检索结果-内蒙古大学图书馆

21st IEEE International Conference on High Performance Computing and Communications, 17th IEEE International Conference on Smart City and 5th IEEE International Conference on Data Science and Systems, HPCC/SmartCity/DSS 2019

作者： Zhou, Xin Li, Rongchun National Key Laboratory for Parallel and Distribution Processing National University of Defense Technology Changsha China

ISBN: (纸本)9781728120584

Turbo product codes (TPC) are a class of forward error correction (FEC) codes. They have good bit error rate (BER) performance at high code rate. It is relatively simple to implement the encoder of TPC and the decoding complexity of their decoder is reasonable. Therefore, TPC are widely used in various places such as satellite communication systems and data storage systems. In this paper, a parallel TPC decoder based on GPUs is proposed. All rows or columns of the two-dimensional product code matrix are decoded simultaneously in this proposed decoder. A parallel elementary decoder is designed to simplify the decoding process of TPC which are constructed by extended Hamming codes. The calculations of test patterns and valid codewords are parallelled to reduce decoding latency. In order to further improve the decoding throughput, we present the multi-channels TPC decoder. The performance of the parallel decoder is measured on different GPUs. The experiment result shows that the decoding latency is reduced significantly compared with the TPC decoder based on a CPU. In addition, throughputs of proposed GPU decoder achieve 30Mbps on Nvidia RTX 2080 Ti and 38 Mbps on Nvidia Titan V, which are 44 times and 54 times of the CPU-based decoder. © 2019 IEEE.

关键词： Graphics processing unit

来源：评论

学校读者我要写书评

暂无评论

A Novel network request scheduler for a large scale storage system

A Novel network request scheduler for a large scale storage ...

引用

作者： Qian, Yingjin Barton, Eric Wang, Tom Puntambekar, Nirant Dilger, Andreas National Laboratory for Parallel and Distributed Processing Changsha China Sun Microsystems Inc Lustre Group PH United States

This paper presents a novel Network Request Scheduler (NRS) for a large-scale, LustreTM storage system. It proposes a quantum-based, Object Based Round Robin (OBRR) NRS algorithm that reorders the execution of I/O requests per data object, presenting a workload to backend storage that can be optimized more easily. According to the drawback of static deadlines in large-scale workloads, it proposes a novel two-level deadline setting strategy that not only avoids starvation, but also guarantees that urgent I/O requests are serviced in a specified time period. Via a series of simulation experiments using a Lustre simulator, it demonstrates that I/O performance increases as much as 40% when using the OBRR NRS algorithm, and the two-level deadline setting strategy can avoid starvation and ensure that urgent I/O requests are serviced in the required time. © 2009 Springer-Verlag.

关键词： Scheduling

来源：评论

学校读者我要写书评

暂无评论

A pipelining strategy for accelerating convolution neural networks on ARM CPUs

A pipelining strategy for accelerating convolution neural ne...

引用

作者： Zhou, Xin Dou, Yong Li, Rongchun Zhang, Peng Liu, Yuntao National Key Laboratory for Parallel and Distribution Processing National University of Defense Technology Changsha China

Convolution is a primary operation in convolution neural networks. The speed of inference is mainly decided by the speed of the convolutional layer. Improving the performance of embedded processors makes it possible to process the inference on embedded devices. In this article, a pipelining strategy of single instruction and multiple data (SIMD) instructions is proposed to finely optimize the process of the 3 × 3 convolution on ARM-based CPUs. We implement the SIMD group to improve the efficiency of the SIMD pipeline. A tiling method is exploited to increase data reuse during the process. An evaluation model is proposed to guide the design of the tiling method and register allocation. The speed of our implementation is 5.18 times of the GNU compiler collection compiled unoptimized version on RK3288. The effect of our optimizing method is measured by a performance profiling tool, the performance information suggests that the pipelining strategy has a significant effect for both normal and depthwise separable convolution. By implementing multithread processing, the speedup achieves 18.3 compared with the single thread unoptimized version. © 2021 John Wiley & Sons Ltd

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

Efficiently indexing sparse wide tablesin cloud computing

引用

Information Technology Journal 2013年第20期12卷 5415-5423页

作者： Huang, Bin Peng, Yuxing Department of Computer Huaihua University 418008 Huaihua China School of Computer Science Wuhan University 430072 Wuhan China National Laboratory of Parallel and Distributed Processing National University of Defense Technology 410073 Changsha China

