Adaptive VNF Scaling Approach with Proactive Traffic Prediction in NFV-enabled Clouds

Cloud computing and network function virtualization (NFV) are emerging as key technologies to meet different service requirements of users. With NFV, service providers can flexibly deploy network services on clouds, significantly improving user experience and network scalability. Due to the flow fluctuation and demand dynamic of users, most existing resource allocation methods are reactive and may lead to high delay and resource consumption. In this paper, we propose an adaptive NFV resource allocation approach with proactive traffic prediction. First, a prediction model based on Long Short-Term Memory (LSTM) is designed to predict user demands. Next, with predicted demands, we construct a cost minimization model of virtual network function (VNF) scaling and flow routing, and design a proactive VNF scaling method combining horizontal scaling and vertical scaling. Then we utilize Markov approximation method to approximate VNF scaling problem, and design a polynomial algorithm based on Markov chain to find the near-optimal solution. The simulation shows that the proposed approach is more efficient and save up to 29% of cost with the better convergence.