• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    一種面向云計算非比例資源消耗特性的虛擬機(jī)放置算法

    2019-09-10 07:22:44羅香玉辛剛桂小林
    關(guān)鍵詞:云計算

    羅香玉 辛剛 桂小林

    摘 要:虛擬機(jī)放置是云計算中的一個基本問題。 通過將多臺虛擬機(jī)聚集在單臺物理機(jī)上,云計算可極大降低系統(tǒng)的資源以及能源消耗。虛擬機(jī)放置的目標(biāo)之一是開啟最少數(shù)量的物理機(jī)來滿足所有虛擬機(jī)的資源需求。一個重要的挑戰(zhàn)在于各虛擬機(jī)對不同類型資源消耗的比例往往與物理機(jī)所配備的各類資源的比例并不相同。一旦物理機(jī)上某類資源消耗殆盡,該物理機(jī)上其它類型的資源將無法得到利用,隨之導(dǎo)致所開啟物理機(jī)總數(shù)以及能耗成本增加。文中借助多種不同配置的物理機(jī)來解決上述問題。首先,虛擬機(jī)被劃分為若干子集合。劃分的約束條件是保證各個虛擬機(jī)子集合作為一個整體所消耗的各類資源恰與某一類型物理機(jī)各類資源的配置成比例。然后,利用同構(gòu)環(huán)境的虛擬機(jī)放置算法,完成各虛擬機(jī)子集合在相應(yīng)類型物理機(jī)上的放置。實驗結(jié)果表明,文中算法能夠協(xié)調(diào)各類資源的使用,從而有效減少物理機(jī)使用總量,降低能耗成本13.0%~57.6%.

    關(guān)鍵詞:虛擬機(jī)放置;非比例資源消耗;云計算;能耗節(jié)約;多維度資源

    中圖分類號:TP 393 ? ? ? ? ? 文獻(xiàn)標(biāo)志碼:A

    文章編號:1672-9315(2019)05-0889-09

    Abstract:Virtual machine placement is a basic problem in cloud computing.By consolidating several virtual machines onto one single physical machine,cloud computing reduces both resource costs and energy consumption.One of the optimization objectives of virtual machine placement is to use a minimum number of physical machines to accommodate all the virtual machines requested by customers.The challenge lies in that the multi-dimensional resources required by a virtual machine are typically not proportional to that provided by a physical machine.Once a single dimension of resource is exhausted in a physical machine,the rest of all the other dimensions of resources will stay unutilized,leading to a great amount of resource waste.This paper proposes a new virtual machine placement algorithm that mixes multiple kinds of physical machines to tackle the problem of unsufficientresource utilization.Firstly,virtual machines are divided into several subsets.For each subset as a whole,different dimensions of resources requested are approximately proportional to that provided by a kind of physical machine.Secondly,virtual machines in each subset are separately placed onto the corresponding kind of physical machines.Experimental results show that the proposed algorithm coordinates the utilization of different types of resources and achieves power reduction ranging from 13.0% to 57.6%.

    Key words:virtual machine placement;non-proportional resource consumption;cloud computing;power reduction;multidimensional resources

    0 INTRODUCTION

    Nowadays cloud computing is a popular way of offering computation services.It enables customers to enjoy computation services as conveniently as they enjoy electricity and water[1].Customers are typically served by virtual machines.By consolidating several virtual machines onto one single physical machine,the whole cloud computing system achieves both more efficient utilization of resources and lower power consumption.

    A cloud computing system is a large-scale distributed system composed of thousands of or even more physical machines[2].A basic problem is how to place the virtual machines such that the number of the consumed physical machines can be minimized while satisfying multiple resource constraints.

    To accommodate several virtual machines,a physical machine is required to provide available CPU,memory and other resources no less than that requested by the virtual machines.Once a single dimension of resource is exhausted in a physical machine,all the rest of the other dimensions of resources on it will stay unutilized[3].Therefore,an ideal virtual machine placement algorithm should assign each physical machine with several virtual machines that just use up each dimension of resource on it.

    However,the challenge lies in that different dimensions of resources required by a virtual machine are typically not proportional to that provided by a physical machine.Existing studies mainly aim to make the best-effort optimization of resource utilization under the assumption that the candidate physical machines are deterministic and unchangeable.They rarely investigate how to adaptively adjust the candidate physical machines’ configuration according to the virtual machines’ requirements to make further improvements.In the situations that the virtual machines differ significantly with the physical machines in the proportions of the multiple dimensions of resources,the best-effort optimization is usually unsatisfactory.

    Our contributions mainly include the following three aspects.Firstly,we redefine the virtual machine placement problem and divide it into two sub-problems.One is how to adaptively adjust the configuration of the physical machines according to the requirements of the virtual machines.The other is how to map the virtual machines to the candidate physical machines.Secondly,we propose an algorithm,namely CORE(COordinating multiple REsources),to resolve the problem.Finally,extensive experiments have been conducted to evaluate the efficiency of the algorithm.The results show that it can achieve power reduction ranging from 13.0% to 57.6%.

    The paper is organized as follows.Section 1 summarizes the related work.Section 2 explains the motivations of our work.Section 3 redefines the virtual machine placement problem.Section 4 elaborates the CORE algorithm.Section 5 gives the experimental results and Section 6 concludes the whole paper.

    1 RELATED WORK

    The virtual machine placement algorithm greatly affects the performance and the efficiency of clouds and attracts many researchers’ attention[4].