For its perception of unlimited resources and infinite scalability, Cloud Computing has emerged as a pervasive paradigm for hosting? data-centric applications in large computing infrastructures. The data produced by these applications are essentially sparse and wide and may change schema frequently, traditional relational data model is inappropriate for their data management. A new data model, called Sparse Wide Table, was introduced for this task. Unfortunately, we have to face many challenges in building the secondary index for Sparse Wide Table in cloud, as the distributed and column-oriented storage which eliminates a number of NULLs. In this study, we present a three-level index scheme for efficient data processing in the Cloud. Our approach can be summarized as follows. First, we build an index for each column by which the records can be rebuilt easily. Second, we build a bitmap index for each storage node which only indexes the data residing on the node. Third, we organize the storage nodes as a structured overlay and each node maintains a portion of the global index for the all different data. The global index is a bitmap index to indicate the node each data resides in. Finally, based on the three-level index scheme, some query algorithms are implemented. We conduct extensive experiments on a LAN and the results demonstrate that our indexing scheme is dynamic, efficient and scalable. © 2013 Asian Network for Scientific Information.

关键词： Cloud computing

来源：评论

学校读者我要写书评

暂无评论

A clustering-based approach for mining dockerfile evolutionary trajectories

引用

Science China(Information Sciences) 2019年第1期62卷 211-213页

作者： Yang ZHANG Huaimin WANG Vladimir FILKOV Key Laboratory of Parallel and Distributed Computing National University of Defense Technology College of Computer National University of Defense Technology DECAL Lab University of California Computer Science Department University of California

Dear editor,Docker1), as a de-facto industry standard [1], enables the packaging of an application with all its dependencies and execution environment in a light-weight, self-contained unit, i.e., *** launching the container from Docker image, developers can easily share the same operating system, libraries, and binaries [2]. As the configuration file, the dockerfile plays an important role,

关键词： A clustering-based approach for mining dockerfile evolutionary trajectories

来源：评论

学校读者我要写书评

暂无评论

A distributed paging RAM grid system for wide-area memory sharing 20

A distributed paging RAM grid system for wide-area memory sh...

引用

20th IEEE International parallel and distributed processing Symposium, IPDPS 2006

作者： Chu, Rui Xiao, Nong Zhuang, Yongzhen Liu, Yunhao Lu, Xicheng National Key Laboratory for Parallel and Distributed Processing ChangSha HuNan China Hong Kong University of Science and Technology Clear Water Bay Kowloon Hong Kong

ISBN: (纸本)1424400546

Memory-intensive applications often suffer from the poor performance of disk swapping when memory is inadequate. Remote memory sharing schemes, which provide a remote memory that is faster than the local hard disk, are able to improve the performance of such applications. Due to the limitation of being applicable within single clusters only, however, most of the previous remote memory mechanisms, such as the network memory scheme, fail to be extendable into a large scale, distributed, heterogeneous, and dynamic environment. In this work, we propose a service-oriented grid memory sharing scheme, distributed Paging RAM Grid (DPRG). We study the properties and criteria of large scale memory sharing, and then design major operations and optimizations to fit the usage of grid systems. We collect trace from our grid environment, and evaluate DPRG through comprehensive trace-driven simulations. Results show that DPRG significantly outperforms existing remote memory sharing schemes and supports grid computing applications effectively. © 2006 IEEE.

关键词： Storage allocation (computer)

来源：评论

学校读者我要写书评

暂无评论

Efficient Large Models Fine-tuning on Commodity Servers via Memory-balanced Pipeline parallelism 25

Efficient Large Models Fine-tuning on Commodity Servers via ...

引用

25th IEEE International Conferences on High Performance Computing and Communications, 9th International Conference on Data Science and Systems, 21st IEEE International Conference on Smart City and 9th IEEE International Conference on Dependability in Sensor, Cloud and Big Data Systems and Applications, HPCC/DSS/SmartCity/DependSys 2023

作者： Liu, Yujie Lai, Zhiquan Liu, Weijie Wang, Wei Li, Dongsheng College of Computer National University of Defense Technology National Key Laboratory of Parallel and Distributed Computing Changsha China

ISBN: (纸本)9798350330014

Large models have achieved impressive performance in many downstream tasks. Using pipeline parallelism to fine-tune large models on commodity GPU servers is an important way to make the excellent performance of large models available to the general public. Previous solutions fail to achieve an efficient memory-balanced pipeline parallelism. In this poster, we introduce a memory load-balanced pipeline parallel solution. This solution balances memory consumption across stages on commodity GPU servers via NVLink bridges. It establishes a new pathway to offload data from GPU to CPU by using the PCIe link of adjacent GPUs connected by the NVLink bridge. Furthermore, our method orchestrates offload operations to minimize the offload latency during large model fine-tuning. Experiments demonstrate that our solution can balance the memory footprint among pipeline stages without sacrificing training performance. © 2023 IEEE.

