Accelerating temporal action proposal generation via high performance computing

Temporal action proposal generation aims to output the starting and ending times of each potential action for long videos and often suffers from high computation cost. To address the issue, we propose a new temporal convolution network called Multipath Temporal ConvNet (MTCN). In our work, one novel high performance ring parallel architecture based is further introduced into temporal action proposal generation in order to respond to the requirements of large memory occupation and a large number of videos. Remarkably, the total data transmission is reduced by adding a connection between multiplecomputing load in the newly developed architecture. Compared to the traditional Parameter Server architecture, our parallel architecture has higher efficiency on temporal action detection tasks with multiple GPUs. We conduct experiments on ActivityNet-1.3 and THUMOS14, where our method outperformsother state-of-art temporal action detection methods with high recall and high temporal precision. In addition, a time metric is further proposed here to evaluate the speed performancein the distributed training process.     

一种面向高性能计算的多FPGA互连结构及划分方法

针对高性能计算系统在大规模通信互连中面临的性能、成本及功耗等问题,融合新兴的高速互连技术,结合大规模、超大规模系统通信的局部性和异构性,提出基于多FPGA的混合层级高速互连结构,并给出基于集群的多FPGA逻辑功能划分方法。该方法能够根据不同应用自定义设计高效互连网络,降低大规模计算系统的互连成本和开销。通过应用实例实验证明,该方法能够实现大规模设计向多FPGA高性能计算平台的快速映射,加速高性能可配置计算系统的设计实现。     

Extending next generation network (NGN) architecture for connection-oriented transport

Next Generation Network (NGN) is the architecture of the International Telecommunication Union-Telecommunication Standardization Sector (ITU-T in short) supporting the provisioning of Quality of Service (QoS)-guaranteed services over different packet transport technologies. Such capability derives from the effectiveness of a dynamic resource control performed by the Resource Admission Control Function (RACF) at service set-up.Control Plane (CP)-enabled connection-oriented transport networks can guarantee the QoS support for new bandwidth-greedy NGN services across the optical transport segment thanks to the ability of automatic path set-up and traffic segregation. But the Internet Engineering Task Force (IETF) standard for the CP in transport networks, i.e., the Generalized Multi-Protocol Label Switching (GMPLS) is not yet included within the NGN supported transport technologies.In this work, we outline architectural guidelines and design strategies for ITU-T RACF employment across GMPLS-controlled networks while providing a viable solution for dynamic resource control that takes into account operational issues for the integration of GMPLS capabilities within NGN architecture (i.e., supported interfaces, actual node capabilities).An NGN prototype implementing the proposed architectural enhancement is also presented as a proof of concept. The prototype highlights how the extended ITU-T NGN can set-up Multimedia over IP (MoIP) services using GMPLS-controlled transport objects.     

A differentiated services architecture for multimedia streaming in next generation Internet

This paper presents a Differentiated Services (Diffserv or DS) architecture for multimedia streaming applications. Specifically, we define two types of services in the context of Assured Forwarding (AF) per hop behavior (PHB) that are differentiated in terms of reliability of packet delivery: the High Reliable (HR) service and the Less Assured (LA) service. We propose a novel node mechanism called Selective Pushout with Random Early Detection (SPRED) that is capable of simultaneously achieving the following four objectives: (1) a core router does not maintain any state information for each flow (i.e., core-stateless); (2) the packet sequence within each flow is not re-ordered at a node; (3) packets from HR service are delivered more reliably than packets from LA service at a node during congestion; and (4) packets from TCP traffic are dropped randomly to avoid global synchronization during congestion. We show that SPRED is a generalized buffer management algorithm of both tail-dropping and Random Early Detection (RED), and combines the best features of pushout (PO), RED and RED with In/Out (RIO) mechanisms. Simulation results demonstrate that under the same link speed and network topology, network nodes employing our Diffserv architecture have substantial performance improvement over the current Best Effort (BE) Internet architecture for multimedia streaming applications.     

