面向多应用混部的性能保障方法综述

数据中心的高投入和低资源利用率一直是云服务提供商关注的问题.面对这个难题,直接的解决方案是在同等资源上混合部署更多的应用以提高资源使用效率.然而,由于混部应用对共享资源的竞争导致了应用间的性能干扰,从而影响了应用的性能、服务质量(quality of service,QoS)和用户满意度,因此如何保障应用的性能已成为混部场景下的关键问题.着重从应用和集群特征分析(基础)、干扰检测(前提)、单节点资源分配(微观层面策略)和集群作业调度(宏观层面策略)4个方面阐述多应用混部性能保障的相关背景、挑战和关键技术.在不同的混部场景下,由于应用和集群特征等不同,性能保障工作所面临的挑战和问题复杂度也各异,例如单位资源上混合部署的应用数量会直接影响到搜索资源空间的时间开销,应用的运行方式会影响到共享资源的竞争强度.因此,从问题复杂度角度出发,从应用和集群特征、资源干扰维度和混部应用个数3个维度对相关研究工作面临的挑战进行讨论和分析.探讨了面向高密度混部场景应用性能保障方法的发展方向和挑战,认为全栈式的软硬件协同方法是保障高密度混部下应用性能的趋势,该方法有助于全面地提升应用性能的可靠性和数据中心的资源利用率.

面向高密度混部的动态资源分配方法

当前的无服务计算提供商采用了一种灵活度低、固定CPU和内存分配比例的耦合式资源分配策略.随着更多类型应用被部署在无服务计算平台中,该策略已无法满足函数应用的多样化资源需求.由于函数应用的资源分配粒度小、部署密度高,若将CPU与内存资源的分配进行解耦,需解决资源配置空间爆炸问题.提出Semi-Share,一个面向无服务计算的解耦式资源管理系统,为函数寻找最优资源配置的同时降低混部函数之间的干扰.为解决资源配置空间爆炸问题,Semi-Share构建了一个2层资源分配架构,将资源配置空间划分为多个子空间来降低问题复杂度.第1层是函数分组,基于函数的资源使用特征和历史负载信息进行函数分组,根据分组将资源配置空间划分为多个子空间.第2层是资源分配,利用贝叶斯优化和加权打分函数来指导模型在资源配置空间中朝正确的方向搜索,降低时间开销.实验结果显示,Semi-Share相较于被广泛使用的梯度下降搜索法降低了平均85.77%的时间开销,并为函数带来平均42.72%的性能提升;与同样使用贝叶斯优化的耦合式资源分配系统COSE相比,Semi-Share能带来平均32.25%的性能提升.

面向混部云失败批处理作业的预测算法

为了降低混部云失败批处理作业的风险,使用K-means聚类算法将批处理作业分为四类,在分类的基础上提出了二层嵌套分类模型(two-layer nested classification model,TLNM),实现了基于TLNM的预测算法。基于Ali Trace 2018数据集上的实验结果表明,该算法的接受者操作特性(receiver operating characteristic,ROC)曲线明显优于其他常用分类器,ROC曲线下面积(即AUC)可以达到0.978,表明该算法具有良好的分类性能。同时召回率可以达到0.951,通过混淆矩阵可以看出TLNM算法能够准确预测出执行失败的批处理作业。

混部集群资源利用分析

现代互联网数据中心的规模随应用服务需求的增长而越来越大,但数据中心资源利用率低已逐步成为云计算进一步发展的制约因素.为了提高数据中心的资源利用率,云服务提供商将在线服务和离线批处理作业混合部署到同一个生产集群中.但混合部署增加了数据中心管理系统复杂性,对数据中心系统调度和工作负载分配提出了新的挑战.本文从资源使用角度出发,统计分析了阿里巴巴最新发布的包含4034台机器长达8天的混部集群日志数据集,刻画了机器对离线批处理任务与在线服务容器资源分配策略,以及离线批处理作业与在线服务之间的相互干扰.并根据不同的负载特征,以多种方式对机器进行分类,研究机器分工对集群效率提升的意义.通过分析阿里巴巴集群日志数据集,我们发现:(1)集群中43.271%的机器存在容器对CPU核心“超订”现象,而内存不存在“超订”现象;(2)集群中存在“备用节点”,确保集群出现故障时,任务能及时被转移到“备用节点”继续执行;(3)延迟敏感的在线任务的CPU利用率较低,但对内存资源的需求比较高,而离线批处理作业的CPU利用率较高,在线任务和离线任务的资源占用互补;(4)混合部署显著提高了CPU利用率,而内存可能是限制集群性能的主要因素;(5)集群中容器分布存在不平衡性;(6)离线任务的混合部署导致容器内存利用率有所下降,且当在线服务资源需求激增时,调度器缺少一定的容错性和健壮性;(7)离线任务如何部署与容器当前性能紧密相关,尤其是容器CPI指标,与离线任务部署呈现显著相关性.本文对集群负载特性、资源使用及离线和在线任务相互干扰进行研究,有助于其他研究人员对集群系统调度和负载分配分析优化,以提高现代数据中心的资源利用率.