关键词： Program processors

来源：评论

学校读者我要写书评

暂无评论

A distributed Approach to Consistent Order Delivery of Concurrent Events in Asynchronous DVEs

A Distributed Approach to Consistent Order Delivery of Concu...

引用

2010 2nd International Conference on Education Technology and Computer(第二届IEEE教育技术与计算机国际会议 ICETC 2010)

作者： Hangjun Zhou Wei Zhang Yuxing Peng Sikun Li Key Laboratory of Science and Technology for National Defence of Parallel and Distributed Processing Key Laboratory of Science and Technology for National Defence of Parallel and Distributed Processing

ISBN: (纸本)9781424463671;9781424463688;9781424463701

In large-scale asynchronous distributed virtual environments(DVEs), one of the difficult problems is to deliver the concurrent events in a consistent order at each node. Generally, the previous consistency control approaches can be classified into two categories: causal order and time stamped order. However, causal order approaches can merely preserve the cause-effect relation of events and time stamped order approaches seem intrinsically complex to be used in serverless large-scale asynchronous DVEs. In this paper, we proposed a novel distributed algorithm to identify the concurrent events and preserve the consistent order delivery of them at different nodes. Simulation studies are also carried out to compare the performance of this algorithm with that of the previous ones. The results show that the new algorithm can effectively deliver the concurrent events in consistent order at each node and is more efficient than the previous algorithms in large-scale asynchronous DVEs.

关键词： distributed virtual environments concurrent events distributed consistency algorithm

来源：评论

学校读者我要写书评

暂无评论

Topology awareness algorithm for virtual network mapping

引用

Journal of Zhejiang University-Science C(Computers and Electronics) 2012年第3期13卷 178-186页

作者： Xiao-ling LI Huai-min WANG Chang-guo GUO Bo DING Xiao-yong LI Wen-qi BI Shuang TAN National Key Laboratory of Parallel and Distributed Processing National University of Defense Technology.Changsha 410073China School of Computer National University of Defense TechnologyChangsha 410073China China Electronic Systems Engineering Corporation Beijing 100039China The Northern Institute of Electronic Equipment of China Beijing 100083China

Network virtualization is recognized as an effective way to overcome the ossification of the Internet. However, the virtual network mapping problem (VNMP) is a critical challenge, focusing on how to map the virtual networks to the substrate network with efficient utilization of infrastructure resources. The problem can be divided into two phases: node mapping phase and link mapping phase. In the node mapping phase, the existing algorithms usually map those virtual nodes with a complete greedy strategy, without considering the topology among these virtual nodes, resulting in too long substrate paths (with multiple hops). Addressing this problem, we propose a topology awareness mapping algorithm, which considers the topology among these virtual nodes. In the link mapping phase, the new algorithm adopts the k-shortest path algorithm. Simulation results show that the new algorithm greatly increases the long-term average revenue, the acceptance ratio, and the long-term revenue-to-cost ratio (R/C).

关键词： Network virtualization Ossification Virtual network (VN) mapping Substrate network (SN) Topology awareness Acceptance ratio

来源：评论

学校读者我要写书评

暂无评论

Effective top-down active learning for hierarchical text classification

引用

17th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2013

作者： Li, Xiao Ling, Charles X. Wang, Huaimin Department of Computer Science University of Western Ontario Canada National Laboratory for Parallel and Distributed Processing National University of Defense Technology China

ISBN: (纸本)9783642374555

Hierarchical text classification is an important task in many real-world applications. To build an accurate hierarchical classification system with many categories, usually a very large number of documents must be labeled and provided. This can be very costly. Active learning has been shown to effectively reduce the labeling effort in traditional (flat) text classification, but few works have been done in hierarchical text classification due to several challenges. A major challenge is to reduce the so-called out-of-domain queries. Previous state-of-the-art approaches tackle this challenge by simultaneously forming the unlabeled pools on all the categories regardless of the inherited hierarchical dependence of classifiers. In this paper, we propose a novel top-down hierarchical active learning framework, and effective strategies to tackle this and other challenges. With extensive experiments on eight real-world hierarchical text datasets, we demonstrate that our strategies are highly effective, and they outperform the state-of-the-art hierarchical active learning methods by reducing 20% to 40% queries. © Springer-Verlag 2013.

关键词： Text processing

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：