Dynamic pricing scheme for IaaS cloud platform based on load balancing: A Q-learning approach

In the era of cloud computing, Infrastructure-as-a-Service (IaaS) cloud providers can provide their various resources as virtual machine instances, which will later be allocated to users. The two main challenges they face are to set optimal prices and to improve utilization level of computing resources. In this paper, we formulate these challenging problems with dynamic programming approach involving customer utility function with service quality gap model and consumer inertia under discrete finite horizon Markovian decisions. Combining with the advantages of load balancing in resource allocation, we develop a novel Dyna-f Q-Learning approach to obtain the optimal solution for dynamic pricing problems. Numerical illustrations show that our proposed algorithm is more efficient than conventional method whether in service pricing or resource allocation.