混部数据中心负载特征及其任务调度优化分析

随着现代互联网数据中心的规模越来越大,数据中心面临着能耗、可靠性、可管理性与可扩展性等方面的挑战。同时,数据中心承载的服务多样,既有在线Web服务,也有离线批处理任务。在线任务要求较低的延迟,而离线任务要求较高的吞吐量。为了提高服务器利用率,降低数据中心能耗,当前数据中心往往将在线任务和离线任务混合部署到同一个计算集群中。在混部场景下,如何同时满足在线和离线任务的不同要求,是目前面临的关键挑战。分析了阿里巴巴于2018年发布的含有4034台服务器的混部计算集群在8天内的日志数据(cluster-trace-v2018),从静态配置信息、动态混部运行状态、离线批处理作业DAG依赖结构等出发,揭示其负载特征,包括任务倾斜与容器部署的相关关系等,根据任务依赖关系与关键路径,提出了相应的任务调度优化策略。

混部负载场景下离线负载资源调度策略研究

混部负载是当前业界提高数据资源利用率的重要手段,其原理是将在线负载和离线负载共同放置于同一数据中心、共享资源,在保证在线负载服务质量的前提下,将空闲资源分配给离线负载。当前针对混部负载中离线负载的资源调度采用传统的公平或者短作业优先等策略,并未考虑在线负载资源需求波动对离线负载运行的影响。为了达到进一步提升资源利用率和作业吞吐率的目的,提出基于负载完成时间预判的模拟退火资源分配策略。结果表明,该策略比公平策略和短作业优先策略在平均资源利用率上分别提高了7.8%和15.5%,在吞吐率上分别提高了38.2%和29.1%。

论宋代部色体系

"部",即乐部制度,"色",即"执色"体系。宋代的部色体系经历了"教坊四部"、"四部合一与执色的设置"以及"教坊十三色"三个阶段。本文详细考证这一历史进程中宋代部色体系的名实、功能、管理机制以及演变过程,并将部色体系与宋代艺术实践相联系,以探讨宋代部、色体系的功能特征及其转化的深层原因。

从明清郴州桂阳籍诗人用韵考郴州官话韵母特点

搜集并整理明清时期郴州桂阳籍诗人的诗歌韵文,分析古体诗以及赋的韵部混押情况,结合明代韵书《字学元元》俗音,对比《湖南方言调查报告》中1935年记录的郴州各县方音,考证出明清时期郴州官话韵母特点。

A prediction comparison between univariate and multivariate chaotic time series

The methods to determine time delays and embedding dimensions in the phase space delay reconstruction of multivariate chaotic time series are proposed. Three nonlinear prediction methods of multivariate chaotic time series including local mean prediction, local linear prediction and BP neural networks prediction are considered. The simulation results obtained by the Lorenz system show that no matter what nonlinear prediction method is used, the prediction error of multivariate chaotic time series is much smaller than the prediction error of univariate time series, even if half of the data of univariate time series are used in multivariate time series. The results also verify that methods to determine the time delays and the embedding dimensions are correct from the view of minimizing the prediction error.

