Working Set Prediction for Live Virtual Machine Migration

Abstract

Live migration of virtual machines has been a powerful tool to facilitate system maintenance, load balancing, fault tolerance, and powersaving in clusters or data centers. In pre-copy approach that is mainly used in live migration, VM service downtime is expected to be minimal but total migration time is prolonged by iterative copy operations and the significant amount of transferred data during the whole migration process. In this paper, we present a framework that includes pre-processing phase, push phase and stop phase, for reducing the amount of transferred data in pre-copy based live migration. In pre-processing phase of the framework, we propose Least Recently Used (LRU) based working set prediction algorithm in physical memory of virtual machine. The proposed algorithm classify working set list that collects the most recent used memory pages and inactive list that collects the least recent used memory pages. Applying the proposed working set prediction algorithm, we can reduce the amount of iteration time of push phase and reduce the total migration time. From calculation result compared with pre copy based live migration that is not include pre-processing phase, the proposed framework can reduce the total migration time.