基于性能计数事件的计算机功耗估算模型

精准快速获取计算机系统的实时功耗是功耗优化研究的基础,因此提出并建立了一种高精度的计算机功耗估算模型。通过分析统计系统运行时代表性的性能计数事件,应用机器学习理论分析性能事件与功耗的关系,建立多核计算机系统实时功耗估算模型。模型构建时使用多元线性回归(multiple linear regression,MLR)方法以及支持向量回归(support vector regression,SVR)方法分析两者关系,并对两种方法建立的功耗估算模型进行了对比分析。实验结果表明,基于性能事件的功耗估算模型可准确估计计算机实时功耗,估算误差不高于3%。与已有模型相比较,该估算模型精度更高、通用性更好。

城市民宿道德-性能复合型负面事件对购买意愿的影响——以偷拍为例

通过情景模拟问卷调查,从道德-性能复合型负面事件角度切入,分析城市民宿偷拍负面事件对购买意愿的影响程度与机理,探讨先前城市民宿入住经验、消费者人口统计特征对购买意愿影响的调节作用.结果表明,负面事件发生后,消费者购买意愿显著降低;道德-性能复合型负面事件既通过影响感知质量,又通过影响品牌形象负向影响消费者购买意愿;道德-性能复合型负面事件对消费者购买意愿的伤害程度随先前城市民宿入住经验、性别、学历发生变化,没有入住经验、女性、高学历消费者,购买意愿受损程度显著高于有入住经验、男性、中低学历消费者.

移动节点随机事件捕获性能的统计分析

在事件随机变量参数未知的指数分布下,讨论分布参数的Bayes估计。首先,通过不同调度方案分析移动节点随机事件捕获的性能;其次,采用随机事件参数未知下的在线变速调度方案和随机事件参数已知下的离线匀速调度方案,研究移动节点事件捕获性能对工作状态速度的影响;最后,采用MATLAB软件进行仿真,结果表明:在线变速调度方案下的捕获性能优于离线变速或匀速调度方案、在线匀速调度方案,且在线变速调度方案是离线匀速调度方案的一种实时修正。

基于功耗计数器的处理器功耗实时估算方法

针对现有的2类处理器功耗实时估算方法的不足,本文开发了一种基于功耗计数器的处理器功耗实时估算方法。该方法结合了基于性能事件计数器和基于电路信号方法的优点,利用功耗计数器记录与处理器功耗密切相关的电路信号的变化次数,能够以较小的观测粒度实时估算处理器的功耗,并且具有较低的硬件开销。本文在龙芯GS364处理器上实现并评估了该方法。结果表明,使用16个功耗计数器可以在512个时钟周期的粒度内实现0.83%的估算误差,与此同时,它的硬件开销只占处理器总面积的0.063%。

SNMP网络性能管理系统的设计与实现

首先介绍了SNMP的基本原理和网络性能管理的系统模型。较为详细的阐述了一个基于SNMP的网络性能管理系统的设计与实现,其特点是基于策略执行、区分服务及分布式管理等,能以丰富的图表等形式直观的反映出当前的网络性能状况。

SDH传输系统的性能监视

首先提出性能监视的总体概念，然后从性能事件处理、性能数据采集以及性能历史数据处理3个方面对性能监视进行了详细分析。

采用机器学习算法的软件能耗感知模型及其应用

针对软件开发人员不了解代码的能耗,或者即使能测得软件运行时的能耗但由于缺乏对影响能耗的各因素了解的问题,在对4种机器学习算法实验的基础上,选用岭回归算法建立了以软件性能事件为特征的软件能耗模型。以收集的软件能耗测量值和软件性能事件数据作为训练样本,利用岭回归机器学习算法建立平均能耗模型,用平均能耗模型来预测其他软件的平均能耗;并在此基础上,利用软件运行的实时特征提出了实时的能耗模型。实验结果表明:以测量的能耗为基准,采用所提出的平均能耗模型对软件能耗进行预测的误差率在9%以内;与平均能耗模型相比,实时能耗模型的预测误差率更低。采用本文所提出的两种能耗模型,软件设计人员不需要对自己的软件进行测量和训练就可以较准确地估算出软件的能耗并能解释能耗产生的原因,最后根据预测结果对软件的能效进行优化。

一种改进的整机系统实时功率建模方法

本文通过考察处理器、内存和磁盘三个部件的活动信息,包括硬件性能事件和部件利用率,构建整机系统实时功率模型.对于处理器和内存的功耗,通过特征选择找到合适的处理器内部的性能事件,结合CPU功耗状态C-States,建立低误差的功耗模型;对于没有提供性能事件的磁盘,通过利用率对部件功耗进行估算.采用回归方法分别建立并综合两个模型即可得到整机的实时功率模型.实验使用了真实系统Eureqa建模和PosgreSQL系统运行TPC-H负载对两种功率模型进行验证,结果表明:基于硬件性能事件与C-States的实时处理器与内存功率模型的误差在3%以下;结合性能事件、C-States与利用率的整机功率模型其相对误差在4%~10%之间,优于仅使用利用率的整机功率模型,最大时提升可达7%.