MPFS: A truly scalable router architecture for next generation Internet

A new generation architecture of IP routers called massive parallel forwarding and switching （MPFS） is proposed, which is totally different from modern routers. The basic idea of MPFS is mapping complicated forwarding process into multilevel scalable switch fabric so as to implement packet forwarding in a pipelining and distributed way. This processing mechanism is named forwarding in switching （FIS）. By interconnecting multi-stage, lower speed components, called forwarding and switching nodes （FSN）, MPFS achieves better scalability in forwarding and switching performance just like MPP. We put emphasis upon IPv6 lookup problem in MPFS and propose a method for partitioning IPv6 FIB and mapping them to switch fabric. Simulation and computation results suggest that MPFS routers can support line-speed forwarding with a million of IPv6 prefixes at 40 Gbps. We also propose an implementation of 160 Tbps core router based on MPFS architecture at last.     

A human-in-the-loop manufacturing control architecture for the next generation of production systems

In recent years, the introduction of Industry 4.0 technologies in the manufacturing landscape promoted the development of smart factories characterised by relevant socio-technical interactions between humans and machines. In this context, understanding and modelling the role of humans turns out to be crucial to develop efficient manufacturing systems of the future. Grounding on previous researches in the field of Human-in-the-Loop and Human Cyber-Physical Systems, the paper aims at contributing to a deep reflection about human-machine interaction in the wider perspective of Social Human-in-the-Loop Cyber-Physical Production Systems, in which more agents collaborate and are socially connected. After presenting an evolution of manufacturing control organisations, an architecture to depict social interactions in smart factories is proposed. The proposed architecture contributes to the representation of different human roles in the smart factory and the exploration of both hierarchical and heterarchical data-driven decision-making processes in manufacturing.     

Building a next generation Internet with source address validation architecture

The IP packet forwarding of current Internet is mainly destination based. In the forwarding process, the source IP address is not checked in most cases.This causes serious security, management and accounting problems. Based on the drastically increased IPv6 address space, a ＂source address validation architecture＂ （SAVA） is proposed in this paper, which can guarantee that every packet received and forwarded holds an authenticated source IP address. The design goals of the architecture are lightweight, loose coupling, ＂multi-fence support＂ and incremental deployment. This paper discusses the design and implementation for the architecture, including inter-AS, intra-AS and local subnet. The performance and scalability of SAVA are described. This architecture is deployed into the CNGI-CERNET2 infrastructure a large-scale native IPv6 backbone network of the China Next Generation Internet project. We believe that the SAVA will help the transition to a new, more secure and dependable Internet.     

下一代电信网络性能监控系统设计

为了主动发现和跟踪解决电信网络服务质量问题,支持网络管理业务的平滑过渡,提出面向性能问题生命周期管理构建性能监控系统。参考下一代运营操作系统和软件,构建了端到端的性能问题生命周期管理流程和性能监控共享信息数据模型,设计了以组件为基础的系统结构。在运营商网络管理系统的基础上对系统进行实现和集成应用,应用表明,性能监控系统能有效支撑对性能问题的发现、分析和解决效果评估,并支持灵活的系统集成和持续的业务演进。     

An architecture for high performance engineering informationsystems

Commercially available database systems do not meet the information and processing needs of design and manufacturing environments. A new generation of systems-engineering information systems-must be built to meet these needs. The architectural and computational aspects of such systems are addressed, and solutions are proposed. The authors argue that a mainframe-workstation architecture is needed to provide distributed functionality while ensuring high availability and low communication overhead, that explicit control of metaknowledge is needed to support extendibility and evolution, that large rule bases are needed to make the knowledge of the systems active, and that incremental computation models are needed to achieve the required performance of such engineering information systems     

高性能计算机系统相对持续性能度量模型

高性能计算机系统的持续性能是反映实际领域应用中高性能计算机系统性能强弱的重要度量标准。简单介绍了高性能计算机系统常用的性能评价方法,结合应用基准程序集,提出了相对持续性能的度量模型。实验基于高性能计算气象应用评测,结果表明应用相对持续性能度量模型可区分不同厂商的高性能计算机系统的性能强弱,为高性能计算机系统的选择提供参考,并在一定程度上反映了气象应用本身的可扩展性。     

