Application-aware Multi-path Network Resource Allocation System Based On SDN

The development of Internet of Things has gradually transformed the"Internet of Everything"into a reality from a beautiful technological vision.The massive interconnection of sensing devices poses new challenges for traditional network architectures.First,IoT network equipments which carry different applications are numerous.These network applications may have different needs for network resources.The second is the heterogeneity of the network.Different access methods mean that the link status of the network is different and always changing.To intelligently schedule network resources in such a complex network situation,There are two main problems need to be solved:the first is how to determine the network resource requirements of the specific application;the second is how to allocate network resources for the specific application based on the current network resource status.For the first problem,the methods used in traditional traffic engineering mainly include:traffic identification technology based on data header information such as ports,DPI technology,and traffic identification technology based on protocol behavior mode.In this paper we deeply investigates the limitations of traditional traffic identification methods and proposes a traffic recognition algorithm based on machine learning technology.By selecting appropriate traffic characteristics,it solves the problems encountered by traditional recognition algorithms.For the second problem,this paper takes advantage of the separation of the SDN architecture which separates the control and forwarding plane,and implements a traffic scheduling algorithm based on traffic type and current network resource status in the control plane.NFV is used to solve traffic sampling problems in SDN networks in this paper,it can further decouple the traffic identification function and network resource scheduling algorithm,which improves the robustness of the system.The results show that the traffic classification model in the system can achieve the identification rate of 91.10%.Compared with other traditional scheduling strategies,the SDN-based application-aware multi-path network resource scheduling system proposed in this paper can make full use of multi-path transmission resources in the network to ensure the personalized resource requirements of different services.