Linux操作系统网络软件时延的研究

Linux操作系统在高性能计算环境、机群系统、高性能安全网关等方面日益发挥着重要作用 ,因此 ,了解和掌握Linux操作系统的网络软件性能具有重要意义 利用Netslice———一种自动分析网络软件时延的工具 ,对Linux操作系统的网络软件时延进行深入和全面的测评与分析 实验发现 ,在发送同样大小的数据包时 ,网络软件的接收时延远远大于发送时延 进一步的研究指出 ,非对齐数据的拷贝和上下文切换是造成接收额外开销的主要原因 实验同时指出 ,Lin

程序能耗测量分析工具FPowerTool及其能耗优化实践

能耗感知编程(EAP)是通过优化软件的能效来降低计算系统能耗的一种新途径。它把能耗作为主要指标引入软件开发的过程,通过调整代码的编写方式,降低程序的执行能耗。能耗感知编程的实现面临能耗热点发现、耗能原因确定和问题代码定位等问题。为了解决这些问题,提出了一种程序能耗与性能事件协同测量与分析的新方法EPC。首先,简要论述EPC的基本原理和基于该方法的程序能耗测量与分析工具FPowerTool的实现。然后,介绍如何对能耗和性能事件做关联分析,找出影响程序能耗的主要因素。最后,通过一组程序优化案例,说明如何通过关联分析定位与高能耗相关的程序代码,进而改变代码编写和数据放置与访问方式,使程序的执行能耗得到降低。实验结果表明,基于EPC方法提供的能耗感知和分析能力,通过改进数据定义、赋值、放置与访问方式等可以改善程序性能和能效。

通用图形处理器功耗估算模型

为精准快速地获得GPU功耗数据,提出一种基于硬件性能计数事件的通用图形处理器(GPGPU)功耗估算方法。通过分析GPGPU程序运行时的功耗分布情况,选择一组与应用程序运行功耗密切相关的硬件性能计数事件集合,使用反向传播人工神经网络分析硬件性能计数事件与实时功耗间的关系,最终建立GPGPU功耗估算模型。实验结果表明,与多元线性回归的功耗估算模型相比,该模型具有更高的估算准确性和通用性。

传输通道误码问题处理

本文首先对同步传输系统（SDH）中误码的度量、误码检测机理以及误码对传输设备所承载业务的影响进行了阐述。然后对误码产生的原因进行了详细解析，并详细介绍了实际工作中传输设备误码问题处理的一般方法和步骤。

利用深度学习的硬件计数器复用估计算法

利用深度学习方法,为硬件计数器复用(multiplexing,MPX)提供结果精度更高的估计模型。通过对MPX估计得到的结果与实际采集的真实数据进行相似性分析,证明相同程序多次运行之间得到的硬件计数值是线性相关的。采用神经网络多层感知器(multilayer perceptron,MLP)和双向门控神经网络(bidirectional gated recurrent unit,Bi-GRU)这2种深度学习模型,对MPX数据进行拟合。基于动态时间规整(dynamic time warping,DTW),提出一个全新的评估MPX数据精度的指标DTW-cost。实验结果表明,同时收集15个硬件事件数据时,MLP方法拟合得到的13个高性能计算应用平均准确率比现有使用最广的固定插值法高出10.53%,最多可提升19.8%;而在MLP表现较差的事件上,Bi-GRU方法得到的平均准确率提升了28.8%。

Design, Analysis and Installation of Offshore Instrumented Moored Data Buoy System

This paper discusses the analysis done on the meteorological ocean buoy mooring used for monitoring the Indian seas. Based on the extreme environmental parameters experienced by the buoys, mooring loads are analyzed using offshore dynamic analysis software. The results obtained are validated with the tension recorder installed in one of the moorings, and the results are found to comply with an accuracy of better than 1%. The successful on demand performance of the mooring during major cyclones in the Bay of Bengal and the vital meteorological and oceanographic information provided by the buoy during these disastrous cyclonic events validates the mooring design, and proves the data availability for societal needs. The time critical data assimilated in the cyclone prediction models have given confidence to improve the country＇s weather prediction and climate modelling capabilities.

Television Sports News Between Performance, Violence and Spectacle

The purpose of this paper is to analyze today＇s construction of sports news, and to see the way that media present the live sports events and how do livc sports events become global breaking news. We started with the theoretical background and we analyzed the audiences of the TV channels dedicated to sports. Media and violence is a well-known subject, and many theorists have analysed it but little research has been conducted about the effects that violence in sport news may have on people or society. Therefore, the aim of this study is to provide insights into these effects, to try to demonstrate that today＇s sports news does not focus on pcrformancc any more, but they have become a spectacle of violence and gossip. After reviewing the literature in order to gain a perspective on this topic, we focused on the complex reconstructive and subjective nature of news rcception, the different ways the audience constructs meaningful intcrpretations from the news, exposure to news and the construction of violent images of reality.