    According to the assumption for the resources,existing virtual machine placement algorithms could be classified into single dimensional resource oriented and multi-dimensional resource oriented algorithms.F.Pan et al.proposed a single resource oriented algorithm that only considers the CPU resource[5].L.Chen et al.proposed a multi-dimensional resource oriented algorithm RIAL that considers CPU,memory and the network resources[6].It assigns different weights to different resources and calculates the resource intensity of each physical machine based on the weights.By this way,the multi-dimensional problem is transformed into a single dimensional one.R.Li et al.proposed a true multidimensional solution which aims to keep balanced usage of each dimensional resource[7].However,the physical machines were assumed to be determined in advance and only best-effort utilization was provided.Besides,the relationships among the resource utilization,the characteristics of virtual machine requirements and the configuration of the physical machines were not investigated.

    According to the assumption for the characteristic of the workload,existing virtual machine placement algorithms can be classified into static and dynamic ones.Static algorithms assume that the workload of each virtual machine is constant.Many approximation algorithms for bin packing can be used for static virtual machine placement[8].Besides,genetic or other intelligent optimization algorithms can also work[9].Dynamic algorithms[10-11]migrate virtual machines from one physical machine to another as the workload changes.

    From the point of view of the optimization objectives,existing virtual machine placement algorithms can be classified into single-objective oriented and multi-objective oriented ones.The optimization objects include minimizing SLA violation,maximizing resource utilization,minimizing energy consumption,etc[12-13].W.Wang et al.proposed a single objective oriented algorithm that aims for energy minimization[14].J.Xu and J.Fortes proposed an algorithm that simultaneously minimizes resource wastage,power consumption and thermal dissipation costs[15].H.Zhao proposed an algorithm that ensures both low power consumption and high performance guarantee[16].

    Besides,the state-of-art research also addresses the virtual machine placement problem in edge cloud systems[17-19].

    However,to the best of our knowledge,existing literatures mainly concentrate on virtual machine placement optimization under the assumption that the candidate physical machines are determined in some artificial way.There is no solution that automatically changes the candidate physical machines according to the virtual machines’ resource requirements.

    2 MOTIVATIONS

    For economic and environmental reasons,cloud computing aims to employ a minimum number of physical machines to accommodate the virtual machines requested by customers,reducing both resource costs and power consumption.For a given set of virtual machines,both the resource costs and the power consumption are affected by not only the virtual machine placement algorithm itself,but also the configuration of the physical machines.

    Suppose that there are 500 virtual machines.For half of them,each one requires 3 cores CPU and 0.5 GB RAM.For the other half,each one requires 1 core CPU and 1.5 GB RAM.Therefore,the total amount of CPU required by the virtual machines equals 1 000 cores,and the total amount of required RAM equals 500 GB.With the physical machines configured with 4 cores CPU and 8 GB RAM,the least number of the powered-on physical machines cannot be smaller than 250 and the utilization efficiency of RAM cannot be greater than 25%,whatever virtual machine placement algorithm is adopted.However,with the physical machines configured with 4 GB RAM and 8 cores CPU,only 125 physical machines need to be powered on and both the two types of resources can be sufficiently utilized.

    For the above scenario,changing the physical machine’s configuration leads to more efficient resource utilization and lower power consumption.However,it is not always practical to do so in reality.Virtual machines are created and removed dynamically,and each of them may have different resource requirements.Hence the optimal physical machine configuration changes frequently.It is not practical to alter the physical machine’s configuration all the time.

    Therefore,we mix several kinds of physical machines to mimic physical machines with arbitrary kind of configuration.We assume that the cloud providers purchase several kinds of physical machines with different resource configurations and the number of each kind of physical machines is large enough.For a given set of virtual machines,the placement algorithm automatically adjusts the number of each kind of physical machines to power on,ensuring that the multiple resources provided by the whole powered-on physical machines always proportional to that requested by the whole virtual machines.For cloud providers,the most important thing is not to purchase but to power on the least number of physical machines,because the budget for power consumption is much higher than the infrastructure costs.

    In a word,for a given set of virtual machines,the efficiency of resource utilization is greatly affected by the physical machines’ configuration.With unsuitable physical machine configuration,resource waste is inevitable,whatever placement algorithm is adopted.We aim to mix several kinds of physical machines to keep the mimic configuration always suitable to the virtual machines,so that the placement results can be improved compared with those obtained with a deterministic configuration.

    3 PROBLEM STATEMENT

    Suppose that there are K kinds of physical machines with different resource configurations.The number of each kind of physical machine is large enough for accommodating an arbitrary set of virtual machines.The i th kind of physical machine is represented with a vector pi composed of C elements with each element pij corresponding to the amount of the j th type of resource provided by the i th kind of physical machine.There are N virtual machines to be placed onto the physical machines.Each virtual machine is expressed with a vector vi′ also composed of C elements,with each element vi′j representing the amount of the j th type of resource requested by the i′ th virtual machine.Here 1≤i≤K,1≤i′≤N,1≤j≤C,K,N and C are known numbers.

    The virtual machine placement problem is divided into two sub-problems.Firstly,for accommodating a given set of virtual machines,how many each kind of physical machines should be powered on? Secondly,how to map the virtual machines to the powered-on physical machines? The optimization objective is to minimize the total number of the powered-on physical machines while satisfying each virtual machine’s resource requirements.

    4 THE CORE ALGORITHM

    4.1 Basic idea

    Let ni denote the number of the powered-on physical machines with the i th kind of configuration and VSi denote the set of the virtual machines mapped to the i th kind of physical machines.Hence {VSi|1 ≤i≤K} defines a partitioning on the whole set of the virtual machines.

    Once we obtain a proper partitioning of the virtual machines,through placing the virtual machines in each subset VSi to the ith kind of physical machines,ni can be figured out.Therefore,the partitioning of the virtual machines is at the heart of the problem.

