面向多处理器虚拟机的动态NUMA方法

发布时间：2018-03-29 16:50

本文选题：虚拟机监控器　切入点：多处理器　出处：《小型微型计算机系统》2015年04期

【摘要】：NUMA内存技术在增加了计算机系统内存带宽的同时,其跨节点访存延迟高的特点也增加了虚拟化平台的内存管理难度.现有的多核虚拟机NUMA优化方法大多为静态方法,即无需动态调节虚拟机状态.这些方法大都限制了虚拟化监控器调度虚拟CPU的灵活性,降低了虚拟化平台负载均衡的能力,而影响虚拟机的运行效率.针对该问题,提出动态NUMA方法——在告知虚拟机NUMA拓扑结构的情况下,该方法通过调节虚拟机内操作系统NUMA相关数据结构方法,达到配合虚拟机监控器任务调度器的目的,实现虚拟机运行效率的提升.相较于静态方法,本文提出调节虚拟机NUMA数据结构,以适应虚拟机监控器调度器的动态方法.此方法提高了虚拟CPU的调度灵活性,提高了虚拟化平台负载均衡的能力,从而保证了多处理器虚拟机的运行效率.
[Abstract]:The NUMA memory technology in the computer system to increase memory bandwidth at the same time, the inter node latency is high also increased the difficulty of management memory virtualization platform. The existing multi-core NUMA virtual machine optimization methods are static method, which is required to dynamically adjust the state of the virtual machine. These methods are mostly limited to virtualization monitor the scheduling of virtual CPU flexibility, reduces the load balancing ability of virtual platform, which affects the efficiency of the virtual machine. Aiming at this problem, put forward the dynamic NUMA method in inform the topological structure of NUMA virtual machine case, the method by adjusting the virtual machine operating system NUMA data structure method, with the virtual machine monitor the task scheduler to implement a virtual machine to enhance the operating efficiency. Compared with static method, this paper adjust the data structure of NUMA virtual machine, in order to adapt to the virtual machine monitor This method improves the scheduling flexibility of virtual CPU, improves the load balancing ability of virtualization platform, and ensures the operation efficiency of multiprocessor virtual machine.

【作者单位】：复旦大学软件学院;复旦大学上海市数据科学重点实验室;上海交通大学软件学院并行与分布式系统研究所;
【基金】：国家自然科学基金项目(61003002)资助
【分类号】：TP302

【参考文献】