Working Set Prediction using LRU cache with Splay Tree Algorithm in Live VM Migration

Abstract

Live virtual machine (VM) migration provides great benefits for load balancing, power management, fault tolerance and other system maintenance issues in modern data centers. VM live migration basically consists of transferring its memory pages from a source server to a destination server. The amount of transferred memory pages affected the downtime and application performance of virtual machines. In pre-copy approach in live VM migration, total migration time is prolonged the significant amount of transferred data during the whole migration process. In this paper, we propose a working set prediction algorithm which combine LRU (Least Recently Used) cache with splay tree algorithm. Deploying splay tree algorithm with LRU, the proposed algorithm predict the memory pages of the most recent process and define as the working set. Applying the proposed algorithm, we can reduce the amount of transferred memory pages during migration process by transferring the working set in last round. Experiment demonstrates that compared with XEN’s default pre-copy based migration algorithm, the proposed algorithm can reduce 23.67% of the total transferred memory pages during migration process.