TanDEM-X Autonomous Formation Flying System： Flight Results

The TanDEM-X mission is a scientific and commercial Earth observation mission comprising two satellites flying in close formation. The formation maintenance can be advantageously performed by an onboard autonomous system, which reduces the operational efforts, provides a shorter reaction time in case of contingencies and increases the control performance. The TanDEM-X Autonomous Formation Flying （TAFF） system has been developed for this purpose and is intended to replace the ground-based formation keeping activities during routine operations. TAFF has been activated for the first time in October 2010 for commissioning, during which the autonomous usage of thrusters was prohibited. Afterwards, a closed-loop campaign was successfully conducted in March 2011, demonstrating the capability of TAFF to maintain autonomously the formation. After a brief technical description of the system, the paper presents the key results gained during the commissioning phase and the closed-loop campaign,

Damage Detection of Offshore Jacket Structures Using Frequency Domain Selective Measurements

The development of damage detection techniques for offshore jacket structures is vital to prevent catastrophic events. This paper applies a frequency response based method for the purpose of structural health monitoring. In efforts to fulfill this task, concept of the minimum rank perturbation theory has been utilized. The present article introduces a promising methodology to select frequency points effectively. To achieve this goal, modal strain energy ratio of each member was evaluated at different natural frequencies of structure in order to identify the sensitive frequency domain for damage detection. The proposed methodology opens up the possibility of much greater detection efficiency. In addition, the performance of the proposed method was evaluated in relation to multiple damages. The aforementioned points are illustrated using the numerical study of a two dimensional jacket platform, and the results proved to be satisfactory utilizing the proposed methodology.

Factors Affecting Polymer Translocation Through a Nanopore in a Membrane

Monte Carlo simulations were used to study the translocation of a flexible polymer through a pore in a membrane, assuming an attractive interaction between the monomers and the membrane on the trans side of the membrane and no interaction on the cis side. For the case T〈Tc （the temperature corresponding to the minimum in the translocation time τ）, the value of τdecreases with increasing temperature, whereas for T〉Tc, τ increases with increasing temperature. The translocation time depends on the absorbed energy uo in a nontrivial way. The value of τ increases initially upon increasing uo before it begins to decrease. The variation of the translocation time with respect to the solvent quality was also studied. It showed that there is a transition, as the solvent quality improves from ＂poor＂ to ＂good＂： when εAB〈εc （the interaction energy corresponding to the minimum in τ）, τdecreases with increasing the value of εAB; when εAB〉εc, τincreases with increasing εAB- When the chain length was changed, it was found that when the absorbed energy uo was greater than uc,τ was proportional to N1.602; for uo〈uc, τ∝N2.248. As the solvent quality improved from ＂poor＂ to ＂good,＂ the translocation probability increased initially before becoming stable.

New Precoded Spatial-Multiplexing for an Erasure Event in Single Frequency Networks

In this paper, we propose a new spatial-multiplexing(SM) scheme employing an orthogonal precoder over Rayleigh-fading channels for an erasure event in single frequency networks(SFN). To optimize the precoder, the average bit error rate(BER) is evaluated and minimized through a mathematical analysis. Compared to an ordinary SM, the proposed scheme guarantees identical BER performance under non-erasure fading channels and achieves a greatly improved performance under erasure fading channels, especially for a higher erasure-ratio and SNR values. This improvement is mainly due to the increase in the diversity gain incurred by the optimized precoder over the erasure event. We also compare the performance of the proposed SM to that of the conventional constellation-rotation(CR) scheme applied to the single antenna SFN systems. The results of a computer simulation show that the performance of the new scheme is more effective than that of a conventional CR across all simulation cases.

The Post-Merger Performance： Evidence From Italy

The objective of this paper is to investigate whether mergers create value for shareholders in both the short and long term. For this purpose, 120 announcements of mergers that were registered in Italy during the period 1994-2006 among listed companies were examined. The short-term analysis was conducted using the event study methodology in order to estimate the cumulative abnormal returns （CARs） in the time window around the announcement date （-10, ＋10）. In this work, the sample of 120 mergers was divided into two sub-samples： the first considers the mergers that were carried out in all sectors of the economy, and the second focuses only on bank mergers. From the results obtained it would appear that, while the sub-sample of all mergers registered a statistically significant value creation for the shareholders of both the bidder and target companies, values also confirmed by combined analysis, the second sub-sample registered negative values for bidder companies and positive values for target companies. Negative values also seem to be confirmed by the results of the combined analysis both at the date of announcement and throughout the entire period of observation. For the long-term analysis, the Buy and Hold Abnormal Returns （BHARs） methodology was used, with which it was possible to observe the returns for three years. In the 36 months following the merger, the portfolios showed a significant destruction of value