林闖

摘 要：在云計(jì)算日趨流行的背景下，虛擬數(shù)據(jù)中心以其靈活、低成本、易部署等方面的優(yōu)勢(shì)，成為云基礎(chǔ)設(shè)施的首選。服務(wù)調(diào)度和資源分配作為虛擬數(shù)據(jù)中心的核心問(wèn)題，其性能和效率直接影響著上層云服務(wù)質(zhì)量。而系統(tǒng)虛擬化特性和機(jī)制給服務(wù)調(diào)度和資源分配問(wèn)題的研究帶來(lái)了新的挑戰(zhàn)。圍繞虛擬數(shù)據(jù)中心服務(wù)調(diào)度和資源分配問(wèn)題，以理論模型、機(jī)制設(shè)計(jì)和特性分析為重點(diǎn)，開展深入研究。階段性成果包括以下內(nèi)容：（1）對(duì)虛擬數(shù)據(jù)中心體系結(jié)構(gòu)和系統(tǒng)虛擬化機(jī)制進(jìn)行抽象，采用層次化的思想，利用隨機(jī)Petri網(wǎng)模型和組合模型相結(jié)合的方法，建立了虛擬數(shù)據(jù)中心的數(shù)學(xué)模型，對(duì)系統(tǒng)虛擬化引入的服務(wù)器整合和動(dòng)態(tài)遷移等特性和機(jī)制進(jìn)行描述和分析，并量化評(píng)價(jià)了虛擬數(shù)據(jù)中心的可信賴性。（2）提出了虛擬數(shù)據(jù)中心的服務(wù)調(diào)度框架，并建立了以優(yōu)化可用性和響應(yīng)時(shí)間為目標(biāo)的虛擬數(shù)據(jù)中心服務(wù)調(diào)度的一般數(shù)學(xué)模型。首次引入一型和二型模糊邏輯系統(tǒng)解決虛擬化的動(dòng)態(tài)特性和可用性需求的不確定性帶來(lái)的模糊性，在此基礎(chǔ)上提出了一種虛擬數(shù)據(jù)中心的服務(wù)調(diào)度算法。與傳統(tǒng)的調(diào)度算法相比，該算法能夠更好的保證可用性，同時(shí)實(shí)現(xiàn)了更好的響應(yīng)時(shí)間性能。（3）著眼于數(shù)據(jù)中心的能耗和負(fù)載均衡，提出了虛擬數(shù)據(jù)中心資源分配問(wèn)題的最優(yōu)化數(shù)學(xué)模型。借助于經(jīng)典的多維裝箱問(wèn)題，并有效應(yīng)對(duì)異構(gòu)資源需求和能耗特性帶來(lái)的復(fù)雜性，設(shè)計(jì)了能耗優(yōu)化的虛擬數(shù)據(jù)中心資源分配近似算法。實(shí)現(xiàn)了降低能耗，提高資源利用率，同時(shí)達(dá)到良好的負(fù)載均衡。（4）總結(jié)出了三種典型的虛擬數(shù)據(jù)中心管理系統(tǒng)結(jié)構(gòu)。結(jié)合虛擬數(shù)據(jù)中心的特性和機(jī)制，對(duì)三種不同的虛擬數(shù)據(jù)中心管理系統(tǒng)的性能和可靠性進(jìn)行了建模，并進(jìn)行了量化分析比較。提出了為構(gòu)建虛擬數(shù)據(jù)中心管理系統(tǒng)選擇適當(dāng)結(jié)構(gòu)的幾條基本原則，為虛擬數(shù)據(jù)中心工程實(shí)踐提供理論指導(dǎo)和參考。

關(guān)鍵詞：云計(jì)算 虛擬數(shù)據(jù)中心 服務(wù)調(diào)度 資源分配

Abstract：Focusing on the model， scheme design and performance analysis， this report carry on an intensive research on the service scheduling and resource provisioning of virtual data center. The work is summarized as follows：（1）Taking the hierarchy methodology， the model of virtual data center is built combining the stochastic petri nets and combinatorial model， based on the abstract of virtual data center structure and virtualization mechanisms. The characteristics and mechanisms of system virtualization are studied and analyzed， and the dependability of virtual data center is evaluated.（2）The framework model of service scheduling of virtual data center is proposed， and the optimization model is built with availability and responsiveness as the optimization objects. In order to overcome the scheduling difficulties caused by the uncertain workload of virtualized server nodes and the vagueness of availability， a graceful fuzzy prediction method based on the type-I and type-II fuzzy logic systems is first given. Then a novel dynamic scheduling algorithm named SALAF is designed. Experimental results show that the proposed algorithm SALAF can improve the total availability of the virtual data center while providing good responsiveness performance.（3）Aiming at the requirement of energy conservation and load balance， the optimization model of the resource provisioning is built. By defining the concept of energy marginal cost to relieve the complexity of heterogeneous resource demands and energy profits， an Energy-Efficient Dynamic Resource Provisioning （E2DRP） algorithm is proposed， by means of the multi-dimensional bin packing problem. The experimental results show that the proposed algorithms can improve energy efficiency， meanwhile achieve better server utilization and load-balance.（4）Three typical structures of the virtual data center management system （VDMS） are summarized， which are centralized， hierarchical and peer-to-peer structures. And the performance and reliability of the VDMS with the three typical structures is analyzed and evaluated. Some useful rules and conclusions are drawn and proved which are directive and with reference value for the construction of the virtual infrastructure management systems with higher performance， fault-tolerance and scalability.

Key Words：Cloud Computing；Virtual Data Center；Service Scheduling； Resource Provisioning

閱讀全文鏈接（需實(shí)名注冊(cè)）：http：//www.nstrs.cn/xiangxiBG.aspx？id=51858&flag=1