A Simulation-based Migration Algorithm Minimizing the Number of Migrations in Server Virtualization Environments

Migration systems in server virtualization environments distribute workloads and solve resource shortage problems by moving a virtual machine in hotspot to a physical server having residual resources. Most of the migration systems use first fit algorithm known as an efficient solution for bin packing problem to determine migration policies. However, the algorithm cannot provide an optimal migration scenario, because the resource usage of virtual servers continuously changes. Consequently, the migration systems based on the algorithm issue more migrations and increase the cost of the physical servers. In this paper, we propose a simulation-based migration manager producing an optimal migration policy for its own window by simulating the forecasted CPU usages of virtual machines. The simulation-based migration algorithm assumes that every physical server with enough residual capacity to serve a victim can be a candidate destination and simulates a migration on each server. The algorithm finally selects a migration scenario whose expected number of migrations is the lowest. The system consequently, provides applications with high quality of service through accurate analysis and reduces the migration cost by minimizing the number of migrations.