    The CORE algorithm works in three steps.Firstly,it partitions the whole virtual machine set into K subsets ensuring that each subset VSi consumes the multiple resources proportionally to the provisioning of the ith kind of physical machines.Secondly,it separately calculates the mappings between the virtual machines in each subset VSi and the ith kind of physical machines.Thirdly,it merges the results of the second step,the number of each kind of physical machines to be powered on(i.e.,ni)is obtained,and the mappings between the whole virtual machines and the whole powered-on physical machines are also figured out.

    4.2 Elaboration of the algorithm

    The CORE algorithm is depicted in Algorithm 1.In the algorithm,there are two inputs PS and VS.PS is the set of physical machine types with each element pi representing the configuration of the i th type of physical machine,and VS is the set of virtual machines with each element vi′ representing the resource requirements of the vi′ th virtual machine.Both pi and vi′are vectors composed of C elements.The j th element of pi denoted by pij represents the amount of the j th type of resource provided by the i th kind of a single physical machine,and the j th element of vi′ denoted by vi,j represents the amount of the j th type of resource requested by the i′th virtual machine.There are two outputs NS and MS.NS is a set of numbers with each element ni representing the number of the powered-on physical machines with the i th kind of configuration.MS is a set of tuples with each element indicating that the i′ th virtual machine is placed on the s(i′) th physical machine with the k(i′) th kind of configuration.Here k(i′) is a integer between 1 and K,and s(i′) is an integer between 1 and nk(i′).Recall that nk(i′) represents the number of the powered-on physical machines with the k(i′) th kind of configuration.

    Algorithm 1:The CORE algorithm

    The main body of the algorithm calls three sub-algorithms as shown in Algorithm 1.The partitioning sub-algorithm is responsible for partitioning the virtual machine set VS into K subsets,the placement sub-algorithm is responsible for placing the virtual machines in a subset VSi to the i th kind of physical machines,and the merging sub-algorithm is responsible for combining the placement results to obtain the number of the powered-on physical machine for each kind of configuration,and the mappings between the whole virtual machines and the whole powered-on physical machines.In the algorithm description,π represents the partition on VS and π={VSi|1 ≤ i ≤ K}.MSi represents the mapping of the virtual machines in subset VSi and the i th kind of physical machines.

    4.2.1 The partitioning sub-algorithm

    As discussed in the subsection 4.1,the partitioning of the virtual machines is at the heart of the problem.For the virtual machines as a whole,the multiple resources requested may be not proportional to any kind of the existing physical machines.With proper partitioning,the multiple resources requested by each virtual machine subset could be proportional to a certain kind of physical machines.

    Let Rij denote the amount of the j th type of resource requested by the virtual machines of the subset VSi.VSi satisfies proportional resource consumption means that for each j between 1 and C,Rij/pij is almost the same.We define θ to measure the degree of proportionality,with θ=min{Rij/pij|1≤j≤C}/max{Rij/pij|1≤j≤C}.θ is a number between 0 and 1.A larger θ means a higher degree of proportionality.It should be noted that max{Rij/pij|1≤j≤C} is a lower bound of the number of the ith kind of physical machines to power on.

    The partitioning sub-algorithm aims to guarantee that each virtual machine subset VSi satisfies proportional resource consumption,and the lower bound of the whole powered-on physical machines,i.e.,1≤i≤K(max{Rij/pij|1≤j≤C}),is minimized.The partitioning sub-algorithm adopts a greedy strategy described in Algorithm 2.For each virtual machine,the algorithm assigns it to the subset that makes the least increase of 1≤i≤K(max{Rij/pij|1≤j≤C}).

    4.2.2 The placement sub-algorithm

    The placement sub-algorithm is responsible for placing the virtual machines in a subset VSi to the ith kind of physical machines.Although other placement strategies also work,we adopt the FirstFit strategy for simplicity.Since the partitioning sub-algorithm ensures that the multiple resources requested by the virtual machine subset VSi are approximately proportional to that provided by the ith kind of physical machines,even if the FirstFit strategy is qualified to generate ideal placement results.Experimental results also validate the conjecture,as exposed in Section 5.The placement sub-algorithm is described in Algorithm 3.In the description, means that the k th virtual machine in VSi is placed onto the s(k)th physical machine with the i th kind of configuration satisfying 1≤k≤|VSi| and 1≤s(k)≤ ni.

    The results indicate that,the 2nd configuration can lead to high utilization of the two types of resources as well as low power consumption.Moreover,the number of the consumed physical machines(.i.e.,266)is very near to nmin(i.e.,253).Therefore,as long as the configuration of the physical machines is suitable to the virtual machines,even if the First Fit algorithm can generate near-optimal placement results.More complex placement algorithms are not very necessary.

    The results also indicate that,with unsuitable configuration,at least one type of resource generates a great amount of waste.Meanwhile,there is little room for further optimization by improving placement strategy,since the number of physical machines consumed already approaches the lower bound nmin.With the 1st configuration,the number of the consumed physical machines equals 507 while nmin equals 496.With the 3rd configuration,the number of the consumed physical machines equals 530 while nmin equals 506.Whatever virtual machine placement algorithm is adopted,the reduction of the number of the consumed physical machines cannot be greater than 4.5%.

    In summary,compared to the placement strategy,the configuration of the physical machine really matters.The most important thing is to make the configuration always suitable to the virtual machines.Once the configuration is determined,there is little room for further optimization whatever placement strategy is adopted.Meanwhile,the requirements of the virtual machines are dynamic,and any single kind of physical machine configuration can not always meet the virtual machines’ requirements.Therefore,the CORE algorithm adaptively adjusts the number of each kind of physical machines powered on to make the different kinds of physical machines as a whole always suitable to the requirements of the virtual machines.

