Survivable SFC deployment method based on federated learning in multi-domain networks

Abstract

Abstract In the multi-domain network scenario, in order to improve the survivability of service function chain in the face of network failure, most methods solve this problem through virtual network function backup mechanism. However, the traditional multi-domain SFC deployment method lacks a SFC partition mechanism for backup resource consumption, and does not consider the isolation and privacy requirements between different network domains. In view of the above problems, this paper proposes a reliability partition scheme based on reinforcement learning in SFC partition stage, which can ensure that VNF is backed up while maintaining good load balance and low inter-domain transmission delay, and improve the reliability of SFC. Then, this paper proposes a VNF backup mechanism with minimum resource fluctuation in the VNF mapping stage, and uses the ILP model to determine the backup scheme of each VNF, so as to ensure the minimum fluctuation of resource occupancy of the entire network. Finally, this paper proposes a multi-domain SFC deployment and backup algorithm based on Federated learning (FA-MSDB). Each domain is trained locally to improve the load balance of node resources and reduce the transmission delay in the domain, the global and local models are updated periodically to deploy and backup VNF requests. The experimental results indicate that compared with other three version of FA-MSDB and other three benchmark algorithms, FA-MSDB can effectively improve the survival rate of SFC, reduce the overall transmission delay, and ensure good inter-domain and intra-domain load balance.