面向高性能计算的分布式故障定位框架

针对高性能计算系统中故障定位难度高且实时性差的问题,提出了一种基于消息传递的故障定位框架（MPFL）,包括基于树形拓扑的故障检测（TFD）和故障分析（TFA）算法。首先,在并行作业初始化时,将所有参与计算的节点进行逻辑上的树形划分,生成故障定位树（FLT）,并将故障定位任务分布到节点上;然后,当消息库、操作系统等组件检测到节点异常状态时,基于TFD算法分析作业的FLT结构,根据负载平衡、性能开销等因素选择接收异常状态的节点;最后,节点利用TFA算法对接收到的异常状态进行推理得出故障,TFA算法使用基于规则的事件关联,并基于消息传递设计轻量级的主动探测,将两种方式相结合,提高了故障分析的准确性。实验以模拟节点停机故障为定位目标,并以NPB-FT与NPB-IS为基准测试,在集群上对MPFL框架进行了评估。实验结果表明,MPFL框架在故障定位能力与开销节省方面表现突出。     

High-speed,in-band performance measurement instrumentation for next generation IP networks

Facilitating always-on instrumentation of Internet traffic for the purposes of performance measurement is crucial in order to enable accountability of resource usage and automated network control, management and optimisation. This has proven infeasible to date due to the lack of native measurement mechanisms that can form an integral part of the network’s main forwarding operation. However, Internet Protocol version 6 (IPv6) specification enables the efficient encoding and processing of optional per-packet information as a native part of the network layer, and this constitutes a strong reason for IPv6 to be adopted as the ubiquitous next generation Internet transport.In this paper we present a very high-speed hardware implementation of in-line measurement, a truly native traffic instrumentation mechanism for the next generation Internet, which facilitates performance measurement of the actual data-carrying traffic at small timescales between two points in the network. This system is designed to operate as part of the routers’ fast path and to incur an absolutely minimal impact on the network operation even while instrumenting traffic between the edges of very high capacity links. Our results show that the implementation can be easily accommodated by current FPGA technology, and real Internet traffic traces verify that the overhead incurred by instrumenting every packet over a 10 Gb/s operational backbone link carrying a typical workload is indeed negligible.     

Superscalar architecture design for high performance DSP operations

This paper evaluates the possibility of using a general purpose superscalar architecture as the main computational engine for high performance DSP algorithms. Real-time sample rate conversion (SRC) in a software defined radio (SDR) has been taken as an example representing a class of computationally demanding DSP tasks. This scenario corresponds to digital filters operating at a high sampling rate in intermediate frequency (IF) stage of a multi-standard wireless transceiver. However, instead of a dedicated signal processing engine, a superscalar processor is designed for SRC implementation. An iterative, SimpleScalar based architectural modeling tool has been developed to analyze various parameters of superscalar processors. Both power and performance metrics have been taken under consideration to come up with an efficient design. It has been shown that the resulting superscalar architecture can provide a fully programmable solution capable of supporting future wireless communication standards in real-time. The design methodology explored in this work can be extended to obtain efficient processor architectures for a range of other applications.     

集群式高性能计算系统研究

研究了集群的系统结构和主要优势,以及集群式高性能计算系统的诞生;分析了集群式高性能计算系统的架构和构建方式,集群构建包括网络部署、存储系统、计算节点、管理节点、登录节点等部分。在此基础上构建了基于Linux的集群式高性能计算系统。     

高性能交换结构综述

简述交换结构分类,从交换体系结构、调度算法和QoS保证等方面分别介绍了当前研究较多的单级、多级和多平面多级等多种交换网络,并简述这些交换网络的关键要素和研究结果,对比它们的实现复杂度、可扩展性、吞吐率、QoS保证等,指出高速交换结构发展中值得研究的若干问题,为进一步研究高速交换结构提供参考.