmitted datasets from other clouds, and the average time of transmitted datasets between clouds. We also prove that the scheduling method based on the data layout increases the number of completed tasks before their deadlines and reduces the average execution time.

关键词： scheduling algorithms Resource management Layout Data mining

来源：评论

学校读者我要写书评

暂无评论

Adaptive multi-tier intelligent data manager for Exascale 23

Adaptive multi-tier intelligent data manager for Exascale

引用

20th ACM International Conference on Computing Frontiers (CF)

作者： Carretero, Jesus Garcia-Blas, Javier Aldinucci, Marco Besnard, Jean Baptiste Acquaviva, Jean-Thomas Brinkmann, Andre Vef, Marc-Andre Jeannot, Emmanuel Miranda, Alberto Nou, Ramon Riedel, Morris Torquati, Massimo Wolf, Felix Univ Carlos III Madrid Madrid Spain Univ Torino Turin Italy CINI HPC KTT Lab Turin Italy ParaTools SAS Bruyeres Le Chatel France DataDirect Networks Storage Paris France Johannes Gutenberg Univ Mainz Mainz Germany INRIA Bordeaux France Barcelona Supercomp Ctr Barcelona Spain Juelich Supercomp Ctr Julich Germany Univ Pisa Pisa Italy CINI HPC KTT Lab Pisa Italy Tech Univ Darmstadt Darmstadt Germany

ISBN: (纸本)9798400701405

The main objective of the ADMIRE project1 is the creation of an active I/O stack that dynamically adjusts computation and storage requirements through intelligent global coordination, the elasticity of computation and I/O, and the scheduling of storage resources along all levels of the storage hierarchy, while offering quality-of-service (QoS), energy efficiency, and resilience for accessing extremely large data sets in very heterogeneous computing and storage environments. We have developed a framework prototype that is able to dynamically adjust computation and storage requirements through intelligent global coordination, separated control, and data paths, the malleability of computation and I/O, the scheduling of storage resources along all levels of the storage hierarchy, and scalable monitoring techniques. The leading idea in ADMIRE is to co-design applications with ad-hoc storage systems that can be deployed with the application and adapt their computing and I/O behaviour on runtime, using malleability techniques, to increase the performance of applications and the throughput of the applications.

关键词： High-Performance I/O Malleability Monitoring scheduling algorithms Predictive models

来源：评论

学校读者我要写书评

暂无评论

Service-aware and Interference Avoiding Multi-node Parallel Resource scheduling Algorithm for Smart Grid Networks 22

Service-aware and Interference Avoiding Multi-node Parallel ...

引用

22nd IEEE International Conference on Communication Technology, ICCT 2022

作者： Hu, Chengbo Zhu, Xueqiong Lu, Yongling Wang, Zhen Yang, Jinggang Sun, Rong Xue, Hai State Grid Jiangsu Electric Power Company Ltd. Research Institute Electric Power System Artificial Intelligence State Grid Corporation Joint Laboratory Nanjing211103 China

ISBN: (纸本)9781665470674

The efficiency of wireless resource scheduling is a technical challenge in the fields of smart grid networks when time slot, frequency, and channel resources are limited and transmission service requests appearing. In this paper, a novel multi-node cooperative parallel resource scheduling scheme is proposed to solve this problem. Based on the perception of global network transmission services and the topology of scheduling services, multi-node cooperative parallel resource scheduling is performed. The computer simulation verifies that the proposed algorithm can maximize the number of parallel business data streams, satisfy the QoS quality of power services, and save network resources. © 2022 IEEE.

关键词： scheduling algorithms

来源：评论

学校读者我要写书评

暂无评论

SOWalker: An I/O-Optimized Out-of-Core Graph Processing System for Second-Order RandomWalks

SOWalker: An I/O-Optimized Out-of-Core Graph Processing Syst...

