GOSBMB:基于客户操作系统行为的虚拟机内存均衡方法

GOSBMB:Dynamic Memory Balancing Based on Guest Operating System Behaviors for Virtual Machines

在虚拟机环境中为客户操作系统分配内存资源时,需在性能和资源利用效率之间进行权衡.本文提出一种基于客户操作系统行为的虚拟机内存均衡方法GOSBMB(Guest Operating System Behaviors based Memory Balancer).该方法在尽量降低性能损失的前提下,根据客户操作系统中进程工作集和磁盘页面缓存对虚拟机监控器呈现的行为特征,以对客户操作系统透明的方式估计它们各自的内存需求,并按需动态调整内存资源.在Xen上实现了GOSBMB原型系统,实验表明,使用GOSBMB动态调整内存资源时,在节约内存资源达69.6%的情况下,客户操作系统的性能损失低于7.6%.

The application performance is as important as the resource utilization when we allocate memory for guest operating systems in virtualized environments.This paper presents a memory-balancing method for virtual machines,named GOSBMB（Guest Operating System Behaviors based Memory Balancer）.It aims to minimize the performance loss brought by balancing memory.The memory need of the guest operating system consists of two parts which respectively,cover the amount of the memory consumed by the process working sets,and the memory used as the page cache.GOSBMB estimates the memory need by observing VMM architecturally visible events transparently.Then it allocates the memory resource for the guest operating systems on demand.We have implemented a working prototype of GOSBMB on Xen platform.The memory can be saved up to 69.6% with only 7.6% performance loss when we use GOSBMB to balance memory dynamically.