Prediction Based Dynamic Configuration of Virtual Machines in Cloud Environment

Abstract

In cloud world software and computer infrastructure (virtual machine, network, storage etc) are given as metered service. In infrastructure providers (Amazon EC2, Rack Space etc.) The Virtual Machine (VM) size is static. Depending on the capacity of the VM the prices are fixed and pay as you go model is used. Since the VM size is static there is a high possibility of mismatch between application resource requirement and VM capacity. If the VM capacity is more than the application requirement, even though user is not utilizing the entire VM capacity he will be unnecessarily paying the extra money and also in this case resource utilization is not efficient. If the VM capacity is less than the application requirement then the application performance will degrade. With this motivation this work presents a prediction based dynamic configuration framework for virtual machines which predicts the required computing resources of the application and configures the VM as per the application requirement. To capture the characteristic of cloud application RAIN (which is a workload generation toolkit) has been used. To evaluate our prediction model we are using Olio which is a cloud benchmark application. With our approach an average error rate of 8% - 10% is observed.