Resource provisioning under uncertainty in cloud computing

Abstract

In this thesis, we mainly focus on the resource provisioning in cloud computing. Resources can be provisioned from cloud providers to cloud consumers through two options, i.e., reservation and on-demand. The reservation option is cheaper and able to guarantee the availability and prices of resources. However, a cloud consumer has to purchase the reservation option with prior commitment for specific resources. Due to uncertainties, the common problems encountered in resource provisioning with the two options are overprovisioning and underprovisioning. In this thesis, we consider different uncertainties in the resource provisioning problems, i.e., uncertainties of resource demand, resource price, power price, and availability of resources. For our major contributions, we propose the resource provisioning algorithms and framework to deal with the uncertainties for three cloud stakeholders, namely cloud consumer, cloud provider, and cloud retailer. The contributions are as follows: First, we propose novel algorithms for a cloud consumer to provision resources from cloud providers. The algorithms can minimize the expected resource provisioning cost incurred by overprovisioning and underprovisioning of resources, while the uncertainties are taken into account. We formulate optimization models to obtain the optimal solution for the algorithms. The models are derived by stochastic programming with two- and multi-stage recourse so that the optimal solution from the algorithms can be applied for long-term resource provisioning plans. We also apply the robust optimization to handle the impact of the uncertainties on the optimal solution. The performance evaluation shows that the proposed algorithms have the lowest resource provisioning cost when they are compared with other well-known algorithms. To reduce the computational complexity of the algorithms, we also apply Benders decomposition and sample-average approximation methods.