SDNFV-based routing service composition model

Abstract

With the rapid development of Internet, varieties of new types of applications are emerging out. Accordingly, the user communication demands for diversified applications exhibit more and more complex characteristics. The conventional approach to deal with this issue for Internet Service Provider (ISP) is continuous purchasing of new physical network equipment, which inevitably causes high technical costs and operating expense. Fortunately, the paradigms of decoupling control plane from data plane in Software Defined Networking (SDN) and decoupling functional services from underlying physical equipment in Network Function Virtualization (NFV) bring significant insights to deal with this challenging issue. Accordingly, with appropriately reusing diversified software-based routing functions and then adaptively selecting them to compose customized routing services, a novel SDNFV (i.e., SDN and NFV) based routing service composition model is proposed. In addition, considering continuously generated information by large-scale network communication activities, we combine machine learning with the proposed model. According to the user feedbacks for the provided services, the appropriate routing function selection and service composition is trained and optimized by the method of multi-layer feed-forward neural network. Simulation results verify the feasibility of the proposed model.