本文選題：數(shù)據(jù)中心 + 虛擬化��； 參考：《上海交通大學(xué)》2013年碩士論文

【摘要】：隨著云計算和虛擬化的興起和發(fā)展，數(shù)據(jù)中心在保證應(yīng)用的服務(wù)質(zhì)量的同時，能耗的管理也越來越受到重視。然而由于應(yīng)用的復(fù)雜性增強(qiáng)，負(fù)載的動態(tài)變化性增多，能夠在保證數(shù)據(jù)中心服務(wù)質(zhì)量的同時，自適應(yīng)地調(diào)節(jié)數(shù)據(jù)中心的資源分配，并進(jìn)一步的降低數(shù)據(jù)中心能耗的系統(tǒng)還有待進(jìn)一步的研究。 本文分析了現(xiàn)有的虛擬化數(shù)據(jù)中心能耗管理的技術(shù)和方法，提出了基于智能控制理論進(jìn)化控制算法的數(shù)據(jù)中心能耗管理的自適應(yīng)控制系統(tǒng)。系統(tǒng)由資源監(jiān)測器、資源協(xié)調(diào)器和資源分配器組成。本文的主要創(chuàng)新點(diǎn)包括以下三點(diǎn)： 1）系統(tǒng)能夠?qū)崿F(xiàn)能耗的自適應(yīng)的管理。能夠根據(jù)負(fù)載的動態(tài)變化進(jìn)行資源的動態(tài)分配。系統(tǒng)能夠?qū)崿F(xiàn)能耗自適應(yīng)管理的前提是能夠進(jìn)行能耗和性能的監(jiān)測以及精度較高的在線預(yù)測。監(jiān)測由資源監(jiān)測器實(shí)現(xiàn)，資源監(jiān)測器負(fù)責(zé)監(jiān)測物理集群的整體能耗和每個虛擬機(jī)的性能，反饋給模型預(yù)估器進(jìn)行預(yù)估。模型預(yù)估器采用微粒蟻群算法對物理集群的能耗和虛擬機(jī)資源分配之間的模型以及虛擬機(jī)的吞吐量和虛擬機(jī)資源分配之間的模型進(jìn)行在線辨識，并把辨識的能耗和性能模型的結(jié)果反饋給資源優(yōu)化器。 2）系統(tǒng)能夠?qū)崿F(xiàn)能耗和性能的統(tǒng)一管理。在保證數(shù)據(jù)中心服務(wù)質(zhì)量的前提下，進(jìn)一步地降低能耗。系統(tǒng)能耗的降低主要通過資源優(yōu)化器進(jìn)行。資源優(yōu)化器按照預(yù)估模型，采用遺傳算法對關(guān)于能耗和性能的目的效用函數(shù)進(jìn)行優(yōu)化，得到分配給每個虛擬機(jī)的資源優(yōu)化結(jié)果，然后通過Xen虛擬機(jī)管理器進(jìn)行資源的分配，以此來實(shí)現(xiàn)能耗和性能的統(tǒng)一管理。 3）系統(tǒng)具有較高的可擴(kuò)展性。系統(tǒng)分別使用微粒蟻群算法和遺傳算法進(jìn)行模型的建模和優(yōu)化，由于智能控制理論算法在模型求解過程中，不需要考慮模型的具體特性，使得該方法能夠適應(yīng)復(fù)雜的非線性系統(tǒng)和模型。同時，在算法的實(shí)現(xiàn)過程中，有多個參數(shù)可供設(shè)定和調(diào)節(jié)，適用于約束條件較多的模型，提高了系統(tǒng)的改進(jìn)度和可擴(kuò)展性。 本文為了對系統(tǒng)的有效性和穩(wěn)定性進(jìn)行驗(yàn)證，搭建了基于Xen的測試平臺，使用TPC-W進(jìn)行負(fù)載的生成。通過對實(shí)驗(yàn)結(jié)果的分析，本系統(tǒng)能夠保證應(yīng)用的服務(wù)質(zhì)量，自適應(yīng)地進(jìn)行數(shù)據(jù)中心虛擬機(jī)的資源的分配，同時帶來了一定的能耗節(jié)約。
[Abstract]:With the rise and development of cloud computing and virtualization, data centers pay more and more attention to the management of energy consumption while ensuring the service quality of applications.However, due to the complexity of the application and the increasing dynamic variation of the load, the resource allocation of the data center can be adjusted adaptively while ensuring the quality of service in the data center.And the system of further reducing the energy consumption of data center still needs further research.This paper analyzes the existing technologies and methods of virtual data center energy management, and proposes an adaptive control system for data center energy management based on intelligent control theory evolutionary control algorithm.The system consists of resource monitor, resource coordinator and resource allocator.The main innovations of this paper are as follows:1) the system can realize adaptive management of energy consumption.The dynamic allocation of resources can be made according to the dynamic change of load.The premise that the system can realize adaptive management of energy consumption is to monitor energy consumption and performance and to predict on line with high precision.The monitoring is implemented by a resource monitor, which is responsible for monitoring the overall energy consumption of the physical cluster and the performance of each virtual machine, and feedbacks to the model predictor for prediction.The model predictor uses particle ant colony algorithm to identify the model between the energy consumption of physical cluster and the allocation of virtual machine resources, and the model between throughput of virtual machine and resource allocation of virtual machine.The results of the identified energy consumption and performance models are fed back to the resource optimizer.2) the system can realize unified management of energy consumption and performance.Under the premise of ensuring the service quality of the data center, the energy consumption is further reduced.The reduction of system energy consumption is mainly carried out by the resource optimizer.According to the prediction model, the resource optimizer uses genetic algorithm to optimize the purpose utility function about energy consumption and performance, and obtains the optimization results of resources allocated to each virtual machine, and then allocates the resources through the Xen virtual machine manager.To achieve the unified management of energy consumption and performance.3) the system has high expansibility.The system uses particle ant colony algorithm and genetic algorithm to model and optimize the model, because the intelligent control theory algorithm does not need to consider the specific characteristics of the model in the process of solving the model.The method can adapt to complex nonlinear systems and models.At the same time, in the implementation of the algorithm, there are many parameters to be set and adjusted, which is suitable for the model with more constraints, and improves the improvement and expansibility of the system.In order to verify the validity and stability of the system, a test platform based on Xen is built and the load is generated by TPC-W.Through the analysis of the experimental results, the system can guarantee the quality of service of the application, adaptively allocate the resources of the virtual machine in the data center, and at the same time bring about some energy saving.
【學(xué)位授予單位】：上海交通大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2013
【分類號】：TP308

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 顧振宇;張申生;李曉勇;;Xen中Credit調(diào)度算法的優(yōu)化[J];微型電腦應(yīng)用;2009年02期

，

本文編號：1768674