项目名称: 虚拟化数据中心内存资源预测与动态调配
项目编号: No.61272158
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 罗英伟
作者单位: 北京大学
项目金额: 84万元
中文摘要: 内存资源对于整个数据中心的性能有着显著的影响。在虚拟化数据中心内,由于大量虚拟机并发运行并且虚拟机在不同时刻对内存资源的需求是动态变化的,如何在单个服务器以及整个数据中心内进行内存资源的动态调配,使得分配给虚拟机的物理内存大小能够满足其上应用程序不断变化的内存需求,是提高内存资源利用率和提升整个数据中心综合性能的关键。本项目主要研究一种预测式的内存资源管理方法,对虚拟机的内存资源进行在线的监控和预测,并根据预测结果动态调配虚拟机的内存资源。这里,研究的关键是低开销、高精确度的内存工作集在线跟踪与预测方法,及面向单机和全局的动态内存调配策略。我们拟利用硬件计数器,对传统的内存预测机制进行扩展和改进,研究并实现一个低开销、高精确度的内存工作集跟踪机制。在获得了精确的内存需求之后,我们将通过气球技术、远程内存技术以及虚拟机在线迁移技术来实现单机和全局的内存资源的动态调配。
中文关键词: 数据中心;工作集;内存管理;缓存管理;虚拟化优化
英文摘要: Memory system has a significant impact on the performance of a data center for cloud computing. In a virtualized data center, multiple virtual machines share each individual physical host and thus share the physical memory. Due to the dynamic memory demand of virtual machines, it is advised to allocate memory for each virtual machine on the fly accordingly in order to improve the overall performance of the data center. This on-demand memory allocation requires accurate memory demand prediction that yields a minimal overhead. We propose a predictive, dynamic, center-wide memory balancer that adjusts memory across the whole data center. Our research starts with the design and implementation of a low overhead, highly accurate, on-the-fly memory demand predicator which improves over conventional memory demand predictors and is based on on-line monitoring with hardware performance counters. With the accurate memory prediction, we further propose techniques that adjust memory allocation across the whole data center using ballooning, remote memory, and live migration.
英文关键词: Data Center;Working Set Size;Memory Management;Cache Management;Virtualization Optimization