| After entering information age,various network applications have sprung up in an endless stream,and the functions and structures of Internet have become increasingly complex.In order to alleviate problems faced by traditional network architecture,Software-Defined Network(SDN)emerged as an innovative network architecture with separation of data layer and control layer.However,there still bring new problems: how to deploy controllers reasonably to enhance network performance while paying the minimum cost;how to efficiently calculate deployment strategies to adapt to dynamic changes on large-scale network topologies.In response to the above problems,this paper proposes two distributed controller deployment management strategies.(1)In SDN,due to uneven distribution of data streams,controllers should be more inclined to be deployed on nodes with higher data stream density.We propose a distributed controller deployment algorithm based on node weights(NWDP).The node weights are measured according to request flow density of nodes and their own degrees.To improve computational efficiency,some nodes with larger weights are extracted as controller candidate set.Then,the appropriate controller deployment strategy is determined by minimizing the network propagation latency and deployment cost.The simulation results show that NWDP algorithm can minimize the flow establishment cost between the switch and the controller,and improve the quality of network service.(2)We propose a Graph Convolution Network(GCN)model based on traffic characteristics(FF-GCN).By extracting the structural features and traffic features of the network topology as input,through multi-layer graph convolution layer.The multi-modal structure between nodes are captured,and FF-GCN model with higher accuracy is generated for the prediction of the nodes classification to realize a more reasonable distributed deployment of controllers.Experiments show that the model can achieve 80% prediction accuracy in a large-scale network(702 nodes,811 edges). |