Currents and mixing in the northern South China Sea

We investigated the vertical distribution of current velocity data of the entire water column at a site on the continental shelf of the northern South China Sea(SCS) from August 4 to September 6,2007,and found that the characteristics of barotropic and baroclinic tides are mainly diurnal.During the observation period,we also estimated the mixing before and after the passage of Typhoon Pabuk.We found that the internal-wave-scale dissipation rate,the turbulent dissipation rate,and the mixing rate in every water layer increased by about an order of magnitude after the typhoon passage.We analyzed a case of abrupt strong current and calculated the mixing rate before,during,and after the typhoon event.The results show that the internal-wave-scale dissipation rate and the mixing rate in every water layer increased by about two orders of magnitude during the event,while the turbulent dissipation rate increased by about an order of magnitude.Passage of the abrupt strong current could also have increased the mixing rate of affected seawater by more than an order of magnitude.However,the passage of the typhoon differed in that there was an increase in mixing only in the lower layer where the abrupt strong current was particularly strong.The variation of the mixing rate may help us to understand the effects of typhoons and abrupt strong currents on the mixing of seawater.

Experimental and theoretical analysis on the natural frequency of external prestressed concrete beams

Analyses and computations of the natural frequencies of external prestressed concrete structures are the bases for studying the dynamic properties of such structures. We carded out dynamic tests on three types of external simply supported beams, pulling the beams gradually. Then vertical loads were applied to the beams while the frequencies and related coefficients were measured at every step. We calculated natural frequencies and the results indicate that natural frequencies increase as the prestressing force rises in the pre-crack stage, and decrease significantly during the post-crack stage. Substantial incoincidences exist between the calculated and experimental results for the frequency and its tendency to changel Based on the experimental results, we modified the stiffness and other parameters in the equations. The results calculated using the modified equations agree with experimental results well, so the modified eauations can be used nractically.

Research of Underwater Bottom Object and Reverberation in Feature Space

The critical technical problem of underwater bottom object detection is founding a stable feature space for echo signals classification. The past literatures more focus on the characteristics of object echoes in feature space and reverberation is only treated as interference. In this paper, reverberation is considered as a kind of signal with steady characteristic, and the clustering of reverberation in frequency discrete wavelet transform （FDWT） feature space is studied. In order to extract the identifying information of echo signals, feature compression and cluster analysis are adopted in this paper, and the criterion of separability between object echoes and reverberation is given. The experimental data processing results show that reverberation has steady pattern in FDWT feature space which differs from that of object echoes. It is proven that there is separability between reverberation and object echoes.

Local Tensile Strength of Connection Between H-Steel Beam Flange and Concrete-Filled Circular Column Tube with Through Diaphragm

This paper focused on investigating local tensile strength of connection between steel beam flange and concrete-filled circular column tube with through diaphragm. Three specimens were designed and tested to failure, and the structure behavior was studied by experiment and FEM analysis. On the basis of the results obtained, an estimation for local plastic and ultimate strengths of the connections using yield line theory was attempted, which results in a good prediction.

Mechanics of flow and mixing at antroduodenal junction

The morphology of tissue structures composing the pyloric orifice is thought to play a role in effectively mixing aqueous gastric effluent with duodenal secretions. To understand the physical mechanisms leading to efficient digestion requires computational models that allow for analyses of the contributions of individual structural components. Thus, we have simulated 2-D channel flows through representative models of the duodenum with moving boundary capabilities in order to quantitatively assess the importance of notable features. A well-tested flow solver was used to computationally isolate and compare geometric and kinematic parameters that lead to various characteristics of fluid motion at the antroduodenal junction. Scalar variance measurement was incorporated to quantify the mixing effectiveness of each component. It was found that the asymmetric geometry of the pyloric orifice in concert with intermittent gastric outflow and luminal constriction is likely to enhance homogenization of gastric effluent with duodenal secretions.

Probabilistic Teleportation of a Three-Particle GHZ State via Two Three-Particle Entangled W States

A scheme for teleporting an unknown three-particle GHZ state from a sender to either one of two receivers is proposed. In this scheme, the quantum channel is composed of two non-maximally three-particle entangled W states. An unknown three-particle GHZ state can be perfectly teleported probabilistically if the sender performs two generalized Bell-state measurements and the Hadamard operation while either one of two receivers introduces an ancillary particle which is one of the final three particle constituting the teleported state, then performs the controlled-not operation with the ancillary particle as the target bit and introduces an appropriate unitary transformation with the help of the other receiver＇s simple measurements. All kinds of unitary transformations are given in detail. The present scheme may be directly generalized to teleport an unknown multiparticle GHZ state via two three-particle entangled W states used as the quantum channel.