引用

USENIX Annual Technical Conference (USENIX ATC)

作者： Wu, Yutong Shi, Zhan Huang, Shicai Tian, Zhipeng Zuo, Pengwei Fang, Peng Wang, Fang Feng, Dan Huazhong Univ Sci & Technol Wuhan Natl Lab Optoelect Wuhan Peoples R China

ISBN: (纸本)9781939133359

Random walks serve as a powerful tool for extracting information that exists in a wide variety of real-world scenarios. Different from the traditional first-order random walk, the second-order random walk considers recent walk history in selecting the next stop, which facilitates to model higher-order structures in real-world data. To meet the scalability of random walks, researchers have developed many out-of-core graph processing systems based on a single machine. However, the main focus of out-of-core graph processing systems is to support first-order random walks, which no longer perform well for second-order random walks. In this paper, we propose an I/O-optimized out-of-core graph processing system for second-order random walks, called SOWalker. First, we propose a walk matrix to avoid loading non-updatable walks and eliminate useless walk I/Os. Second, we develop a benefit-aware I/O model to load multiple blocks with the maximum accumulated updatable walks, so as to improve the I/O utilization. Finally, we adopt a block set-oriented walk updating scheme, which allows each walk to move as many steps as possible in the loaded block set, thus significantly boosting the walk updating rate. Compared with two state-of-the-art random walk systems, GraphWalker and GraSorw, SOWalker yields significant performance speedups (up to 10.2x).

关键词： scheduling algorithms

来源：评论

学校读者我要写书评

暂无评论

Average Utility Function Maximization-Based Multi-UAV Cooperative Perception and Trajectory Optimization 34

Average Utility Function Maximization-Based Multi-UAV Cooper...

引用

IEEE 34th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC)

作者： Pu, Renyan Chai, Rong Sun, Ruijin Li, Lifan Chongqing Univ Posts & Telecommun Sch Commun & Informat Engn Chongqing 400065 Peoples R China Xidian Univ State Key Lab Integrated Serv Networks Xian Peoples R China China Mobile IoT Co Ltd Hong Kong Peoples R China

ISBN: (纸本)9781665464833

In recent years, unmanned aerial vehicles (UAVs) have been experienced rapid development and have been widely used in many fields. Equipped with both communication modules and sensing modules, UAVs are capable of achieving integrated communication and target detection, thus greatly improving spectrum efficiency and system performance. In this paper, we consider a scenario where multiple UAVs collaborate to detect targets and transmit the collected data to a central UAV. Addressing the problem of communication and perception scheduling, we first analyze the detection performance and communication performance, and then formulate the joint communication and perception scheduling problem into an average utility function maximization problem. To solve the formulated problem, we first consider the dynamic characteristics of the environment, and model the problem as a Markov decision process. Regarding the UAVs as multiple agents, we propose a multi-agent double deep Q-network based joint communication and perception scheduling algorithm. Simulation results demonstrate the effectiveness and superiority of the algorithm.

关键词： scheduling algorithms

来源：评论

学校读者我要写书评

暂无评论

A Scalable Heuristic Time-Sensitive Traffic scheduling Algorithm for In-Vehicle Network 5

A Scalable Heuristic Time-Sensitive Traffic Scheduling Algor...

引用

5th IEEE International Conference on Hot Information-Centric Networking, HotICN 2022

作者： Huang, Zhiqiu Zhu, Hailong Zhang, Huayu Huang, Tao Beijing University of Posts and Telecommunication BeijingBJ 100876 China Purple Mountain Laboratories NanJingNJ 211111 China

ISBN: (纸本)9798350335200

The rapid development of autonomous driving has put forward higher requirements for the bandwidth and certainty of data transmission in the vehicle. As an important standard in Time-Sensitive Network (TSN), IEEE 802.1Qbv provides the possibility to achieve the mixed transmission of different traffic classes in In- Vehicle Network (IVN). However, the mainstream method to solve the Gate Control List (GCL) by Satisfiability Modulo Theories (SMT) is too complicated, which is not suitable for large-scale traffic scenarios. And the proposed heuristic algorithms are not scalable enough to meet all the constraints of IVN. Therefore, an advanced Earliest Window heuristic scheduling algorithm is proposed in this paper, which can satisfy all the constraints of scheduling problems in multi-period, multi-queue and multi-hop scenarios with controllable complexity. The Frame Replication and Elimination for Reliability (FRER) mechanism of IEEE 802.1CB is also introduced in the algorithm to ensure the reliable transmission of critical data flows. © 2022 IEEE.

关键词： scheduling algorithms

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：