VM scaling based on Hurst exponent and Markov transition with empirical cloud data

• Using statistical measures to find out features for virtual machines (VM).
• Exploring empirical data of VMs for resource management in cloud environment.
• Hurst exponent and Markov transition to analyze the trend of resource usage for VMs.
• Effectively to predict and monitor the usage of VMs for resource allocation.

One of the major benefits of cloud computing is virtualization scaling. Compared to existing studies on virtual machine scaling, this paper introduces Hurst exponent which gives additional characteristics for data trends to supplement the often used Markov transition approach. This approach captures both the long and short-term behaviors of the virtual machines (VMs). The dataset for testing of this approach was gathered from the computer usage of key servers supporting a large university. Performance evaluation shows our approach can assist prediction of VM CPU usage toward effective resource allocation. In turn, this allows the cloud resource provider to monitor and allocate the resource usage of all VMs in order to meet the service level agreements for each VM client.