摘要：

针对内存预拷贝过程中迁移时间较长和内存页反复重传的特点，改进传统的内存动态迁移机制，引入马尔科夫预测模型，提出基于脏页概率预测的工作集测定算法。利用脏页的历史操作访问情况预测其下一轮迭代被修改的概率，只传输预测概率较低的页。实验结果表明，该算法缩短了迁移总时间和停机时间，能有效支持虚拟机动态迁移。

关键词: 内存迁移, 虚拟机, 虚拟机动态迁移, 马尔科夫模型, 可写工作集

Abstract:

Aiming at longer time and memory pages repeated retransmission in the process of memory pre-copy, this paper optimizes the mechanism of memory pre-copy migration and uses Markov prediction model to improve the algorithm that reckon the working set of memory dirty page, designs an new algorithm that calculate working set of memory dirty page by forecasting the probability of dirty pages. This algorithm calculates probability of being modified next round of iteration using dirty pages history of the operation visits, only the memory pages with a lower probability can be translated. Experimental results show that new algorithm shortens the total time of migration and downtime, and effectively support dynamic migration of Virtual Machine(VM).

Key words: memory migration, Virtual Machine(VM), VM live migration, Markov model, Writable Working Set(WWS)

中图分类号: 

• TP317.1