虚拟环境下Web服务动态负载均衡策略改进

为了提高Web服务集群的伸缩性和自动化能力,从虚拟化和负载均衡两方面研究集群系统,对现有负载采集策略做了改进,设计并实现了一种可根据负载值自动控制集群规模的模型XCluster。新模型运行在Xen提供的虚拟化环境中,实时监视宿主机层和虚拟机层的负载状态,随着集群系统总负载的增长,逐渐引入新的虚拟机来扩大集群规模,同时将任务合理分配到各个虚拟机节点上;当总负载下降时,逐渐关闭虚拟机缩小集群规模,释放出来的硬件资源又可以提供给其他集群系统使用。理论分析和实验结果表明,XCluster只需占用很少的网络通信量完成信息收集和命令下达,能够充分利用虚拟机易于管理的优势完成后端节点的调度,并且在任务总量相同的情况下,使用尽可能少的集群节点来执行任务。

Improved dynamic load-balancing strategy of web servers in virtual environment

In order to improve the scalability and automation of web cluster servers, we study many cluster systems from the aspects of virtualization and load balancing, propose a load information collection algorithm based on the Xen server virtualization platform, design and implement a modeling system called XCluster which can automatically control the cluster scale according to the load status. A monitoring service in the new model collects and quantifies the loads of the hosts and active nodes in real time. VM controllers will be notified to deploy more VMs from an optimum load host when light load nodes are in short supply; however, when there are too many light load nodes, a number of VMs will be shutdown to release more hardware resources. Theoretical analysis and experimental results show that the XCluster can make full use of the unique capabilities of the virtualization platform which can manage all back end nodes in an economical way. By scheduling VMs, the clusters can avoid resource waste, yet maintaining the applications quality of service within an acceptable size.