Performance and power modeling and evaluation of virtualized servers in IaaS clouds

Abstract

In this paper, Stochastic Activity Networks (SANs) are exploited to model and evaluate the power consumption and performance of virtualized servers in cloud computing. The proposed SAN models the physical servers in three different power consumption and provisioning delay modes, switching the status of the servers according to the workload of the corresponding cluster if required. The Dynamic Voltage and Frequency Scaling (DVFS) technique is considered in the proposed model for dynamically controlling the supply voltage and clock frequency of CPUs. Thus, Virtual Machines (VMs) on top a physical server can be divided into several power consumption and processing speed groups. According to the workload of the system and the number of waiting requests, the proposed SAN decides to scale up or down the VMs, so it helps the overall system to save power when it still preserves satisfiable performance. After modeling the servers and VMs using SAN formalism, some performance related measures together with the power consumption metric are defined on the proposed model. The results obtained by solving the proposed SAN model configured with real data show the prominence of the proposed model in comparison with some baselines and previously proposed models.