Dynamic Multicast-oriented Virtual Network Function Placement with SFC Request Prediction

This paper studies the multicast-oriented virtual network function placement (MVNFP) problem in a dynamic environment, which considers the prediction of multicast service function chain requests (MSRs). The end-to-end delay, computing resource consumption, and request processing delay are considered in the objective function, with bandwidth and computing resources constrained. An online dynamic MVNFP algorithm with two stages, called ODMVP, is proposed to address the problem above. In the pre-placement stage, a bidirectional long short-term memory with attention and incremental window update mechanisms is used to predict the incoming MSRs. Grey wolf optimizer is then adopted to pre-place VNFs and determine the corresponding routes based on the prediction result. In the repair stage, a fast greedy VNF placement scheme is developed to handle the actual arrival MSRs that are not successfully predicted. Experimental results show that the ODMVP outperforms other state-of-the-art algorithms in terms of prediction, solution quality, and response.