SR-FIB: A NewIP Forwarding Information Base via Neural Networks

Aiming at the problem of fast indexing and efficient storage of addresses in the SRv6 protocol of NewIP network, a forwarding information base (FIB) that supports fast retrieval of IPv6 addresses was designed, called SR-FIB. The neural network was used to learn the distribution characteristics of IPv6 addresses in the memory and build an index mapping model that can achieve uniform address mapping. The storage structure and retrieval algorithm that can effectively support the mechanism of longest prefix matching were designed. The simulation results show that SR-FIB can achieve rapid retrieval of IPv6 addresses under the SRv6 protocol. Under the requirement of 1% network packet loss rate, the memory consumption is only 26.14 MB and the throughput is 11.48 MPPS, which can meet the requirements of network retrieval.