Network bandwidth aware dynamic automated framework for Virtual Machine Live Migration in cloud environments

Live migration is a very important feature of virtualisation, a running VM can be seamlessly moved between different physical hosts. The source VM’s CPU state, storage, memory and network resources can be completely moved to a target host without disrupting the users or running applications. Live VM migration is an extremely powerful tool in many key scenarios such as load balancing, online maintenance, proactive fault tolerance and power management. There are four steps involved in the live VM migration, the setup stage, memory transfer stage, VM storage transfer stage and the network clean up stage. The most important part of live VM migration is transferring the main memory state of the VM from the source to the destination host which can consume a significant amount of network bandwidth in a short period of time. Modern cloud based data centres generate a significant amount of network traffic apart from VM live migration traffic. If VM migration occurs during a peak time, VM migration and user traffic will compete for network bandwidth, then the data centre’s network may not have enough resources to support both VM migration and demands of application users, which would create a bottleneck in the network. Therefore, this research presents a centralised, bandwidth aware, dynamic, and automated framework for live VM migration in Cloud environments. The proposed framework adopted a heuristic approach, and it provides guaranteed bandwidth for VM live migration by controlling user traffic on the network while scheduling live VM migration in an efficient manner. The framework consists with two main components, The Central Controller and the Local Controller. The Local Controller is responsible for collecting resources usage data from VMs and PMs however the Central Controller makes global management decisions. The Central Controller is based on four algorithms which are called a migration policy. The migration policy contains the following algorithms: the host overloaded detection, host underloaded detection, VM selection and VM placement algorithms which are proposed in this research. The proposed migration policy has been implemented in CloudSim and evaluated against two benchmark migration policies in CloudSim. Five evaluation metrics have been used in the simulation to evaluate the performance of the proposed migration policy. The results reveal that the proposed migration policy outperformed the two benchmark policies