浅谈数据中心网络架构的发展

网络是数据中心的一个重要基础设施,而通常分为两大部分：前端计算网络以及后端存储网络.后端的存储网络目前主要是FC网络,不过随着IP技术的发展,采用FCoE技术融合前后端网络成为一种趋势.本文主要关注前端计算网络.     

Off-line optimization and control for real time integration of a three-phase hydrogenation catalytic reactor

In order to develop and test the integration procedure, in this paper a real time process integration involving the optimization and control of the process is presented, in this case, with the two-layer approach. The used optimization algorithms were Levenberg–Marquardt and SQP that solve a non-linear least square problem subject to bounds on the variables. The two-layer approach is a hierarchical control structure where an optimization layer calculates the set points and manipulated variables to the advanced controller, which is based on the dynamic matrix control with constraints (QDMC). The non-isothermal dynamic model of the three-phase slurry catalytic reactor with appropriate solution procedure was utilized in this work (Vasco de Toledo, E. C., Santana, P. L., Maciel, M. R. W., & Maciel Filho, R. (2001). Dynamic modeling of a three-phase catalytic slurry reactor. Chemical Engineering Science, 56, 6055–6061). The model consists on mass and heat balance equations for the catalyst particles as well as for the bulk phases of gas and liquid. The model was used to describe the dynamic behavior of hydrogenation reaction of o-cresol to obtain 2-methil-cyclohexanol, in the presence of a catalyst Ni/SiO₂.     

A reactive extension of the OpenMusic visual programming language

Objectives: OpenMusic (OM) is a domain-specific visual programming language designed for computer-aided music composition. This language based on Common Lisp allows composers to develop functional processes generating or transforming musical data, and to execute them locally by demand-driven evaluations. As most historical computer-aided composition environments, OM relies on a transformational declarative paradigm, which is hard to conciliate with reactive data-flow (an evaluation scheme more adequate to the development of interactive systems). We propose to link these two evaluation paradigms in the same and consistent visual programming framework.Methods: We establish a denotational semantics of the visual language, which gives account for its demand-driven evaluation mechanism and the incremental construction of programs. We then extend this semantics to enable reactive computations in the functional graphs.Results: The resulting language merges data-driven executions with the existing demand-driven mechanism. A conservative implementation is proposed.Conclusions: We show that the incremental construction of programs and their data-driven and demand-driven evaluations can be smoothly integrated in the visual programming workflow. This integration allows for the propagation of changes in the programs, and the evaluation of graphically designed functional expressions as a response to external events, a first step in bridging the gap between computer-assisted composition environments and real-time musical systems.     

Coding scheme based on spherical polar coordinate for control over packet erasure channel

Like feedback strategy, coding scheme is an important part of networked control system design as well. On the basis of spherical polar coordinate, a novel coding scheme is proposed for stabilization problem of discrete linear time invariant system subject to packet erasure channel with feedback. The coding scheme uses encoder without access to control inputs. In the case that a decoder does not use control inputs, a definite relation between the quantized data and the corresponding quantization error is established, which helps to analyze the stability of system, and a selective quantization method is adopted, by which finite data rate is obtained. In the case that a decoder uses control inputs, instead of quantizing the system state at each time step as usual, the encoder quantizes the initial state all the time by updating quantizer. Sufficient conditions guaranteeing the system stable are presented for two cases, respectively, and the corresponding design methods for coding schemes are given. Copyright © 2012 John Wiley & Sons, Ltd.     

基于R软件的金融资产收益波动率建模分析

金融资产收益波动率是金融计量分析的核心议题,该文考虑对其建模分析的计算机实现问题。利用R软件的强大计算与绘图功能,给出金融资产收益波动率各个模型的模拟实现以及对实际金融市场数据的各种建模分析。     

kinect深度数据随机跳变误差分析及修正算法研究

采用微软空间扫描设备kinect的测距数据会产生随机跳变误差现象。通过分析误差原因及实验验证了跳变误差随距离增加曲线上升的特点。为修正该误差设计了一个基于像素滤波技术的误差修正算法。在kinect有效距离内连续观测三个待测物的空间深度数据和滤波数据。经实验验证在使用10帧深度数据可以将跳变误差在1.5m以内的跳变误差控制在5mm以内;1.5~3m以内的随机跳变误差控制大部分在10mm以内;超过3m的随机误差均匀分布在25mm以内。     

基于k-means改进算法的入侵检测系统研究

入侵检测技术作为一种主动的网络安全防护技术越来越引起研究者的关注,该文以k-means算法为基础,对基于k-means算法的入侵检测系统进行了研究和分析,指出了传统k-means算法的不足,提出了相应的改进策略,在此基础上完成基于k-means改进算法的入侵检测系统的研究。     

基于增强ETL过程的大数据策略研究

大数据不仅意味着大机遇,同样带来了大噪声。因此,任何具有价值的大数据策略都必须给出滤除噪音,同时对真正相关的数据进行有效且高效的处理和解决办法。该文分析了大数据背景下企业的数据策略和现有IT解决方案中存在的问题,提出通过优化ETL过程来增强早期数据整合的解决方案,以提高企业大数据策略的成本收益。     

视频服务器拥塞控制算法的研究与实现

分析现有网络拥塞控制算法的优缺点,在此基础上,根据多媒体数据和嵌入式设备的特点提出改进的基于视频服务器的拥塞控制算法。此算法以UDP协议为基础,通过控制传输速率来降低丢包率和网络时延,提高了嵌入式设备的多媒体数据传输质量。同时,将此算法应用于基于TMS320DM6467T硬件开发平台和Linux操作系统的视频服务器上,并通过实验验证了算法的可靠性与实用性。     

回归神经网络预测模型归一化方法分析

该文运用几种常见数据归一化方法分别对自回归神经网络动态预测模型的预测性能进行分析,结果说明不同数据归一化处理对模型的性能影响非常明显,运用最大运算法进行归一化处理要优于其它几种常见归一化方法。