Co-pyrolysis characteristics of typical components of waste plastics in a falling film pyrolysis reactor

Waste plastics mainly come from MSW and usually exist in the form of mixed plastics. During the co-pyrolysis process of mixed plastics, various plastic components have different physicochemical properties and reaction mechanisms. Considering the high viscosity and low thermal conductivity of molten plastics, a falling film pyrolysis reactor was selected to explore the rapid co-pyrolysis process of typical plastic components(PP, PE and PS).The oil and gas yields and the compositions of pyrolysis products of the three components under different ratios at pyrolysis temperatures were analyzed to explore the co-pyrolysis characteristics of PP, PE, and PS. The study is of great significance to the recycling of waste plastics.

Effects of rotational speed and fill level on particle mixing in a stirred tank with different impellers

The particle mixing was studied in a cylindrical stirred tank with elliptical dished bottom by experiments and simulations.The impeller types used were double helical ribbon(HR) + bottom HR,pitched blade ribbon + bottom HR,inner and outer HR + bottom HR,and pitched blade ribbon + Pfaudler + bottom HR labeled as impellers Ⅰ to Ⅳ,respectively.The quantitative correlations among the rotational speed,fill level and power consumption for impeller Ⅰ and impeller Ⅱ were obtained by experiments to validate the discrete element method(DEM) simulations.The particle mixing at different operating conditions was simulated via DEM simulations to calculate the mixing index using the Lacey method,which is a statistical method to provide a mathematical understanding of the mixing state in a binary mixture.The simulation results reveal that as the rotational speed increases,the final mixing index increases,and as the fill level increases,the final mixing index decreases.At the same operating conditions,impeller Ⅲ is the optimal combination,which provides the highest mixing index at the same revolutions.

Effect of Social Capital on Performance of Smallholder Farmer Organizations in Western Kenya

Development literature has recently promoted the use of producer organizations in linking farmers to better-paying commodity markets. However, empirical studies find mixed performance of such organizations. This study examines the producer organizations＇ internal factors that may explain the differences in the performance of producer organizations. The study specifically analyzes the role of social capital, within a producer organization, on the performance of such organization using quantitative techniques. The level of commercialization is used as proxy of organization＇s performance. The study finds that social capital positively affects the performance of producer organizations. These findings imply that development strategies that target the promotion commercialization of smallholder agriculture through producer organizations should pay attention to the internal factors within such organizations.

A Comparative Study of the Performance of Symmetric and Asymmetric Mixed-conducting Membranes

According to the configuration,mixed-conducting membranes are classified as symmetric membranes and asymmetric membranes consisting of a thin dense layer and a porous support.In this study,these two kinds of SrCo0.4Fe0.5Zr0.1O3-δ oxide-based membranes were systematically compared in terms of oxygen permeability and chemical stability,and their differences were elucidated by means of the theoretical calculation.For the oxygen permeability,the asymmetric membrane was greater than the symmetric membrane due to the significant decrease of bulk diffusion resistance in the thin dense layer of the asymmetric membrane.In regard to the chemical stability,the increase of oxygen partial pressure on the asymmetric membrane surface at CH4 side produced the stable time of over 1032h in partial oxidation of methane at 1123K,while the symmetric membrane was only of 528h.This study demonstrated that the asymmetric membrane was a promising geometrical configuration for the practical application.

A Hybrid Algorithm Based on Differential Evolution and Group Search Optimization and Its Application on Ethylene Cracking Furnace

To find the optimal operational condition when the properties of feedstock changes in the cracking furnace online,a hybrid algorithm named differential evolution group search optimization(DEGSO) is proposed,which is based on the differential evolution(DE) and the group search optimization(GSO).The DEGSO combines the advantages of the two algorithms:the high computing speed of DE and the good performance of the GSO for preventing the best particle from converging to local optimum.A cooperative method is also proposed for switching between these two algorithms.If the fitness value of one algorithm keeps invariant in several generations and less than the preset threshold,it is considered to fall into the local optimization and the other algorithm is chosen.Experiments on benchmark functions show that the hybrid algorithm outperforms GSO in accuracy,global searching ability and efficiency.The optimization of ethylene and propylene yields is illustrated as a case by DEGSO.After optimization,the yield of ethylene and propylene is increased remarkably,which provides the proper operational condition of the ethylene cracking furnace.