    5.3 Evaluation of CORE’s performance

    In the experiment,we use the synthetic virtual machine generator to obtain 1 000 virtual machines’ resource requirements.The parameters are set as follows:a1=1,b1=7,a2=1 and b2=19.In the experiment,the total amount of CPU requirement equals 4 046 cores,and the total amount of RAM requirement equals 9 887 GB.

    We adopt the First Fit algorithm to place the virtual machines onto the physical machines with each kind of configuration respectively and adopt the CORE algorithm to place them onto the three kinds of physical machines as well.Besides,we calculate the optimal value of the power consumption and the resource utilization for the three configurations.The results are shown in TABLE Ⅳ.

    The results show that,through properly mixing the three kinds of physical machines,the CORE algorithm reduces the power consumption and improves the resource utilization efficiency.The power consumption reduction ranges from 13.0% to 57.6%.

    For further analysis,we calculate the degree of proportionality θ.We have θ(1)=0.15,θ(2)=0.61 and θ(3)=0.41 respectively with the three different configurations.It means that the second configuration fits the requirements of the virtual machines best,while the first configuration is the worst.Therefore,the situation where only the physical machines with the first configuration are adopted leads to the maximum power consumption,and the CORE reduces the power consumption with the highest ratio(i.e.,57.6%).

    6 CONCLUSION

    The efficiency of virtual machine placement is affected by not only the placement strategy itself but also the configuration of the physical machines.The improper configuration leads to a waste of both resource costs and energy consumption.This paper proposes the idea of mixing several kinds of physical machines to mimic a new kind of configuration that properly fits the resource requirements of the virtual machines.Furthermore,we devise the algorithm CORE to realize the idea.It divides the virtual machines into several subsets,with each subset satisfying proportional resource consumption to a certain kind of physical machines.Experimental studies show that the algorithm achieves 13.0% to 57.6% energy reduction compared with those adopt any single kind of physical machines.

    REFERENCES:

    [1] Zhao L,Lu L,Jin Z,et al.Online virtual machine placement for increasing cloud provider's revenue[J].IEEE Transactions on Services Computing,2017,10(2):273-285.

    [2]Chaisiri S,Lee B,Niyato D.Optimization of resource provisioning cost in cloud computing[J].IEEE Transactions on Services Computing,2012,5(2):164-177.

    [3]Zhang J,He Z,Huang H,et al.SLA aware cost efficient virtual machines placement in cloud computing[C]//Proceedings of IEEE International Conference on Performance,Computing and Communications,2014:1-8.

    [4]Masdari M,Nabavi S,Ahmadi V.An overview of virtual machine placement schemes in cloud computing[J].Journal of Network & Computer Applications,2016,66(C):106-127.

    [5]Pan F,Jiang C,Xu X,et al.Placement strategy of virtual machines based on workload characteristics[J].Journal of Chinese Computer Systems,2013,34(3):520-524.

    [6]Chen L,Shen H,Sapra K.RIAL:resource intensity aware load balancing in clouds[C]//Proceedings of IEEE Conference on Computer Communications(INFOCOM),2014:1294-1302.

    [7]Li R,Zheng Q,Li X,et al.A novel multi-objective optimization scheme for rebalancing virtual machine placement[C]//Proceedings of IEEE International Conference on Cloud Computing,2016:710-717.

    [8]Bansal N,Caprara A,Sviridenko M.Improved approximation algorithms for multidimensional bin packing problems[C]//Proceedings of IEEE Symposium on Foundations of Computer Science,2006:697-708.

    [9]Kaaouache M,Bouamama S.Solving bin packing problem with a hybrid genetic algorithm for VM placement in cloud[J].Procedia Computer Science,2015,60(1):1061-1069.

    [10]Zhang M,Ren H,Xia C.A dynamic placement policy of virtual machine based on MOGA in cloud environment[C]//Proceedings of IEEE ISPA/IUCC,2017:885-891.

    [11]Hyser C,Mckee B,Gardner R,et al.Autonomic virtual machine placement in the data center[R].HP Labs Technical Report,HPL-2007-189,2007.

    [12]Gaggero M,Caviglione L.Model predictive control for energy-efficient,quality-aware,and secure virtual machine placement[J].IEEE Transactions on Automation Science and Engineering,2019,16(1):420-432.

    [13]Guerrero C,Lera I,Bermejo B,et al.Multi-objective optimization for virtual machine allocation and replica placement in virtualized hadoop[J].IEEE Transactions on Parallel and Distributed Systems,2018,29(11):2568-2581.

    [14]Wang W,Jiang Y,Wu W.Multiagent-based resource allocation for energy minimization in cloud computing systems[J].IEEE Transactions on Systems Man & Cybernetics Systems,2017,47(2):205-220.

    [15]Xu J,F(xiàn)ortes J.Multi-objective virtual machine placement in virtualized data center environments[C]//Proceedings of IEEE International Conference on Green Computing and Communications,2010:179-188.

    [16]Zhao H,Wang J,Liu F,et al.Power-aware and performance-guaranteed virtual machine placement in the cloud[J].IEEE Transactions on Parallel & Distributed Systems,2018,29(6):1385-1400.

    [17]Li K,Nabrzyski J.Networked virtual machine placement in edge cloud systems[C]//Proceedings of IEEE International Symposium on Parallel and Distributed Computing,2019:23-31.

    [18]Tziritas N,Koziri M,Bachtsevani A,et al.Data replication and virtual machine migrations to mitigate network overhead in edge computing systems[J].IEEE Transactions on Sustainable Computing,2017,2(4):320-332.

    [19]Tao Z,Xia Q,HAO Z,et al.A survey of virtual machine management in edge computing[J].Proceedings of the IEEE,2019,107(8):1482-1499.

