【摘要】：云計算作為效用計算，并行計算，網(wǎng)格計算等計算模式融合的產(chǎn)物，它通過資源池化的方式，彈性的對外提供計算、存儲以及網(wǎng)絡(luò)服務(wù)。但日益膨脹的大規(guī)模云計算數(shù)據(jù)中心導(dǎo)致了管理困難以及運營成本攀升的窘境，具體表現(xiàn)在：資源利用率仍舊不高，能耗驚人，虛擬機資源分配不合理。具有彈性特征的云計算資源管理主要考慮兩方面的因素：用戶QoS以及云服務(wù)提供商收入。多數(shù)虛擬機用戶仍然按峰值需求預(yù)定虛擬機的資源配置，導(dǎo)致多數(shù)時段的資源閑置，服務(wù)器長時間空載運行，浪費了大量的電力能源。學(xué)術(shù)界諸多文獻針對資源利用率問題提出了面向虛擬機集群的資源分配以及部署方法，而既滿足用戶QoS又不浪費過多資源的虛擬機最佳資源配置與負載類型及強度息息相關(guān)，本文針對以往多數(shù)文獻對單個虛擬機最佳資源配置與負載關(guān)系研究的不足之處展開理論以及實驗分析，并結(jié)合博弈論應(yīng)用于有限資源條件下對不同資源偏好的虛擬機資源重組優(yōu)化問題中。 主要工作總結(jié)如下： 1)基于虛擬機web服務(wù)場景，使用排隊模型針對web服務(wù)請求的平均請求速率、平均服務(wù)時間、資源平均利用率、平均響應(yīng)時間等性能指標(biāo)建模，并結(jié)合經(jīng)典排隊論Little公式推導(dǎo)出了平均響應(yīng)時間與平均請求速率，平均服務(wù)時間的關(guān)系；通過T時間內(nèi)的web服務(wù)運行過程觀察數(shù)據(jù)，包括資源利用率，平均響應(yīng)時間以及平均請求速率，對模型進行了一致性檢驗，計算出web請求的平均服務(wù)時間；在八組以負載強度為控制變量的實驗測試中，得到與負載強度相應(yīng)的虛擬機資源臨界點，當(dāng)虛擬機資源配置低于臨界點，響應(yīng)時間出現(xiàn)急劇上升，最后通過對實驗數(shù)據(jù)的回歸分析得出了服務(wù)請求速率與最佳資源配置的顯著關(guān)系。 2)針對有限資源條件下的虛擬機資源重組問題，根據(jù)虛擬機最佳資源配置決策是否沖突分別提出了解決方法�；谏鲜鲂阅苣Ｐ停�，虛擬機根據(jù)部署的負載類型與強度獨立計算申請所需資源，若虛擬機目標(biāo)資源配置總和未超過初始重組資源總和，則直接根據(jù)各自需要進行資源重分配，否則認為資源重組沖突，針對沖突問題，應(yīng)用博弈談判，經(jīng)過離散迭代求解。實驗證明了該方法在以下幾點的有效性：提高資源利用率，優(yōu)化虛擬機性能，保障資源重分配的公平性。
[Abstract]:Cloud computing is the product of utility computing, parallel computing, grid computing and so on. It provides computing, storage and network services through resource pool. However, the expanding large-scale cloud computing data center has led to the difficulties of management and rising operating costs, which are reflected in: the resource utilization is still not high, the energy consumption is amazing, and the allocation of virtual machine resources is unreasonable. Flexible cloud computing resource management mainly considers two factors: user QoS and cloud service provider revenue. Most virtual machine users still preorder the resource allocation of virtual machine according to the peak demand, which leads to idle resources in most periods, the server running without load for a long time, and wasting a lot of power energy. For the resource utilization problem, many literatures have put forward the resource allocation and deployment method for virtual machine cluster. However, the optimal allocation of virtual machine resources, which not only satisfies the user QoS but also does not waste too much resources, is closely related to the load type and intensity. In this paper, the theoretical and experimental analysis is carried out in view of the shortcomings of most previous literatures on the relationship between optimal resource allocation and load of a single virtual machine. And the game theory is used to optimize the resource recombination of virtual machine with different resource preference under the condition of limited resources. The main work is summarized as follows: 1) based on the virtual machine web service scenario, the average request rate, average service time, average resource utilization and average response time of web service requests are modeled by queuing model. The relationship between average response time, average request rate and average service time is derived by using the classical queuing theory Little formula, and the data of web service running process in T time, including resource utilization, are observed. The average response time and the average request rate were tested to calculate the average service time of the web request, and in eight groups of experiments with load intensity as the control variable, the average service time of the web request was calculated. The critical point of virtual machine resources corresponding to the load intensity is obtained. When the virtual machine resource allocation is below the critical point, the response time increases sharply. Finally, through regression analysis of experimental data, the significant relationship between service request rate and optimal resource allocation is obtained. 2) aiming at the problem of virtual machine resource reorganization under limited resources, According to the optimal resource allocation decision of virtual machine, the solutions are proposed respectively. Based on the above performance model, the virtual machine independently calculates the required resources according to the load type and intensity of the deployment. If the total allocation of the target resources of the virtual machine is not more than the sum of the initial reorganization resources, then the resources are reallocated directly according to their respective needs. Otherwise, it is considered that the conflict of resource recombination is solved by using game negotiation and discrete iteration to solve the conflict problem. Experiments show that the proposed method is effective in the following aspects: improving resource utilization, optimizing the performance of virtual machine and ensuring the fairness of resource redistribution.
【學(xué)位授予單位】：重慶大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：TP393.01

【參考文獻】

相關(guān)期刊論文 前2條

1 華夏渝;鄭駿;胡文心;;基于云計算環(huán)境的蟻群優(yōu)化計算資源分配算法[J];華東師范大學(xué)學(xué)報(自然科學(xué)版);2010年01期

2 周文俊;曹健;;基于預(yù)測及蟻群算法的云計算資源調(diào)度策略[J];計算機仿真;2012年09期

本文編號：2278387