A Distributed Virtual-Machine Placement and Migration Approach Based on Modern Portfolio Theory

Abstract Virtual machine placement and migration (VMPM) are key operations for managing cloud resources. Considering the large scale of cloud infrastructures, several proposals still fail to provide a comprehensive and scalable solution. A variety of approaches have been used to address this issue, e.g., the modern portfolio theory (MPT). Originally formulated for financial markets, MPT enables the construction of a portfolio of financial assets in order to maximize profit and reduce risk. This paper presents a novel VMPM approach applying MPT and incremental statistics computation for VMPM decision-making so as to maximize resource usage while minimizing under and overload. Extensive simulation experiments were conducted using CloudSim Plus, relying on synthetic data, PlanetLab and Google Cluster traces. Results show that the proposal is highly scalable and largely reduces computational complexity and memory footprint, making it suitable for large-scale cloud service providers.