检索结果-内蒙古大学图书馆

POSTER: Layrub: Layer-centric GPU memory reuse and data migration in extreme-scale deep learning systems

ACM SIGPLAN Notices 2018年第1期53卷 405-406页

作者： Liu, Bo Jiang, Wenbin Jin, Hai Shi, Xuanhua Ma, Yang Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Huazhong University of Science and Technology Wuhan430074 China

ISBN: (纸本)9781450349116

Growing accuracy and robustness of Deep Neural Networks (DNN) models are accompanied by growing model capacity (going deeper or wider). However, high memory requirements of those models make it difficult to execute the training process in one GPU. To address it, we first identify the memory usage characteristics for deep and wide convolutional networks, and demonstrate the opportunities of memory reuse on both intra-layer and inter-layer levels. We then present Layrub, a runtime data placement strategy that orchestrates the execution of training process. It achieves layer-centric reuse to reduce memory consumption for extreme-scale deep learning that cannot be run on one single GPU. © 2018 Owner/Author.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

VMgrid: A Virtual Machine supported grid middleware

VMGrid: A Virtual Machine supported grid middleware

引用

IEEE International Conference on Granular computing (GRC)

作者： Song Wu Wei Zhu Wenbin Jiang Hai Jin Cluster and Grid Computing Lab Services Computing Technology and System Lab Huazhong University of Science and Technology Wuhan China

ISBN: (纸本)9781424425129

Many grid middleware such as Globus Toolkit and GPE have started to support virtual machines. They implement the remote control of virtual machine by grid service and make use of virtual machinespsila function of quick deployment to construct the execution environment of grid application fast. In order to support virtual machine in grid environment, VMgrid (Virtual Machine grid) is presented. All services in VMgrid are implemented based on secure shell host protocol (SSH) while Globus Toolkitpsilas Virtual Workspace is on WSRF. This paper gives a clear description of VMgrid. Some comparisons with Globus Toolkitpsilas Virtual Workspace in performance are made in this paper. Performance results show that VMgrid virtual workspace atomic services are more effective than that of Globus.

关键词： Virtual machining Middleware grid computing Operating systems Software Libraries Computers

来源：评论

学校读者我要写书评

暂无评论

Adaptive multi-round scheduling strategy for divisible workloads in grid environments

Adaptive multi-round scheduling strategy for divisible workl...

引用

International Conference on Information Networking

作者： Yongcai Tao Hai Jin Song Wu Xuanhua Shi Services Computing Technology and System Lab Cluster and Grid Computing Lab Huazhong University of Science and Technology Wuhan China

ISBN: (纸本)9781424445264

Scheduling is the key to divisible workload execution. UMR (uniform multi-round) algorithm potentially performs near optimal by improving overlap of communication and computation. However, it is questioned how a static schedule works effectively in dynamic grid environment. The paper proposes an adaptive divisible workload scheduling system, which can adjust the schedule in a proactive way. An adaptive UMR-based multi-round algorithm (called AUMR) is presented and evaluated. In AUMR, if the run-time resource monitor notifies the scheduler of any resource changes, the scheduler will evaluate its impact and adjust the schedule if necessary. The experiment results show a considerable performance improvement by AUMR in dynamic grid environment.

关键词： Adaptive scheduling Processor scheduling grid computing Dynamic scheduling Monitoring Scheduling algorithm Robustness Delay Computers clustering algorithms

来源：评论

学校读者我要写书评

暂无评论

Multi-version Execution for the Dynamic Updating of Cloud Applications

Multi-version Execution for the Dynamic Updating of Cloud Ap...

引用

IEEE Annual International Computer Software and Applications Conference (COMPSAC)

作者： Feng Chen Weizhong Qiang Hai Jin Deqing Zou Duoqiang Wang Services Computing Technology and System Lab Cluster and Grid Computing Lab Huazhong University of Science and Technology Wuhan China

Softwares usually need to be updated to fix bugs or add new features. On the other hand, some critical softwares, such as cloud applications, need to provide service continuously, thus should be updated without downtime. Conventional Dynamic Software Updating (DSU) systems try to update programs while running, but they hardly consider the communication of the program to be updated with other programs, which may lead to some inconsistency problems. We handle the problem with an improved DSU system by using multi-version execution. When a new update arrives, instead of updating the application to the new version, we fork a new process of the old version and dynamically update it to the new version, then make these two versions run concurrently until the update finishes. We implement a prototype system called MUC (Multi-vesion for Updating of Cloud) on Linux. To verify our prototype, we apply MUC to cloud applications Redis and Ice cast, and evaluate the overhead of MUC at runtime.

