[论文笔记] Live Gang Migration of Virtual Machines (HPDC, 2011)

Timespan: 2.1 – 2.2
Umesh Deshpande, Xiaoshuang Wang, and Kartik Gopalan. 2011. Live gang migration of virtual machines. In Proceedings of the 20th international symposium on High performance distributed computing (HPDC '11). ACM, New York, NY, USA, 135-146.. (gs:11)

    作者Umesh Deshpande是Binghamton University的博士生，师从Kartik Gopalan。Kartik Gopalan的研究兴趣是"experimental computer systems, primarily in Resource Virtualization, Computer Networks, Security, and Distributed Systems”。
    他们课题组发表的与本文相关的论文列表：

• Umesh Deshpande, Brandon Schlinker, Eitan Adler, and Kartik Gopalan, Gang Migration of Virtual Machines using Cluster-wide Deduplication, To appear in 13th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid), Delft, The Netherlands, May 2013.
• Umesh Deshpande, Unmesh Kulkarni and Kartik Gopalan, Inter-rack Live Migration of Multiple Virtual Machines, Proc. of the 6th International Workshop on Virtualization Technologies in Distributed Computing, Delft, The Netherlands, June 2012.
• Umesh Deshpande, Beilan Wang, Shafee Haque, Michael Hines, and Kartik Gopalan, MemX: Virtualization of Cluster-wide Memory, In Proc. of 39th International Conference on Parallel Processing (ICPP), San Diego, CA, USA, September 2010.
• Michael Hines, Umesh Deshpande, and Kartik Gopalan, Post-Copy Live Migration of Virtual Machines, In SIGOPS Operating Systems Review, Volume 43, Number 3, pages 14--26, 2009.
• Michael Hines and Kartik Gopalan, Post-Copy Based Live Virtual Machine Migration Using Adaptive Pre-Paging And Dynamic Self-Ballooning, In the International Conference on Virtual Execution Environments (VEE), Washington DC, March 2009. (Among four best papers at VEE 2009 selected for publication in OSR Special Issue).

以下是论文笔记

1. live gang migration
是指将部署于同一台物理机器的一组VM并发进行动态迁移（“simultaneously migrating a group of co-located and live virtual machines”，“simultaneous migration of mutiple active VMs from one physical host to another”）。

这篇论文研究如何对“live gang migration”进行性能优化（total migration time, network traffic overhead, service downtime)，主要使用了de-duplication、differential compression等技术，原型平台基于QEMU/KVM实现、通过改进已有的单VM迁移pre-copy方法而得。

2. 本文在研究live gang migration时实现和比较的相关技术：(S1)

• offline and online de-duplication: at the page and sub-page granularities
• re-hashing: 在memory transfer时针对dirtied pages进行
• differential compression: 针对nearly identical memory content
• offline and online compression of uniform pages: 针对uniform pages

实验对比结果表明：“offline de-duplication mode at page granularity” 能够最大幅度降低total migraiton time和network overhead。

3. uniform page: 是指该页含有的数据全领0或全1（文中没有给出确切定义，猜测如此）(S2)
QEMU/KVM上现有单VM迁移的实现针对uniform pages会先压缩再传输。

4. (S2.2)将live gange migration分成了两个阶段：

• preparation: 定期扫描“memory of co-located VMs”，使用content hashing来识别相同的内存页。第一轮会对每个VM的每个memory page进行扫描，后续的扫描只针对dirtied pages。
• migration: 对于identical pages，第一次传输整个page content，之后再运到只需要传输page identifier。

5. (S2.3)讨论了针对dirtied pages的re-hashing
re-hash的触发时机有两种：

• periodic re-hashing: 在periodic memory scans进行
• online re-hashing: 在VM迁移时进行

6. (S2.4)介绍了针对部分相似的pages，如何定量确定相似度？
文中采用了【Gupta10】的HashSimilarityDetector(k, s, c)方法