当前位置: X-MOL 学术Big Data Res. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
A Service Sustainable Live Migration Strategy for Multiple Virtual Machines in Cloud Data Centers
Big Data Research ( IF 3.3 ) Pub Date : 2021-02-25 , DOI: 10.1016/j.bdr.2021.100213
Anurag Satpathy , Manmath Narayan Sahoo , Ashutosh Mishra , Banshidhar Majhi , Joel J.P.C. Rodrigues , Sambit Bakshi

Virtual machine (VM) migration is an indispensable aspect of a virtualized cloud environment. It assists in resource management by dynamically relocating VMs from one physical machine to another. This is an essential aspect especially for big data applications that are prone to variable workloads and often demand relocation of resources in terms of VMs. However, such applications not only experience stochastic workloads but also have stringent requirements on the maximum tolerable latency. To address such issues, VMs are often relocated using live VM migration. VM migration is associated with overheads, hence, in this paper, we propose a modified serial migration strategy to migrate multiple VMs based on the pre-copy live migration technique. We propose to interleave the pre-copy stages in such a way that a balance is achieved between the migration time and downtime overheads. The proposed technique is compared with the state-of-the-art serial, parallel, and improved serial migration strategies. Concerning downtime, the proposed approach performs exceptionally well compared to both serial and improved serial methods. The downtime of the proposed scheme and parallel are comparative for read-intensive applications (low dirtying rates). However, for write-intensive applications (high dirtying rates) the former significantly outperforms the latter. The migration time performance of the proposed scheme is observed to be much better than that of the parallel technique and is slightly higher than those of serial and improved serial techniques.



中文翻译:

云数据中心中多个虚拟机的服务可持续生存实时迁移策略

虚拟机(VM)迁移是虚拟化云环境必不可少的方面。它通过将虚拟机从一台物理机动态重定位到另一台物理机来辅助资源管理。这是一个必不可少的方面,特别是对于容易产生可变工作负载并且经常需要根据VM进行资源重定位的大数据应用程序而言。但是,这样的应用程序不仅经历随机的工作量,而且对最大可容忍的延迟有严格的要求。为了解决此类问题,通常使用实时VM迁移来重新定位VM。VM迁移与开销相关联,因此,在本文中,我们提出一种改进的串行迁移策略,以基于预复制实时迁移技术迁移多个VM。我们建议以在迁移时间和停机时间开销之间达到平衡的方式交错预复制阶段。将提出的技术与最新的串行,并行和改进的串行迁移策略进行了比较。关于停机时间,与串行方法和改进的串行方法相比,该方法的执行效果非常好。对于读取密集型应用程序(低脏污率),建议方案的停机时间与并行运行时间比较。但是,对于写密集型应用程序(高脏率),前者的性能明显优于后者。观察到所提出的方案的迁移时间性能比并行技术好得多,并且比串行技术和改进的串行技术略高。将提出的技术与最新的串行,并行和改进的串行迁移策略进行了比较。关于停机时间,与串行方法和改进的串行方法相比,该方法的执行效果非常好。对于读取密集型应用程序(低脏污率),建议方案的停机时间与并行运行时间比较。但是,对于写密集型应用程序(高脏率),前者的性能明显优于后者。观察到所提出的方案的迁移时间性能比并行技术好得多,并且比串行技术和改进的串行技术略高。将提出的技术与最新的串行,并行和改进的串行迁移策略进行了比较。关于停机时间,与串行方法和改进的串行方法相比,该方法的执行效果非常好。对于读取密集型应用程序(低脏污率),建议方案的停机时间与并行运行时间是可比较的。但是,对于写密集型应用程序(高脏率),前者的性能明显优于后者。观察到所提出的方案的迁移时间性能比并行技术好得多,并且比串行技术和改进的串行技术略高。与串行方法和改进的串行方法相比,该方法的执行效果非常好。对于读取密集型应用程序(低脏污率),建议方案的停机时间与并行运行时间是可比较的。但是,对于写密集型应用程序(高脏率),前者的性能明显优于后者。观察到所提出的方案的迁移时间性能比并行技术好得多,并且比串行技术和改进的串行技术略高。与串行方法和改进的串行方法相比,该方法的执行效果非常好。对于读取密集型应用程序(低脏污率),建议方案的停机时间与并行运行时间比较。但是,对于写密集型应用程序(高脏率),前者的性能明显优于后者。观察到所提出的方案的迁移时间性能比并行技术要好得多,并且比串行和改进的串行技术要好一些。

更新日期:2021-03-02
down
wechat
bug