    [20]Calheiros R,Ranjan R,Beloglazov A,et al.CloudSim:a toolkit for modeling and simulation of cloud computing environments and evaluation of resource provisioning algorithms[J].Software Practice & Experience,2011,41:23-50.

    猜你喜歡
    云計算
    云計算虛擬化技術(shù)在電信領(lǐng)域的應(yīng)用研究
    基于云計算的醫(yī)院信息系統(tǒng)數(shù)據(jù)安全技術(shù)的應(yīng)用探討
    談云計算與信息資源共享管理
    志愿服務(wù)與“互聯(lián)網(wǎng)+”結(jié)合模式探究
    云計算與虛擬化
    基于云計算的移動學(xué)習(xí)平臺的設(shè)計
    基于云計算環(huán)境下的ERP教學(xué)改革分析
    科技視界(2016年22期)2016-10-18 14:33:46
    基于MapReduce的故障診斷方法
    實驗云:理論教學(xué)與實驗教學(xué)深度融合的助推器
    云計算中的存儲虛擬化技術(shù)應(yīng)用
    科技視界(2016年20期)2016-09-29 13:34:06
    日韩人妻精品一区2区三区| 精品亚洲成a人片在线观看| 久久婷婷青草| 51午夜福利影视在线观看| 免费观看a级毛片全部| 国产精品 欧美亚洲| 丰满迷人的少妇在线观看| 女人爽到高潮嗷嗷叫在线视频| 国产精品久久久久久精品古装| 国产成人系列免费观看| 亚洲国产精品一区二区三区在线| 色综合欧美亚洲国产小说| 男男h啪啪无遮挡| 国产精品熟女久久久久浪| 黑人巨大精品欧美一区二区蜜桃| 19禁男女啪啪无遮挡网站| 一区二区三区精品91| 多毛熟女@视频| 丝袜脚勾引网站| tube8黄色片| 欧美最新免费一区二区三区| 欧美人与性动交α欧美精品济南到| 成人18禁高潮啪啪吃奶动态图| a 毛片基地| 亚洲欧美色中文字幕在线| 欧美日韩综合久久久久久| 操出白浆在线播放| 99久久99久久久精品蜜桃| 999久久久国产精品视频| 亚洲在久久综合| 欧美日韩成人在线一区二区| 两个人看的免费小视频| 亚洲精品国产区一区二| 19禁男女啪啪无遮挡网站| 国产熟女午夜一区二区三区| 校园人妻丝袜中文字幕| 免费黄网站久久成人精品| 亚洲国产日韩一区二区| 久久久精品国产亚洲av高清涩受| 人妻人人澡人人爽人人| 精品视频人人做人人爽| 亚洲一区二区三区欧美精品| 搡老乐熟女国产| 狂野欧美激情性bbbbbb| 国产成人一区二区在线| 亚洲一卡2卡3卡4卡5卡精品中文| 亚洲精品av麻豆狂野| 亚洲成人一二三区av| 午夜福利,免费看| 免费人妻精品一区二区三区视频| 综合色丁香网| 亚洲图色成人| 免费观看av网站的网址| 国产激情久久老熟女| 五月天丁香电影| 日本欧美视频一区| 这个男人来自地球电影免费观看 | 亚洲精品一二三| 日韩人妻精品一区2区三区| 男女边吃奶边做爰视频| 国产亚洲一区二区精品| 国产一区二区激情短视频 | 这个男人来自地球电影免费观看 | 亚洲成色77777| 99热国产这里只有精品6| 久久精品久久精品一区二区三区| 亚洲精品一区蜜桃| 中文字幕人妻丝袜制服| 少妇的丰满在线观看| 欧美人与性动交α欧美软件| 精品一品国产午夜福利视频| 国产精品偷伦视频观看了| 亚洲情色 制服丝袜| 一区二区三区激情视频| 久久鲁丝午夜福利片| 中文字幕人妻熟女乱码| 亚洲国产最新在线播放| 在线观看一区二区三区激情| 日韩av在线免费看完整版不卡| 男女边吃奶边做爰视频| 少妇 在线观看| 热99久久久久精品小说推荐| 亚洲天堂av无毛| 亚洲精品av麻豆狂野| 伊人久久大香线蕉亚洲五| 成人免费观看视频高清| 久久久国产一区二区| 亚洲欧美中文字幕日韩二区| 亚洲第一青青草原| 在线观看免费午夜福利视频| 亚洲综合精品二区| 久久毛片免费看一区二区三区| 亚洲一区二区三区欧美精品| 久久 成人 亚洲| www.