虚拟化系统中的软件自愈相关技术研究

发布时间：2018-07-28 13:27

【摘要】：任何软件系统都不可避免地发生软件衰退,虚拟化系统自然也不例外,因此有必要研究适合于虚拟化系统的自愈技术。虚拟化系统通常采用虚拟机在线迁移技术实现负载均衡、故障缓解等管理功能以提高系统可用性,而软件自愈技术作为一种预防性维护和管理技术,是降低系统衰退导致的系统失效风险、提高系统可用性的更为主动的方法。因此,本文融合了软件自愈技术和虚拟机在线迁移技术,以软件自愈分析模型作为自愈技术研究的核心内容,研究基于虚拟机在线迁移的自愈分析模型和虚拟机在线迁移优化,分析自愈策略的成本,试图为实施自愈技术提供决策依据。本文完成的主要工作和取得的研究成果如下:(1)针对单服务器虚拟化系统的软件衰退问题,采用形式化模型描述了应对衰退的自愈策略,以虚拟机监视器和/或虚拟机集群为自愈对象分别构造了四种不同的自愈策略对应的自愈分析模型,采用了马尔可夫再生理论进行模型分析,并给出了一种最优自愈策略的搜索方法。仿真实验表明合理的自愈策略不仅能有效提升系统可用性,而且能降低系统停机成本。(2)针对多服务器虚拟化系统中虚拟机监视器和虚拟机之间的强依赖关系,利用随机回报网模型分析了基于时间虚拟机冷自愈模型、虚拟机热自愈模型和虚拟机迁移自愈模型,并提出了一个融入负载因素的基于时间和负载的虚拟机迁移自愈模型。仿真实验结果表明无论稳态可用性还是事务丢失数量,虚拟机迁移自愈策略均优于虚拟机冷/热自愈策略;融入负载因素的自愈模型无论在系统可用性和吞吐率等指标方面都优于基于时间策略的自愈模型,且在系统负载动态变化时表现地更为稳定。(3)为了降低多服务器虚拟化系统中基于虚拟机迁移的自愈成本,提出了一个虚拟机动态迁移预拷贝优化算法MSTO。该算法根据时间局部性和空间局部性原理,设计了一种动态的迁移窗口设定和迁移页面选择策略,以适应于动态的页面脏度和迁移链路可用带宽。实验结果表明MSTO在停机时间、总迁移时间和迁移总页数等方面均优于Xen的预拷贝方法。
[Abstract]:Any software system inevitably occurs software decline, virtualization system is no exception, so it is necessary to study self-healing technology suitable for virtualization system. Virtual machine online migration technology is usually used to achieve load balancing, fault mitigation and other management functions to improve system availability, while software self-healing technology as a preventive maintenance and management technology, It is a more active way to reduce the risk of system failure and improve system availability. Therefore, this paper combines the software self-healing technology and virtual machine online migration technology, taking the software self-healing analysis model as the core of the self-healing technology research, studies the self-healing analysis model based on virtual machine online migration and the virtual machine on-line migration optimization. This paper analyzes the cost of self-healing strategy and tries to provide decision basis for implementing self-healing technology. The main work and research results of this paper are as follows: (1) aiming at the problem of software degradation in single-server virtualization system, a formal model is used to describe the self-healing strategy. Taking virtual machine monitor and / or virtual machine cluster as self-healing objects, four self-healing analysis models corresponding to different self-healing strategies are constructed, and Markov reproducing theory is used to analyze the model. A search method of optimal self-healing strategy is given. Simulation results show that reasonable self-healing strategy can not only effectively improve system availability but also reduce system downtime cost. (2) aiming at the strong dependence between virtual machine monitor and virtual machine in multi-server virtualization system. The cold self-healing model, the thermal self-healing model and the migration self-healing model of virtual machine based on time and load are analyzed by using stochastic return net model, and a model of virtual machine migration self-healing based on time and load is proposed. Simulation results show that virtual machine migration self-healing strategy is superior to virtual machine cold / hot self-healing strategy regardless of steady-state availability or transaction loss. The self-healing model incorporating load factors is superior to the time-based self-healing model in terms of system availability and throughput. In order to reduce the self-healing cost of virtual machine migration in multi-server virtualization system, a dynamic migration pre-copy optimization algorithm MSTO is proposed. According to the principle of time locality and spatial locality, this algorithm designs a dynamic migration window setting and migration page selection strategy to adapt to the dynamic page dirt and the available bandwidth of the migration link. The experimental results show that MSTO is superior to Xen in terms of downtime, total migration time and total number of pages.
【学位授予单位】：南京理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP302

【参考文献】