Energy-Aware Profiling and Prediction Modelling of Virtual Machines in Cloud Computing Environments

Cloud Computing has changed the way in which individuals and businesses use IT resources. Instead of buying their own IT resources, they can use the Cloud services offered by Cloud providers with reasonable costs based on a “pay-per-use” model. With the wide adoption of Cloud Computing, the costs for maintaining the Cloud infrastructure have become a vital issue for the providers, especially with the large amount of energy being consumed to operate these resources. Hence, the excessive use of energy consumption in Cloud infrastructures has become one of the major cost factors for Cloud providers. In order to reduce the energy consumption and enhance the energy efficiency of Cloud resources, proactive and reactive management tools are used with consideration of physical resources’ energy consumption. However, these tools need to be supported with energy-awareness not only at the physical machine (PM) level but also at virtual machine (VM) level in order to make enhanced energy-aware decisions. As the VMs do not have physical interface, identifying the energy consumption at the VM level is difficult and not directly measured.
This thesis introduces an energy-aware Cloud system architecture that aims to enable energy-awareness at the deployment and operational levels of a Cloud environment. At the operational level, an energy-aware profiling model is introduced to identify energy consumption for heterogeneous and homogeneous VMs running on the same PM based on the size and CPU utilisation of each VM. At the deployment level, an energy-aware prediction framework is introduced to forecast future VMs’ energy consumption. This framework first predicts the VMs’ workload based on historical workload patterns, particularly static and periodic, using Auto-regressive Integrated Moving Average (ARIMA) model. The predicted VM workload is then correlated to the physical resources within this framework in order to get the predicted VM energy consumption.
The evaluation of the proposed work on a real Cloud testbed reveals that the proposed energy-aware profiling model is capable of fairly attributing the physical energy consumption to homogeneous and heterogeneous VMs, therefore enabling energy-awareness at the VM level. Compared with actual results obtained in this testbed, the predicted results show that the proposed energy-aware prediction framework is capable of forecasting the energy consumption for the VMs with a good prediction accuracy for static and periodic Cloud application workload patterns.
The application of the proposed work is providing energy-awareness which can be used and incorporated by other reactive and proactive management tools to make enhanced energy-aware decisions and efficiently manage the Cloud resources. This can lead towards a reduction of energy consumption, and therefore lowering the cost of operational expenditure (OPEX) for Cloud providers and having less impact on the environment.