A heuristic adaptive threshold algorithm on IaaS clouds

Due to the wide applications of IaaS (Infrastructure as a Service), energy-saving technologies of IaaS clouds has attracted much attention. However, it is very difficult for IaaS cloud providers to guarantee both of energy saving and performance under the condition of satisfying SLA (Service Level Agreement). Recently, adaptive-threshold-based methods are proposed to relieve the trade off between energy saving and satisfying SLA, however, high variable workloads have to be conducted. Thus, a more energy-saving method with lower workloads is desired. In this paper, in order to adaptively discover optimal thresholds, we propose a novel workload prediction-based framework, which seamlessly integrates a feature-selection-based prediction method and a model of measuring the relationship between the energy cost of the migration of virtual machine (VM) and the power incomes when the physical machine (PM) shuts down. Furthermore, a threshold discovering algorithm is designed to dynamically capture reasonable thresholds effectively. Finally, we verify the efficiency and effectiveness of the proposed methods through extensive experiments on Cloud Sim based on Google workload trace data set, and show the significant performance improvement compared with existing techniques. For instance, the proposed methods can improve the energy consumption by 10-20 percents.