自偷自拍.com| 天堂中文最新版在线下载| 国产成人91sexporn| 久久久国产一区二区| 久久久久精品性色| 国产深夜福利视频在线观看| 精品酒店卫生间| 亚洲av成人精品一二三区| 99久久综合免费| 精品亚洲成国产av| 一级,二级,三级黄色视频| 中文字幕最新亚洲高清| 国产欧美日韩一区二区三区在线| 欧美激情高清一区二区三区 | 尾随美女入室| 日本色播在线视频| 国语对白做爰xxxⅹ性视频网站| 欧美精品高潮呻吟av久久| 满18在线观看网站| 免费av中文字幕在线| 欧美成人午夜精品| 久久精品人人爽人人爽视色| 亚洲国产中文字幕在线视频| 中文字幕亚洲精品专区| 制服丝袜香蕉在线| 久久人人爽av亚洲精品天堂| 精品人妻一区二区三区麻豆| 麻豆av在线久日| 亚洲情色 制服丝袜| 久久久久久久大尺度免费视频| 成人免费观看视频高清| 夫妻午夜视频| 久久久久久久大尺度免费视频| 免费高清在线观看日韩| 中文精品一卡2卡3卡4更新| 超碰成人久久| 蜜桃国产av成人99| 别揉我奶头~嗯~啊~动态视频 | 久久久精品94久久精品| 色综合欧美亚洲国产小说| 国产男人的电影天堂91| 欧美久久黑人一区二区| 卡戴珊不雅视频在线播放| 亚洲自偷自拍图片 自拍| 天堂俺去俺来也www色官网| 亚洲国产毛片av蜜桃av| 久久久久精品人妻al黑| 亚洲第一区二区三区不卡| 日本av免费视频播放| 丰满乱子伦码专区| 久久精品亚洲熟妇少妇任你| 久久久久久人妻| 国产xxxxx性猛交| 亚洲人成77777在线视频| 国产一区亚洲一区在线观看| 亚洲精品国产区一区二| 色网站视频免费| 在线观看三级黄色| 亚洲精品国产av蜜桃| 欧美激情高清一区二区三区 | 大香蕉久久网| 可以免费在线观看a视频的电影网站 | 亚洲av日韩精品久久久久久密 | 精品第一国产精品| 涩涩av久久男人的天堂| 午夜日本视频在线| av国产久精品久网站免费入址| 搡老岳熟女国产| 亚洲av中文av极速乱| 啦啦啦中文免费视频观看日本| 高清在线视频一区二区三区| 91国产中文字幕| 亚洲欧美一区二区三区久久| 成人午夜精彩视频在线观看| 成人国产麻豆网| 久久精品亚洲熟妇少妇任你| 美女高潮到喷水免费观看| 成人国产av品久久久| 欧美精品一区二区大全| 91精品国产国语对白视频| 一个人免费看片子| 看非洲黑人一级黄片| 欧美 亚洲 国产 日韩一| 成年动漫av网址| 波多野结衣一区麻豆| 亚洲精品国产av成人精品| 考比视频在线观看| 99热网站在线观看| 亚洲精品在线美女| 亚洲精品aⅴ在线观看| 91老司机精品| 秋霞伦理黄片| 亚洲伊人久久精品综合| 一区二区三区激情视频| 人成视频在线观看免费观看| 热re99久久国产66热| 日韩 欧美 亚洲 中文字幕| 成人手机av| 欧美日韩国产mv在线观看视频| 久久婷婷青草| 十分钟在线观看高清视频www| 狂野欧美激情性bbbbbb| 久久人妻熟女aⅴ| 久久影院123| 亚洲国产精品国产精品| 免费观看a级毛片全部| 热99国产精品久久久久久7| 99re6热这里在线精品视频| 最新在线观看一区二区三区 | 老熟女久久久| 1024视频免费在线观看| 精品卡一卡二卡四卡免费| 秋霞伦理黄片| 爱豆传媒免费全集在线观看| 啦啦啦啦在线视频资源| 精品福利永久在线观看| av电影中文网址| 亚洲男人天堂网一区| 七月丁香在线播放| 欧美 日韩 精品 国产| 男女下面插进去视频免费观看| 日韩成人av中文字幕在线观看| 久久久国产欧美日韩av| 在线观看免费午夜福利视频| 亚洲熟女毛片儿| 香蕉丝袜av| 9色porny在线观看| 在线观看免费视频网站a站| 国产极品粉嫩免费观看在线| 麻豆精品久久久久久蜜桃| 欧美日韩一区二区视频在线观看视频在线| 久久韩国三级中文字幕| 一本一本久久a久久精品综合妖精| 中国三级夫妇交换| 国产一级毛片在线| 日韩av在线免费看完整版不卡| 老汉色∧v一级毛片| 亚洲人成网站在线观看播放| 国产精品一国产av| 日韩 亚洲 欧美在线| 日韩欧美一区视频在线观看| 天天躁夜夜躁狠狠久久av| 热re99久久国产66热| 蜜桃在线观看..| 亚洲av男天堂| 女人被躁到高潮嗷嗷叫费观| 久久精品亚洲av国产电影网| 久久久久精品性色| 精品免费久久久久久久清纯 | 九九爱精品视频在线观看| 老鸭窝网址在线观看| 久久韩国三级中文字幕| 国产探花极品一区二区| 亚洲国产精品一区三区| 在线观看免费视频网站a站| 亚洲国产av新网站| 亚洲国产欧美日韩在线播放| 亚洲精品久久久久久婷婷小说| 一区二区三区乱码不卡18| 中文字幕人妻丝袜一区二区 | 亚洲男人天堂网一区| 大片电影免费在线观看免费| 日日啪夜夜爽| 在线观看免费午夜福利视频| 999精品在线视频| 亚洲av欧美aⅴ国产| 卡戴珊不雅视频在线播放| 亚洲欧美成人精品一区二区| 免费久久久久久久精品成人欧美视频| 免费黄频网站在线观看国产| 国产一区二区在线观看av| 亚洲三区欧美一区| 国产精品 国内视频| 久久精品国产综合久久久| 欧美黄色片欧美黄色片| 国产精品一区二区在线不卡| 在线观看免费视频网站a站| 亚洲av男天堂| 亚洲国产精品国产精品| 一区二区av电影网| 晚上一个人看的免费电影| 国产精品 国内视频| 夜夜骑夜夜射夜夜干| 十八禁网站网址无遮挡| 综合色丁香网| 高清视频免费观看一区二区| 国产视频首页在线观看| 