一种基于网络感知的虚拟机再调度算法  

A Network-Aware VM Re-Scheduling Algorithm

作　　者：罗刚毅[1,2] 钱柱中[1,2] 陆桑璐[1,2] 

机构地区：[1]南京大学计算机软件新技术国家重点实验室,南京210093 [2]南京大学计算机科学与技术系,南京210093

出　　处：《计算机学报》2015年第5期932-943,共12页Chinese Journal of Computers

基　　金：国家自然科学基金(61073028;61202113;61021062);江苏省自然科(BK2011510);江苏省科技支撑项目(BE2013116)资助~~

摘　　要：有效的虚拟机调度策略能够提高数据中心的资源利用率,降低运行时能耗.现有调度算法综合考虑了虚拟机在CPU、内存和网络方面的需求,通过合理部署虚拟机,以期最小化计算、存储与网络的代价.然而,在线的虚拟机部署策略较少考虑由于虚拟机退出所造成的资源利用率下降与网络延迟上升的问题.为此,文中深入研究面向网络感知的周期性资源重配置问题,提出了面向网络感知的虚拟机再调度算法,通过适当的虚拟机迁移,提高部署在虚拟机上任务的性能以及数据中心整体的网络通信效率.算法通过尽可能低代价的虚拟机迁移来提高虚拟机之间的网络通信能力,以提升虚拟机组的整体运行效率,并保持物理机占用但不显著提高.作者通过两个测试平台在真实环境中验证了算法的有效性;通过真实的数据集和模拟实验,在多种虚拟机部署算法下,对比了应用虚拟机再调度算法前后虚拟机的部署效果,验证了该算法能够以较小的代价使得高网络通信代价的任务数明显减少,虚拟机组的网络通信能力显著提高.An effective and efficient VM scheduling algorithm can improve utilization rate of physical servers and lower energy cost. Current VM allocation algorithms focus on the require- ment of CPU, Memory and network bandwidth which trying to allocate VM into physical servers in a low cost way. However, when the jobs are finished, related VMs quit from the system, which leads to the decline of resource utilization and the increasing of transmission delay. This paper studied the network-aware resource reconfiguration problem and proposed a network-aware VM re-scheduling algorithm based on VM live migration. This algorithm focuses on improving communication ability among VMs to promote the overall performance in a way of low migration cost and little extra physical used. Two test beds are deployed in a real environment to examine the effectiveness of the VM reconfiguration algorithm. We compared the result of before and after using our algorithm with different consolidation algorithms based on the real workload data and simulation. The results show that our algorithm considerably reducing the amount of high-delay jobs and improving the communication ability among VMs only with very small cost.

关 键 词：云计算 数据中心 网络感知 虚拟机再调度 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]