MDP and Machine Learning-Based Cost-Optimization of Dynamic Resource Allocation for Network Function Virtualization

The introduction of Network Functions Virtualization (NFV) enables service providers to offer software-defined network functions with elasticity and flexibility. Its core technique, dynamic allocation procedure of NFV components onto cloud resources requires rapid response to changes on-demand to remain cost and QoS effective. In this paper, Markov Decision Process (MDP) is applied to the NP-hard problem to dynamically allocate cloud resources for NFV components. In addition, Bayesian learning method is applied to monitor the historical resource usage in order to predict future resource reliability. Experimental results show that our proposed strategy outperforms related approaches.