Cost Optimization in Cloud Computing

Abstract

In recent years, cloud computing has increased in popularity from both industry and academic perspectives. One of the key features of the success of cloud computing is the low initial capital expenditure needed compared to the cost of planning and purchasing physical machines. However, owners of large and complex cloud infrastructures may incur high operating costs.

In order to reduce operating costs and allow elasticity, cloud providers offer two types of computing resources: on-demand instances and reserved instances. On-demand instances are paid only when utilized and they are useful to satisfy a fluctuating demand. Conversely, reserved instances are paid for a certain time period and are independent of usage. Since reserved instances require more commitment from users, they are cheaper than on-demand instances. However, in order to be cost-effective compared to on-demand instances, they have to be extensively utilized.

This thesis focuses on cost optimization of cloud resources by balancing on-demand and reserved instances. The challenge is to find an optimal resource allocation under uncertainty. In order to solve the problem, this study introduces a theoretical model based on Inventory Theory and a heuristic-based implementation for reserved instances optimization.

The inventory theory model provides a theoretical framework for cost optimization. In addition, the model describes a mathematical method to solve the optimization problem. The heuristic-based implementation analyzes the cloud infrastructure of a company and proposes a purchase plan of reserved instances. The implemented system validates the theoretical finding.

In order to evaluate the proposed approaches, this work describes a set of experiments, using simulations and data from an industry case. The experiments demonstrate the effectiveness of the reserved instances optimizer and the validity of the theoretical model.