A Multi-Dimensional Framework for Virtual Machine Consolidation

Cloud computing is a demand computing model that requires a large number of physical resources and provides services based on the request of each user. A large number of physical servers in data centers have high electrical energy consumption, which causes high operating costs, increases carbon dioxide (CO2) emission. The focus of this paper is on virtual machine consolidation to minimize power consumption, the number of VM migrations and, reducing service level agreement violation. In contrast to the existing works that use CPU utilization for the detection of host overload, a recent study has proposed Multiple Regression Host Overload Detection (MRHOD) and Hybrid Local Regression Host Overload Detection (HLRHOD) algorithms which take multiple factors (CPU, memory, and network bandwidth utilization) into consideration. This paper provides a framework that takes into account multiple factors: CPU, memory, and bandwidth utilization in three terms: host overload detection, VM placement, and service level agreement violation. First, in the host overload detection term, we provide a Separately Local Regression Host Overload Detection (SLRHOD) algorithm that considers CPU, memory, and bandwidth utilization, separately. Second, in terms of VM placement which is an NP-hard problem, the Power Aware Best Fit Decreasing (PABFD) algorithm with consideration of Dot-product (DP) heuristics was proposed. Third CPU and memory take into account the calculation of SLA violation in terms of SLA violation. To evaluate our framework in contrast to existing works, many experiments were performed. For each experiment, we evaluated and compared three objectives, namely energy consumption, service level agreement violation, and the number of VM migrations. Our experiment results show that the Separately Local Regression Host Overload Detection (SLRHOD) algorithm in terms of SLA violations reveals a significant improvement of 80\%. On the other hand, the Separately Local Regression Host Overload Detection (SLRHOD) algorithm saves energy, up to 3\%, compared to the HLRHOD and MRHOD algorithms. Our simulation results show that our proposed algorithm outperforms the existing algorithm and achieves improvements in energy consumption, service level agreement violation, and the number of VM migrations.

A Multi-Dimensional Framework for Virtual Machine Consolidation Keywords:

Cloud Computing , Green Computing , Virtual Machine , Quality of Service , Energy Efficiency