Xen虚拟机迁移机制和负载均衡策略研究

发布时间：2018-01-06 23:34

本文关键词：Xen虚拟机迁移机制和负载均衡策略研究　出处：《山东大学》2012年硕士论文　论文类型：学位论文

更多相关文章： 虚拟化 Xen 动态迁移 负载均衡

【摘要】：虚拟化技术能够在云计算平台中创建多个虚拟机作为虚拟操作平台,满足多用户共享平台时隔离执行环境的需求。并且通过虚拟机动态迁移技术和虚拟机负载均衡策略,可以灵活和动态的管理与调节云计算环境中的资源,使系统资源得到充分利用。在实际应用过程中,由于业务特性等因素,虚拟机动态迁移过程中需要传输大量内存页面,造成动态迁移与应用程序的性能下降,以及负载均衡策略实施效果不佳,影响云计算平台性能。因此,虚拟机动态迁移机制和负载均衡策略是当前虚拟化技术领域的研究热点。虚拟机动态迁移机制主要是通过气球驱动、数据压缩、Copy on Write以及硬件调度等技术进行优化,以减少迭代传输过程中内存页面的传输量,提高动态迁移性能,但是现有技术没有考虑具体业务特征,采用单一策略不能满足不同特征业务的需求,尤其是在I/O密集型业务环境中,动态迁移过程需要传输大量内存脏页,占用过多物理网络带宽,增加迭代时间和宕机时间,从而影响迁移性能和业务运行质量。负载均衡方面主要是利用阈值、集中控制和历史数据预测等方法生成负载均衡策略,在实际应用中,由于业务多样性和业务量突变的影响,造成适用性与兼容性不强,并且缺少数据抖动的处理以及迁移自身开销对业务影响的考虑。针对上述问题,本文提出了面向业务特征的自适应虚拟机动态迁移带宽分配算法,详细分析了虚拟机动态迁移过程中现有物理带宽分配方法存在的不足,将业务特征和服务质量引入带宽分配方案,并提出基于复杂业务环境的负载均衡策略,以增强系统的适用性和鲁棒性。本文主要贡献包括： (1)提出面向业务特征的自适应虚拟机动态迁移带宽分配算法。该算法针对I/O密集型业务环境,通过分析运行业务的特征,利用迁移迭代过程中内存脏页率,以及网络往返时延RTT方法,预测下一时刻运行业务的物理网络带宽使用量,从而自适应分配虚拟机动态迁移使用带宽。在保证迁移性能和系统可靠性的同时,减少迭代时间和宕机时间。 (2)提出了基于复杂业务的动态负载均衡策略。该策略利用已扩展的集群监控软件Ganglia (?)文集和发送物理节点和虚拟节点的性能信息,使用加入权重值的“Black-box and Gray-box"策略,根据业务特征动态的进行数据分析,并动态调整过载次数的阈值以防止突发峰值对负载均衡效果的影响,然后通过贪心算法生成负载均衡策略,以及使用Rainbow虚拟机管理平台执行虚拟机动态迁移。同时利用Ganglia通信机制取代了Sandpiper架构中使用的Twisted通信机制,增加了整体系统的鲁棒性。 (3)实现了基于Xen虚拟机的面向业务特征的自适应虚拟机动态迁移带宽分配算法,实验结果表明,本算法能够在物理带宽资源有限的前提下,合理利用空闲带宽资源,减少虚拟机迭代时间和宕机时间,确保业务服务质量。同时,将基于复杂业务环境的负载均衡策略部署在运行VoIP、Streaming和C-RAN实际通信业务的环境中,并模拟不同时段不同业务的业务量,验证了该策略的适应性、可行性以及负载均衡效果。
[Abstract]:Virtualization technology in cloud computing platform can create multiple virtual machines as virtual operating platform, to meet the multi user sharing platform when isolated execution environment needs. And through the dynamic migration of virtual machines and virtual machine load balancing strategy, flexible and dynamic management and regulation in cloud computing resources, the system resources make full use of it. In the actual application process, due to the factors of business characteristics, transmission of a large number of memory pages to dynamic virtual machine migration process, resulting in decreased performance and dynamic application migration, and the effect of load balancing strategy, influence cloud computing platform performance. Therefore, the dynamic mechanism of virtual machine migration and load balancing strategy research the current hot virtualization technology.
The main dynamic virtual machine migration mechanism is driven by the balloon, data compression, Copy on Write and hardware scheduling technique is optimized to reduce the amount of memory page transmission iteration in the transmission process, improve the dynamic performance of existing technology transfer, but did not consider the specific business characteristics, a single strategy can not meet the different characteristics of the needs of the business, especially in the I/O intensive business environment, dynamic migration process need to transfer large amounts of memory dirty pages, occupying too much physical network bandwidth, increasing the iteration time and downtime, thus affecting the migration performance and operation quality. Load balancing is the use of threshold, centralized predictive control and the historical data generation method of load balancing strategy in practice in the application, due to the impact of diversity of business and business volume caused by mutations, applicability and compatibility is not strong, and the lack of data jitter at And its effect on the business migration overhead into account. According to the above problems, the proposed adaptive bandwidth allocation algorithm for virtual machine migration dynamic business oriented features, a detailed analysis of the shortcomings of the existing physical bandwidth allocation method for dynamic migration of virtual machines in the process of the business characteristics and quality of service into the bandwidth allocation scheme, and proposed load balancing based on the strategy of complex business environment, to enhance the applicability and robustness of the system. The main contributions of this paper include:
(1) the migration of adaptive bandwidth allocation algorithm for virtual machine dynamic business oriented features. The algorithm for the I/O intensive business environment, by analyzing the characteristics of operation, the use of memory dirty pages in the iterative process migration rate and RTT RTT method, the physical network bandwidth prediction next time to run the business use, and adaptive allocation of virtual machine migration using bandwidth. In ensuring the reliability and performance of transfer system at the same time, reduce the iteration time and downtime.
(2) proposed a dynamic load balancing strategy based on complex business. This strategy has been extended by the cluster monitoring software (Ganglia?) performance information collection and transmission of physical nodes and virtual nodes, using weighting value "Black-box and Gray-box" strategy, according to the business characteristic of dynamic data analysis, and dynamic adjustment the number of the threshold to prevent overload on the peak effect of sudden load balancing effect, and then by the greedy algorithm to generate the load balancing strategy, and the use of Rainbow virtual machine management platform to implement the dynamic migration of virtual machines. At the same time using Ganglia communication mechanism to replace the Twisted communication mechanism used in Sandpiper architecture and increase the robustness of the whole system.
(3) the migration of adaptive bandwidth allocation algorithm for virtual machine dynamic business oriented features based on Xen virtual machine, the experimental results show that this algorithm can based on physical bandwidth limited resources, reasonable use of idle bandwidth resources, reduce the iteration time and virtual machine down time, ensure the quality of service. At the same time, the load the complex business environment deployed in operation balancing strategy based on VoIP, Streaming and C-RAN in actual communication business environment, and the simulation of traffic in different periods of different services, this strategy is proved feasible and the adaptability of load balancing effect.

【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP302

【参考文献】