EC2 instance provisioning for cost optimization

A cloud technology has become a popular workflow computing infrastructure. There is large development of large scale workflows having millions of tasks requiring thousands of computation hours. So there is need to optimize the cloud resource allocation. A decision must be taken on appropriate number of VMs along with the allocation of supplementary resources. Viable clouds may have different pricing plans. Provisioning problem thus becomes an NP-hard problem. Currently, many researchers are working on the cost optimization of cloud provisioning. However, most of them mainly considered the task scheduling. In this paper, we devised heuristic based on EDF where decision on number of instances required and pricing plans, and task scheduling are considered together within an optimization process. In order to get the better point for comparison, we find the cost performance by combining two pricing plans: on-demand and reservation. We performed task scheduling in order to minimize number of instances. In order to achieve the cost optimization goal, we applied threshold point calculation on each instance to decide its pricing plan. To make model more realistic we made detailed study of Amazon EC2 computing service. Our cost performance results are anticipated based on Amazon EC2 which can significantly reduce the total cost incurred to customer.