通信学报 ›› 2012, Vol. 33 ›› Issue (Z1): 157-164.doi: 10.3969/j.issn.1000-436x.2012.z1.020

• 学术论文 • 上一篇    下一篇

基于内存分块相异数据的虚拟机同步机制

廖剑伟,陈善雄,李莉   

  1. 西南大学 计算机与信息科学学院,重庆 400715
  • 出版日期:2012-09-25 发布日期:2017-08-03
  • 基金资助:
    西南大学博士基金资助项目;重庆市自然科学基金资助项目;国家自然科学基金资助项目

Memory contents patch based virtual machine synchronization

Jian-wei LIAO,Shan-xiong CHEN,Li LI   

  1. College of Computer and Information Science and Technology,Southwest University of China,Chongqing 400715,China
  • Online:2012-09-25 Published:2017-08-03
  • Supported by:
    The Scientific Research Fund for Doctor of Southwest University of China;The Natural Science Foundation of Chongqing;The National Natural Science Foundation of China

摘要:

定量分析了不同应用程序的内存分块数据相异部分,即某一阶段的内存有效改动页面分块与其他未变化内存分块的数据相异数据比例,提出基于内存数据相异部分的虚拟机同步机制,主虚拟机端通过基于地址和内容的散列函数表寻找与Dirty内存页面分块的最优匹配Non-dirty页面分块,相异数据通过XOR压缩后通过网络发送给备份虚拟机;备份虚拟机解码接收到的同步数据,重组在主虚拟机端的Dirty内存页面,从而完成备份虚拟机的状态同步操作。实验结果表明,与传统的标准异步方式相比,基于内存分块相异数据的虚拟机同步机制可以减少80%左右的同步操作带来的网络通信数据量,大大提高了某些基准测试程序的系统性能。

关键词: 检查点容错技术, 虚拟机同步操作, 内存分块相异数据, 地址内容散列表

Abstract:

A quantitative analysis has been conducted on memory contents difference for various high availability benchmarks.Based on this study memory contents patch based backup virtual machine(VM)synchronization technique was proposed,which finds the best match memory section for each dirty memory section;then compresses the difference contents in these two sections using hash based XOR compression technique in the primary VM; finally,it sends the packed data to the backup VM.For the backup VM part,it first decodes the received data and re-constructs the dirty pages,and then applies these pages to complete the VM synchronization.The experimental results show that compared with regular asynchronous replication,the proposed mechanism can reduce the synchronization data and network traffic by up to 80%,and then benefit to the synchronization to a great extent.

Key words: checkpoint-based fault tolerance, virtual machine synchronization, memory contents patch, address and content hash table

No Suggested Reading articles found!