关键词： Servers Software Synchronization Prototypes Linux Analytical models Process control

来源：评论

学校读者我要写书评

暂无评论

Evaluating Latency-Sensitive Applications' Performance Degradation in Datacenters with Restricted Power Budget

Evaluating Latency-Sensitive Applications' Performance Degra...

引用

International Conference on Parallel Processing

作者： Song Wu Chuxiong Yan Haibao Chen Hai Jin Wei Guo Zhen Wang Deqing Zou Services Computing Technology and System Lab Cluster and Grid Computing Lab Huazhong University of Science and Technology Tencent Inc

ISBN: (纸本)9781467375887

For datacenters with limited power supply, restricting the servers' power budget (i.e., the maximal power provided to servers) is an efficient approach to increase the server density (the server quantity per rack), which can effectively improve the cost-effectiveness of the datacenters. However, this approach may also affect the performance of applications in servers. Hence, the prerequisite of adopting the approach in datacenters is to precisely evaluate the application performance degradation caused by restricting the servers' power budget. Unfortunately, existing evaluation methods are inaccurate because they are either improper or coarse-grained, especially for the latency-sensitive applications widely deployed in datacenters. In this paper, we analyze the reasons why state-of-the-art methods are not appropriate for evaluating the performance degradation of latency-sensitive applications in case of power restriction, and we propose a new evaluation method which can provide a fine-grained way to precisely describe and evaluate such degradation. We verify our proposed method by a real-world application and the traces from Tencent's datecenter with 25328 servers. The experimental results show that our method is much more accurate compared with the state of the art, and we can significantly increase datacenter efficiency by saving servers' power budget while maintaining the applications' performance degradation within controllab.e and acceptable range.

关键词： Servers degradation of performance power budget evaluation method state of the art Budgets

来源：评论

学校读者我要写书评

暂无评论

A high-performance inter-domain data transferring system for virtual machines

Journal of Software

引用

Journal of Software 2010年第2期5卷 206-213页

作者： Li, Dingding Jin, Hai Shao, Yingzhe Liao, Xiaofei Han, Zongfen Chen, Kai Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Huazhong University of Science and Technology Wuhan 430074 China

Virtual machine technology can enhance server utilization and consolidation on an individual physical machine. In a growing number of contexts, many of which re-quire high-performance networking inside of the system. Unfortunately, the development of virtualization technology is inclined to application isolation as yet, leads high communication overheads between virtual machines which are resident on the same physical machine, and also limits the achievable network performance. Some efforts have been paid to bridge the network I/O performance gap between the virtualization and native environment, but the amelioration is not satisfactory. In this paper, we propose a system called IDTS (Inter-Domain Transferring system) which uses the shared memory between different domains (also called hosted or guest OSes) to expand the communication band-width and shorten the latency for network applications running in the co-resident VMs. Evaluations are also presented to prove IDTS's availab.lity and highly efficiency. We believe that IDTS is an effective solution for the inter-domain communication on Xen virtualization platform. © 2010 Academy Publisher.

关键词： Virtualization

来源：评论

学校读者我要写书评

暂无评论

N-Docker: A NVM-HDD Hybrid Docker Storage Framework to Improve Docker Performance 16th

N-Docker: A NVM-HDD Hybrid Docker Storage Framework to Impro...

引用

16th IFIP WG 10.3 International Conference on Network and Parallel computing, NPC 2019

作者： Gu, Lin Tang, Qizhi Wu, Song Jin, Hai Zhang, Yingxi Shi, Guoqiang Lin, Tingyu Rao, Jia National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Huazhong University of Science and Technology Wuhan430074 China State Key Laboratory of Intelligent Manufacturing System Technology Beijing100854 China The University of Texas at Arlington ArlingtonTX76019 United States

ISBN: (纸本)9783030307080

Docker has been widely adopted in production environment, but unfortunately deployment and cold-start of container are limited by the low speed of disk. The emerging non-volatile memory (NVM) technology, which has high speed and can store data permanently, brings a new chance to accelerate the deployment and cold-start of container. However, it is expensive to replace the whole hard disk driver (HDD) with NVM. To achieve the fastest deployment and cold-start with lowest cost, we conduct in-depth analysis on the Top-134 images in Docker Hub and obtain two main insights as: (1) the storing latency of layered image has become the bottleneck of container deployment;(2) only a few image layers are required for container cold-start. Based on these two findings, we propose a NVM-HDD hybrid docker storage framework as N-Docker. It can effectively accelerate container cold-start by detecting the bottleneck layers as well as cold-start required layers and storing them into NVM for faster container startup with limited NVM capacity. Experimental results show that N-Docker can accelerate the container deployment by 1.21X and cold-start by 2.96X. Compared to NVM-Docker, which stores all images into NVM, N-Docker achieves the same performance improvements while reducing the usage of NVM by 88.22%. © 2019, IFIP International Federation for Information Processing.

