面向虚拟化云服务器的智能高速缓存管理

With cloud computing models gaining significant momentum, data centers are increasingly employing virtualization as a means to support a large number of heterogeneous workloads running simultaneously on a multicore server. In such environments, contention for shared cache space can have a destructive effect on performance isolation among virtual machines. However, the existing technology ignores the variety of tenant's service requirements and the inherent characteristics of virtual machines. To this end, we will finish the following research contents in this proposal: 1. The inter-VM elastic cache architecture to optimize cache SLA, including the definition of cache SLA metric, and the design of cache capacity partition with guarantee of SLA. 2. The intra-VM data aware cache management to optimize the locality of shared and private data types under parallel workloads, including the devise of utility monitor hardware and the design of implicit cache partition mechanism based on the insertion of incoming cache lines. 3. The individual cache bank with set cooperation to optimize the access latency of relocated VM, including the estimation of set pressure, and the retention policy of relocated VM working sets. The quality-of-service and performance of last level cache in virtualized cloud servers will be significantly improved through the study of this subject. We will publish 8 to 10 high quality papers on international conferences and domestic or foreign journals, in which 6 to 8 will be indexed by EI or SCI, and finally apply for 1 to 2 patents.

随着云计算的发展，当代数据中心日渐使用虚拟化技术来支持大量异构负载在多核服务器上的并发执行。然而虚拟机间对共享高速缓存的隐式竞争会破坏虚拟机的性能隔离，现有技术又忽略租户多样化的服务需求及虚拟机固有的特征。为此本课题将重点完成以下研究内容：研究面向高速缓存SLA优化的虚拟机间可伸缩Cache结构，包括高速缓存SLA度量的定义，可保证SLA的Cache容量划分算法的设计；研究面向并行负载中共享私有数据局部性优化的虚拟机内数据感知Cache管理策略，包括数据效用监控硬件的设计，基于数据插入位置控制的隐式划分算法的设计；研究面向重定位虚拟机高速缓存访问延迟优化的组间协作单体Cache结构，包括Cache组访问压力的剖析，重定位虚拟机工作集保留策略的制定。本课题将极大改善虚拟化环境下高速缓存的服务质量和性能，预期在国内外期刊和国际会议上发表8-10篇高水平论文，并申请1-2项专利。

随着云计算的发展，当代数据中心日渐使用虚拟化技术来支持大量异构负载在多核服务器上的并发执行。然而虚拟机间对共享高速缓存的隐式竞争会破坏虚拟机的性能隔离，现有技术又忽略租户多样化的服务需求及虚拟机固有的特征。为此本课题资助期内重点完成了以下研究内容：研究面向高速缓存SLA优化的虚拟机间可伸缩Cache结构，包括高速缓存SLA度量的定义，可保证SLA的Cache容量划分算法的设计；研究面向并行负载中共享私有数据局部性优化的虚拟机内数据感知Cache管理策略，包括数据效用监控硬件的设计，基于数据插入位置控制的隐式划分算法的设计；研究面向重定位虚拟机高速缓存访问延迟优化的组间协作单体Cache结构，包括Cache组访问压力的剖析，重定位虚拟机工作集保留策略的制定。本课题的研究成果极大改善虚拟化环境下高速缓存的服务质量和性能，在国内外期刊和国际会议上发表12篇高水平论文，包括CCF A类会议和期刊论文各1篇，有5篇论文被SCI索引，申请5项专利。