A Novel Multi-objective Optimization Scheme for Rebalancing Virtual Machine Placement

Abstract

In cloud computing, load balancing is a very important performance factor. Frequent addition and removal of Virtual Machines (VMs) can cause load imbalance across Host Machines (HMs). Therefore, redistribution of VMs to different HMs needs to be performed periodically. This is called VM load rebalancing (VMrB). Existing VMrB solutions consider either balancing across HMs (inter-HM) or balancing within each individual HM (intra-HM), but not both. In this paper, we give a systematic review of existing VMrB literature, and present a VMrB scheme named MOVMrB that optimizes load balancing of multiple resources across HMs and in each individual HM. Our solution is a true multiple-objective optimization approach that does not scalarize multiple resources into one measurement. Instead, we treat each resource as a separate dimension and convert the load balancing of multiple resources to a complex system optimization problem. In addition, we propose a hybrid VM live migration algorithm that can dramatically speed up the VMrB process. Extensive evaluations using synthetic and real data have been conducted to test the proposed solutions against existing VMrB studies. The results show that our scheme can achieve a more balanced inter-HM and intra-HM load in a more efficient way. To the best of our knowledge, this is the first approach that considers both inter-HM and intra-HM load balancing and does not scalarize multiple resources into one measurement.