Gaussian Process Regression-Based Traffic Prediction and Rate Coordination for Service Chain Congestion Mitigation

With the rapid development of Internet technologies, the network traffic continues to grow exponentially, which leads to serious congestion and makes the service quality difficult to guarantee. Network function virtualization (NFV) has been proposed to ease such a challenge by sharing resource among different virtual network function (VNF). However, such resource sharing among VNFs would worsen traffic congestion and the waste of work due to the processing rates mismatch between upstream and downstream VNFs of the same service chain. Therefore, this work focuses on coordinating the VNF processing rates to jointly avoid resource waste and ease traffic congestion. Specifically, the rate balance process is implemented with two parts which are the rate prediction and matching, during which the backpressure and Gaussian process regression are applied to quickly locate the rate mismatch position and formulate the relationship between network resource and traffic rate respectively. Then, the future traffic rate is predicted and applied to adjust the allocated resource of VNFs in one service chain for mismatch rate coordination. The experimental results indicate that the proposed method eases the traffic congestion and resource waste efficiently and outperforms the other methods in terms of delay, backlog, throughput, etc.