高级检索

    虚拟机快速全系统在线迁移

    Fast and Live Whole-System Migration of Virtual Machines

    • 摘要: 虚拟机全系统在线迁移对于广域网或者没有共享存储的局域网中的虚拟化平台非常重要.然而,全系统映像一般在几十吉字节(gigabyte, GB)量级,迁移这些数据需占用大量I/O带宽且时间开销很大.快速全系统在线迁移方法具有如下特点:文件系统感知的磁盘同步,利用文件系统的磁盘块分配位图,只拷贝实际使用的磁盘块,显著减少了第1阶段同步的磁盘映像数据量;异或压缩,利用映像数据的自相似性,快速有效地减少了迁移最后两个阶段拷贝的数据量;流水迁移,流水化迁移各个阶段,使得读写映像、压缩解压映像、网络收发各阶段充分并行.实验表明,与传统压缩迁移相比,快速迁移可以减少50%的迁移时间,最好情况和平均情况下,能分别降低21.68%和14.48%的停机时间.在网络带宽受限或虚拟机负载过重等极限条件下,快速迁移同样有效.

       

      Abstract: Live and whole-system migration of virtual machines is important for virtual platforms in wide-area network and local-area network, in which shared storage is not deployed. However, the size of whole-system image is often tens of gigabytes, and transferring such much data will occupy too much I/O bandwidth and have serious time overhead. A fast migration method is proposed, which includes three key technologies: file-system-aware block device migration, which utilizes the allocation bitmap of disk blocks in file system to make migration only copy used disk blocks, and reduces half of the disk image data which is transferred during the first migration phase; Xor-based compression, which utilizes the self-similarity of image data to effectively reduce the amount of transferred data in the last two migration phases with low time overhead; parallel migration, which parallelizes each migration phase and overlaps the cost of reading, writing, compressing, uncompressing, sending and receiving image data. Experimental results demonstrate that compared with traditional compression migration, the fast migration method can significantly reduce 50% of migration time, 21.68% of downtime at the best time and 14.48% of downtime on average. It can also speedup migration under extreme conditions, under which network bandwidth is limited or workload in the virtual machine is intensive.

       

    /

    返回文章
    返回