面向分布式虚拟机监视器的节点特征描述机制

摘要

DVMM基于硬件虚拟化技术,在机群硬件之上构建一层系统软件以感知机群的物理资源,并虚拟化全局物理资源,整合构建全局虚拟资源信息呈现给操作系统OS(Operating System)使用.DVMM为多台物理机虚拟为一台虚拟机描述了一个框架,但是此框架存在一个性能上的缺陷:OS对底层硬件拓扑并不真正了解的情况下,可能将不适合在某节点上运行的任务调度到该节点,引发大量的节点间通信,从而导致该任务运行效率显著降低.提出分布式虚拟机监视器DVMM(Distributed Virtual Machine Monitor)中的节点特征描述机制NCM(Node Characterization Mechanism).NCM解决了上述问题,其主要思想是将各个节点的特征描述并保存,为资源管理提供更全面的系统资源描述,呈现系统拓扑结构及节点底层硬件差异供DVMM使用.以上述机制为基础(并在OS的配合下),能够实现任务的高效调度,提高整系统的运行效率,使得DVMM更加完善.%Based on hardware virtualisation technology,DVMM builds a system software tier upon the cluster machines to perceive the physical resources in all nodes at booting time,and virtualises the global physical resource to integrate and construct global virtual resource information for presenting to OS for its use.DVMM describes a framework on the subject that how multiple physical machines are virtualised to a single virtual machine.However there is a pitfall in this framework:in the case of incomplete understanding on underlying hardware topology,OS may dispatch tasks to a node that is not suitable for the task taken,this leads to a lot of communication between the nodes and therefore results in significant deduction of operation efficiency.In this article we study and present the node characterisation mechanism (NCM) in distributed virtual machine monitor (DVMM).NCM solves the above problems.Its main idea is to describe and save the characteristics of each node and to provide a more comprehensive description on system resources for resource management,as well as present the system topology structure and underlying hardware difference of node for DVMM using.Taking the mechanism mentioned above as the basis (and in cooperation with OS),efficient task scheduling can be achieved.Moreover,the operation efficiency of whole system is improved,and the DVMM works more perfect.