Int J Performability Eng ›› 2020, Vol. 16 ›› Issue (5): 728-737.doi: 10.23940/ijpe.20.05.p6.728737

• Orginal Article • Previous Articles     Next Articles

Improved Post-Copy Live Migration with Memory Page Prefetching

Yong Cui*(), Haoran Chen,  and Liang Zhu   

  1. School of Computer and Communication Engineering, Zhengzhou University of Light Industry, Zhengzhou, 450002, China
  • Submitted on ; Revised on ; Accepted on
  • Contact: Yong Cui
  • Supported by:
    This research is supported by the National Natural Science Foundation of China (No. 61906175 and 61902361).


IaaS cloud computing data centers always employ live migration of virtual machines (VM) to achieve dynamic scheduling and management of IT resources. Pre-copy and Post-copy are two prevalent live migration algorithms. Both have pros and cons, and they work well only in suitable scenarios. Contrary to Pre-copy, Post-copy performs better with memory writing workloads running in VM to be migrated. Post-copy switches the VM to the target host immediately once the live migration is launched, and then it requests the source host to transfer memory pages whenever the VM accesses a nonexistent page. Although this Post-copy scheme can guarantee that every memory page needs to copy only once and the migration time is predictable, it can cause a mass of network requests for fault pages, which can lead to a prolonged downtime for VM and a degraded live migration performance. This paper proposes an improved Post-copy scheme with memory page prefetching, which leverages PPM (Prediction by Partial Match) to build a prediction model for memory page accessing and predict the pages to be read soon. Once a network page fault happens, this model determines the following pages to be accessed. Then, this faulted page together with these predicted pages are sent to the target host to avoid subsequent page faults. Experiment results show that this proposed scheme can allow for accurate page prediction and prominently decrease the network page faults and downtime for VM.

Key words: live migration, Post-copy, virtual machine, memory page prefetching, performance optimization