A cloud eco-system : reactive demand control and dynamic pricing methodology

Resources are limited in capacity. In the meanwhile, over-provisioning of resources resulted in server low utilization could be costly to cloud providers. The underlying reasons of the low utilization are multiple-folds, such as uneven application fit where the application cannot fully utilize the resources allocated or the uncertainty in demand forecasts that is introduced by the dramatically varied demand of the cloud resource between peak and non-peak periods. While many research works are devoted to optimize the resource allocation techniques in the effort of achieving higher server utilization, how to control resource demand so the correct level of resource provisioning can be determined has become the next research question. In this thesis, we introduce a pricing methodology with dynamic pricing that intended to induce desired demand pattern and enhances the revenue of a cloud provider. The proposed pricing methodology encourages cloud tenants, whose requested Virtual Machines (VMs) can be allocated easily, to use more cloud service by offering them lower prices and discouraging cloud tenants, whose requested VMs are difficult to allocate, from using cloud service by charging them higher prices. We study our pricing methodology with a combinatorial optimization algorithm, the Knapsack Algorithm and show that the overall revenue is enhanced through evaluations. Then, to achieve fairness among users, we further perform a case study of our pricing methodology with a multi-resource allocation fairness algorithm, the Heterogeneous Dominant Resource Fairness (DRFH) algorithm. Trace-driven simulation results show that the proposed pricing methodology with DRFH can increase the overall revenue by up to 11.60%. Furthermore, we propose a novel cloud federation system that is cognitive to the dynamic prices as a decision making assistant tool for our pricing methodology. The cloud federation system automatically selects and migrates user tasks to a cloud system that is charging at a more affordable rate. We discuss the architectural framework and platform design, provide a mathematical formulation and investigate a total service cost minimization approach with privacy constraints. Simulation results demonstrate the proposed system can lower the cost of cloud services by exploiting the advantages of dynamic prices of multiple clouds.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat