项目名称: 虚拟环境中处理器高速缓存管理的关键技术研究
项目编号: No.61202057
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 张科
作者单位: 中国科学院计算技术研究所
项目金额: 22万元
中文摘要: 虚拟化技术是云计算底层基础架构中的关键技术之一,备受多方关注。然而已有研究工作和商用技术不能在虚拟环境中充分利用多核处理器的私有和共享Cache资源,也未能通过虚拟机管理器对Cache资源进行有效管理。因此,本课题提出一套云计算虚拟环境中多核处理器Cache的管理策略,具体包含三方面研究工作:1)挖掘Cache替换和预取策略的潜在关系使二者结合以提高私有Cache性能;2)采用基于虚拟机访存流特征的管理策略提升虚拟机之间共享Cache的性能;3)在虚拟机管理器(VMM)中使用基于指令动态插入的Cache管理机制以实现Cache的软件控制。通过涵盖从宏观用户体验至微观体系结构各层级指标参数的实验评测方法,提高云计算虚拟环境下多核处理器Cache的利用率和命中率,最终获得虚拟机性能和用户体验的共同提升。本项目拟搭建两套实验和模拟平台,发表8-10篇SCI/EI索引论文,并申请10项以上专利。
中文关键词: 高速缓存;虚拟化;片上总线;资源共享;远程内存访问
英文摘要: Virtualization is a key enabling technology for cloud computing infrastructure, and is attracting growing attention from both academia and industry. However, prior research work and existing commercial technologies could not effectively and efficiently utilize both private and shared cache resources in multi-core processors under a virtualized environment. Moreover, to our best of knowledge, there is no virtual machine manager that could actively influence the cache. Therefore, a set of multi-core cache management policies is proposed for virtualized cloud environment in this proposal. Specifically, we will conduct this research in three aspects: 1) leveraging the unexplored relationship between replacement and prefetching mechanisms in order to increase the performance of private caches; 2) providing different management policies in the shared cache for diverse virtual machines in terms of memory access pattern; 3) actively and dynamically instrument a running virtual machine in the virtual machine manager to control cache behavior. By means of experiments from user experience in macro-scale down to architectural parameters in micro-scale, the utilization and hit rate of caches in multi-processors are expected to boost in the virtualized cloud environment, and the overall performance of virtual machines and use
英文关键词: cache;virtualization;on-chip interconnection;resource sharing;remote memory access