日韩一区二区视频免费看| 99久久精品国产亚洲精品| 多毛熟女@视频| 在线亚洲精品国产二区图片欧美| 亚洲人成电影观看| 亚洲欧美精品综合一区二区三区| 天天操日日干夜夜撸| 王馨瑶露胸无遮挡在线观看| 国产在线免费精品| 日本av手机在线免费观看| 一区二区日韩欧美中文字幕| 又粗又硬又长又爽又黄的视频| 精品亚洲乱码少妇综合久久| 视频区图区小说| 婷婷色av中文字幕| 只有这里有精品99| 久久久久国产精品人妻一区二区| 成人毛片60女人毛片免费| 亚洲婷婷狠狠爱综合网| 国产精品一区二区在线观看99| 久久综合国产亚洲精品| 国产不卡av网站在线观看| av片东京热男人的天堂| 极品少妇高潮喷水抽搐| 桃花免费在线播放| 国产乱人偷精品视频| 日韩伦理黄色片| 青春草国产在线视频| 亚洲精品aⅴ在线观看| 一本大道久久a久久精品| 欧美激情极品国产一区二区三区| 啦啦啦在线免费观看视频4| 人人妻人人爽人人添夜夜欢视频| 黑人欧美特级aaaaaa片| 国产午夜精品一二区理论片| 多毛熟女@视频| 美女中出高潮动态图| 免费不卡黄色视频| 精品人妻一区二区三区麻豆| av片东京热男人的天堂| 91成人精品电影| 视频区图区小说| 久久久久久人人人人人| 欧美人与性动交α欧美精品济南到| 建设人人有责人人尽责人人享有的| 天天操日日干夜夜撸| 国产精品熟女久久久久浪| 国产熟女欧美一区二区| 老司机深夜福利视频在线观看 | 亚洲在久久综合| 成人18禁高潮啪啪吃奶动态图| av视频免费观看在线观看| 亚洲国产欧美在线一区| 搡老岳熟女国产| 一本—道久久a久久精品蜜桃钙片| 国产爽快片一区二区三区| 99香蕉大伊视频| 老汉色∧v一级毛片| 街头女战士在线观看网站| 亚洲欧美激情在线| 亚洲精品久久久久久婷婷小说| 亚洲欧美一区二区三区国产| a级毛片在线看网站| 老司机在亚洲福利影院| 精品免费久久久久久久清纯 | 精品一区二区三卡| 又粗又硬又长又爽又黄的视频| 国产精品久久久av美女十八| 777米奇影视久久| 国产成人系列免费观看| av网站在线播放免费| 免费黄频网站在线观看国产| 亚洲av福利一区| 天天影视国产精品| 777米奇影视久久| 国产日韩欧美在线精品| 国产男女内射视频| 两个人免费观看高清视频| 国产精品欧美亚洲77777| 日韩 欧美 亚洲 中文字幕| av又黄又爽大尺度在线免费看| 男女午夜视频在线观看| 日韩制服骚丝袜av| av在线老鸭窝| 又粗又硬又长又爽又黄的视频| 久久国产精品男人的天堂亚洲| 十分钟在线观看高清视频www| videos熟女内射| 菩萨蛮人人尽说江南好唐韦庄| 男人舔女人的私密视频| 久久久精品国产亚洲av高清涩受| 涩涩av久久男人的天堂| xxx大片免费视频| 波多野结衣av一区二区av| 操美女的视频在线观看| 天天躁狠狠躁夜夜躁狠狠躁| 日韩av免费高清视频| 日韩 欧美 亚洲 中文字幕| 国产亚洲欧美精品永久| 午夜免费观看性视频| 国产男女内射视频| 1024香蕉在线观看| 看免费av毛片| 亚洲av电影在线进入| 亚洲国产最新在线播放| 51午夜福利影视在线观看| www日本在线高清视频| 精品免费久久久久久久清纯 | 80岁老熟妇乱子伦牲交| 欧美久久黑人一区二区| 国产在线免费精品| 女人爽到高潮嗷嗷叫在线视频| 精品国产乱码久久久久久男人| 欧美人与性动交α欧美软件| 国产精品久久久久久人妻精品电影 | 美女高潮到喷水免费观看| 国产成人a∨麻豆精品| 国产男女超爽视频在线观看| 不卡av一区二区三区| 欧美xxⅹ黑人| 黄色视频在线播放观看不卡| 欧美日韩成人在线一区二区| 欧美日韩亚洲高清精品| 久久这里只有精品19| 男女午夜视频在线观看| 日韩av免费高清视频| 赤兔流量卡办理| 丰满迷人的少妇在线观看| 91精品三级在线观看| 久久 成人 亚洲| 久久国产精品大桥未久av| 无限看片的www在线观看| 黄频高清免费视频| 老司机在亚洲福利影院| a 毛片基地| 久久天躁狠狠躁夜夜2o2o | 男女床上黄色一级片免费看| 国精品久久久久久国模美| 波野结衣二区三区在线| 欧美国产精品va在线观看不卡| 满18在线观看网站| 亚洲欧美激情在线| 久久精品久久久久久噜噜老黄| 自线自在国产av| 精品一区二区免费观看| 日本欧美视频一区| 制服人妻中文乱码| 久久久久精品人妻al黑| 99九九在线精品视频| 欧美xxⅹ黑人| 99热全是精品| 日本一区二区免费在线视频| 亚洲国产精品国产精品| 日本猛色少妇xxxxx猛交久久| 亚洲精品一二三| av国产精品久久久久影院| 女性生殖器流出的白浆| 热99国产精品久久久久久7| 色精品久久人妻99蜜桃| 女人被躁到高潮嗷嗷叫费观| 亚洲一级一片aⅴ在线观看| 精品一品国产午夜福利视频| 久久久久久久久免费视频了| 国产国语露脸激情在线看| 