关键词： Containers

来源：评论

学校读者我要写书评

暂无评论

Accelerating GPGPU architecture simulation

Accelerating GPGPU architecture simulation

引用

2013 ACM SIGMETRICS International Conference on Measurement and Modeling of Computer systems, SIGMETRICS 2013

作者： Yu, Zhibin Eeckhout, Lieven Goswami, Nilanjan Li, Tao John, Lizy K. Jin, Hai Xu, Chengzhong Shenzhen Institute of Advanced Technology Chinese Academy of Sciences China ELIS Department Ghent University Belgium Intelligent Design of Efficient Architectures Lab. University of Florida Gainesville FL United States Department of Electrical and Computer Engineering University of Texas Austin TX United States Service Computing Technologies and System Lab./Cluster and Grid Computing Lab. HUST Wuhan China Department of Electrical and Computer Engineering Wayne State University MI United States

ISBN: (纸本)9781450319003

Recently, graphics processing units (GPUs) have opened up new opportunities for speeding up general-purpose parallel applications due to their massive computational power and up to hundreds of thousands of threads enabled by programming models such as CUDA. However, due to the serial nature of existing micro-architecture simulators, these massively parallel architectures and workloads need to be simulated sequentially. As a result, simulating GPGPU architectures with typical benchmarks and input data sets is extremely time-consuming. This paper addresses the GPGPU architecture simulation challenge by generating miniature, yet representative GPGPU kernels. We first summarize the static characteristics of an existing GPGPU kernel in a profile, and analyze its dynamic behavior using the novel concept of the divergence flow statistics graph (DFSG). We subsequently use a GPGPU kernel synthesizing framework to generate a miniature proxy of the original kernel, which can reduce simulation time significantly. The key idea is to reduce the number of simulated instructions by decreasing per-thread iteration counts of loops. Our experimental results show that our approach can accelerate GPGPU architecture simulation by a factor of 88X on average and up to 589X with an average IPC relative error of 5.6%.

关键词： Graphics processing unit

来源：评论

学校读者我要写书评

暂无评论

Vehicle Tracking with Non-overlapping Views for Multi-camera Surveillance system

Vehicle Tracking with Non-overlapping Views for Multi-camera...

引用

IEEE International Conference on High Performance computing and Communications (HPCC)

作者： Wenbin Jiang Chang Xiao Hai Jin Shuo Zhu Zhiwei Lu Services Computing Technology and System Lab Cluster and Grid Computing Huazhong University of Science and Technology Wuhan China

With the rapid development of intelligent video surveillance system for transportation, traditional single-camera-based video analysis has become insufficient. Many researches have focused on the non-overlapping multi-camera target tracking. However, the tracking precision and the computing overhead are still big obstacles. This paper proposes a novel vehicle tracking approach for non-overlapping multi-camera targets with data fusion by using minimum cost and maximum flow method. Structured information of moving targets is extracted and associated with other information such as motion time, the topology of camera network to solve targeted vehicle tracks. Besides, to improve the performance of the process, a parallelization algorithm for camera network topology partitioning is presented, which makes it possible for each sub-graph to track target independently in parallel. The experiment results show that the presented approach is able to perform target vehicle tracking analysis with high efficiency and accuracy.

关键词： Cameras Target tracking Vehicles Feature extraction Network topology Topology

来源：评论

学校读者我要写书评

暂无评论

A Rule-based Framework of Metadata Extraction from Scientific Papers

A Rule-based Framework of Metadata Extraction from Scientifi...

引用

2011 IEEE 10th International Symposium on Distributed computing and Applications to Business,Engineering(第十届电子商务、工程及科学领域的分布式计算和应用国际学术研讨会 DCABES 2011)

作者： Zhixin Guo Hai Jin Cluster and Grid Computing Lab Services Computing Technology and System Lab Huazhong University of Science and Technology Wuhan 430074 China

Most scientific documents on the web are unstructured or semi-structured, and the automatic document metadata extraction process becomes an important task. This paper describes a framework for automatic metadata extraction from scientific papers. Based on a spatial and visual knowledge principle, our system can extract title, authors and abstract from scientific papers. We utilize format information such as font size and position to guide the metadata extraction process. The experiment results show that our system achieves a high accuracy in header metadata extraction which can effectively assist the automatic index creation for digital libraries.

关键词： document metadata information extraction rulebased

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：