Optimizing Virtual Machine Live Migration in Distributed Edge Servers Based on Hybrid Memory

摘要

In order to meet the users' request for high time-efficiency, edge computing migrates services to edge nodes closer to users through the virtual machine (VM) migration technologies. However, migrating VMs from an edge server to another edge server is done under the WAN with weak network conditions, so the migration will take more time than within the data center. In addition, Hybrid memory employing both Dynamic Random Access Memory (DRAM) and Persistent Memory (PM) is increasingly being adopted by edge servers, but lack support for efficient management in existing virtualization technologies. VM migration generates many random write operations on the destination node's memory which causes a decline in memory performance and brings huge wear to the PM. To address these challenges, this paper presents a virtual machine live migration method in distributed edge clouds based on hybrid memory. In this method, we modified the pre-copy algorithm to reduce the migration time, which transferring the cold pages before hot pages through the memory access feature. The virtual persistent memory (VPM) is used to mapping the physical hybrid memory to VM, and then the algorithm will move cold and hot pages to the PM area and DRAM area of the destination VM. Experimental results with several main workloads in edge computing show that our method can improve the performance of the live migration and memory management in destination VM with graceful degradation of CPU performance.