乱人伦中国视频| 一本—道久久a久久精品蜜桃钙片| 日韩视频在线欧美| 两个人看的免费小视频| 夫妻性生交免费视频一级片| 在现免费观看毛片| 国产精品偷伦视频观看了| 亚洲专区中文字幕在线 | 国产精品久久久人人做人人爽| 久久影院123| av有码第一页| 亚洲精品日本国产第一区| 国产精品国产三级专区第一集| 国产精品久久久久久精品古装| 日韩制服骚丝袜av| 一区福利在线观看| 青青草视频在线视频观看| 国产在线免费精品| 亚洲情色 制服丝袜| 19禁男女啪啪无遮挡网站| 国产在视频线精品| 国产黄色免费在线视频| 欧美日韩亚洲高清精品| 成年女人毛片免费观看观看9 | 91精品三级在线观看| 色播在线永久视频| 午夜福利网站1000一区二区三区| 青春草国产在线视频| 人人妻人人澡人人看| 国产深夜福利视频在线观看| 成年女人毛片免费观看观看9 | 国产av码专区亚洲av| av女优亚洲男人天堂| 欧美精品一区二区免费开放| 男女午夜视频在线观看| 久久久久精品国产欧美久久久 | 五月天丁香电影| 在线观看免费日韩欧美大片| 嫩草影视91久久| 国产亚洲欧美精品永久| 久久久久精品人妻al黑| 免费在线观看视频国产中文字幕亚洲 | 亚洲精品国产av蜜桃| 日韩中文字幕视频在线看片| 亚洲欧美成人精品一区二区| 亚洲国产欧美网| 亚洲av欧美aⅴ国产| 黄色 视频免费看| 操美女的视频在线观看| 免费看不卡的av| 国产成人a∨麻豆精品| 永久免费av网站大全| 国产亚洲一区二区精品| 国产97色在线日韩免费| 久久毛片免费看一区二区三区| 高清不卡的av网站| 亚洲七黄色美女视频| 性色av一级| 国产麻豆69| 人妻 亚洲 视频| 久久青草综合色| 丝袜美腿诱惑在线| 国产亚洲av高清不卡| 国产欧美日韩综合在线一区二区| 精品视频人人做人人爽| 亚洲精品av麻豆狂野| 满18在线观看网站| 女人被躁到高潮嗷嗷叫费观| av网站在线播放免费| 精品人妻熟女毛片av久久网站| 女人爽到高潮嗷嗷叫在线视频| 亚洲av中文av极速乱| 黑人巨大精品欧美一区二区蜜桃| 黑人猛操日本美女一级片| 看免费成人av毛片| 欧美国产精品va在线观看不卡| 亚洲av日韩在线播放| 热99国产精品久久久久久7| 亚洲人成电影观看| 91成人精品电影| av网站在线播放免费| 日韩一区二区视频免费看| 中国国产av一级| 国产精品一区二区在线观看99| 一二三四在线观看免费中文在| 欧美日韩福利视频一区二区| 在线观看一区二区三区激情| xxxhd国产人妻xxx| 51午夜福利影视在线观看| 亚洲精品国产区一区二| 自拍欧美九色日韩亚洲蝌蚪91| 亚洲男人天堂网一区| 老司机深夜福利视频在线观看 | 午夜福利网站1000一区二区三区| 女人被躁到高潮嗷嗷叫费观| 国产男女内射视频| 国产免费现黄频在线看| 777久久人妻少妇嫩草av网站| 99香蕉大伊视频| avwww免费| 国产一区二区 视频在线| 国产免费现黄频在线看| 国产亚洲av高清不卡| 女人精品久久久久毛片| 欧美变态另类bdsm刘玥| 精品少妇内射三级| 麻豆av在线久日| 97在线人人人人妻| 两个人看的免费小视频| 亚洲成人av在线免费| 亚洲五月色婷婷综合| bbb黄色大片| 一区二区日韩欧美中文字幕| 午夜福利视频精品| 日韩av不卡免费在线播放| 亚洲精品第二区| 色吧在线观看| 人妻一区二区av| 国产精品熟女久久久久浪| 中文字幕人妻熟女乱码| 考比视频在线观看| 国语对白做爰xxxⅹ性视频网站| 国产精品成人在线| 欧美97在线视频| 国产99久久九九免费精品| 一级片免费观看大全| 免费黄网站久久成人精品| 国产亚洲最大av| 最近中文字幕2019免费版| 韩国av在线不卡| 少妇的丰满在线观看| 在线天堂中文资源库| 在现免费观看毛片| av福利片在线| 欧美av亚洲av综合av国产av | 黄片小视频在线播放| 日韩免费高清中文字幕av| 成人手机av| 男女无遮挡免费网站观看| 亚洲成人手机| 十八禁人妻一区二区| 搡老岳熟女国产| 久久97久久精品| 亚洲国产欧美一区二区综合| 午夜精品国产一区二区电影| 亚洲欧洲日产国产| 人体艺术视频欧美日本| 老司机靠b影院| 男女高潮啪啪啪动态图| av.在线天堂| 999久久久国产精品视频| 国产精品一区二区在线不卡| 51午夜福利影视在线观看| 晚上一个人看的免费电影| 国产精品 欧美亚洲| 黄色视频不卡| 狂野欧美激情性bbbbbb| 亚洲av男天堂| 看免费成人av毛片| 欧美成人午夜精品| 色综合欧美亚洲国产小说| 亚洲成色77777| 建设人人有责人人尽责人人享有的| 日本欧美国产在线视频| 1024视频免费在线观看| 国产成人系列免费观看| 热re99久久国产66热| 91aial.com中文字幕在线观看| 男女无遮挡免费网站观看| 天天躁日日躁夜夜躁夜夜|