A Prediction-Based Multi-Objective VM Consolidation Approach for Cloud Data Centers

Virtual machine (VM) consolidation aims to run VMs on the least number of physical machines (PMs). The optimal consolidation significantly reduces energy consumption (EC), quality of service (QoS) in applications, and resource utilization. This paper proposes a prediction-based multi-objective VM consolidation approach to search for the best mapping between VMs and PMs with good timeliness and practical value. We use a hybrid model based on Auto-Regressive Integrated Moving Average (ARIMA) and Support Vector Regression (SVR) (HPAS) as a prediction model and consolidate VMs to PMs based on prediction results by HPAS, aiming at minimizing the total EC, performance degradation (PD), migration cost (MC) and resource wastage (RW) simultaneously. Experimental results using Microsoft Azure trace show the proposed approach has better prediction accuracy and overcomes the multi-objective consolidation approach without prediction (i.e., Non-dominated sorting genetic algorithm 2, Nsga2) and the renowned Overload Host Detection (OHD) approaches without prediction, such as Linear Regression (LR), Median Absolute Deviation (MAD) and Inter-Quartile Range (IQR).