一种高效的虚拟机动态内存迁移方法

发布时间：2018-03-25 11:29

本文选题：内存迁移　切入点：页内脏偏移段　出处：《计算机科学》2016年04期

【摘要】：内存预拷贝是虚拟机动态迁移的主流策略,但是传统内存预拷贝算法是以脏页量为工作集,这对数据修改量小但在内存中分布较广的虚拟机环境来说,总体呈现迭代次数普遍较高、迁移总时间较长和迁移数据量较大的缺点。针对这一问题,提出了基于页内脏数据的预拷贝算法,该算法用以页为单位的脏数据作为工作集,并引入了新的数据结构记录页内脏数据。相比于以脏页为工作集粒度的传统内存预拷贝算法,该算法的工作集不仅能更准确地反映虚拟机的实际迁移环境,而且粒度明显细化,使得工作集小于预先设定的阈值的几率较大,从而可以减少迁移数据冗余,降低迭代次数,缩短迁移时间,降低迁移带宽。实验表明,该算法具有较高的虚拟机动态迁移效率。
[Abstract]:Memory precopy is the mainstream strategy of virtual machine dynamic migration, but the traditional memory precopy algorithm is based on the dirty page size, which is widely distributed in memory for the virtual machine environment with small data modification. The total number of iterations is generally high, the total migration time is longer and the amount of data transferred is large. In order to solve this problem, a precopy algorithm based on page visceral data is proposed, which uses dirty data per page as the working set. A new data structure is introduced to record page visceral data. Compared with the traditional memory precopy algorithm with dirty pages as the working set, the working set of the algorithm can not only reflect the actual migration environment of virtual machine more accurately, but also refine the granularity obviously. The probability of the working set being less than the pre-set threshold is larger, which can reduce the redundancy of the migration data, reduce the number of iterations, shorten the migration time and reduce the migration bandwidth. Experiments show that the proposed algorithm has a high dynamic migration efficiency of virtual machine.
【作者单位】：四川师范大学计算机科学学院;中国科学院计算技术研究所;
【基金】：国家自然科学基金(61373162) 四川省可视化计算与虚拟现实(KJ201402)资助
【分类号】：TP302

【相似文献】