Utilization based prediction model for resource provisioning

Abstract

Resource provisioning refers to the selection, deployment and management of resources to ensure guaranteed performance for the applications. Efficient resource provisioning is a challenging problem since it is dynamic in nature and requires supporting applications with different performance requirements. In order to provide adequate resources for applications with different requirements that must satisfy expected performance, it is required to predict correct set of resources. Towards this objective, a prediction model for resource provisioning has been developed in this work. The prediction model is trained by the dataset that is created using a benchmark e-Commerce application namely TPC-W that is deployed in Amazon EC2 environment. The experimental results show that the prediction model based on Linear regression exhibits 70 percentage of accuracy, Support Vector Regression shows 68 percentage of accuracy, whereas Multilayer perceptron exhibits 90 percentage of accuracy for the same dataset.