IGP Convergence Optimization with AI

This paper develops an Interior Gateway Protocol (IGP) convergence optimization strategy and proposes an AI-based method to predict a temporary Routing Information Base (RIB) table with high accuracy to be used for packet forwarding while the exact RIB is being computed using a traditional approach, for example Dijkstra algorithm. The proposed method consists of AI-model and the training procedure and considers RIB prediction as a classification task when each Destination in a network is assigned to a class representing corresponding Next Hop. In this work, the complete IGP convergence optimization strategy is implemented and tested in a physical environment for 2 different IGP protocols, namely OSPF and IS-IS. As in traditional IGP process, the AI model takes Link State DataBase (LSDB) and incoming Link State Advertisement (LSA) messages as the input and synthesizes a temporary RIB in the required internal format which is used then for packet forwarding until traditional IGP process finishes exact RIB recomputation. The proposed method is implemented and tested in a physical testbed environment emulating the networks with up to 10000 network elements (routers). The performance evaluation shows methods’ ability to predict RIB table with high accuracy and speed, leading to faster IGP convergence which is observed as 2 to 5 times faster packet-loss elimination in the network after a link failure accident.