A Genetic-Algorithm Solution for Designing Optimal Forwarding Tables

A high speed IP address lookup engine with a reasonable memory cost is a key factor for designing a router at wire speed. This paper proposes a hardware solution that performs each IP address lookup in a few number of memory accesses with minimum amount of memory requirement. Our solution is based on dividing destination IP address into several segments. For finding the optimum address segmentation that minimizes the memory consumption, a genetic-algorithm solution is employed. The genetic program uses benchmark forwarding tables for finding the optimum points of address segmentation. The final result is a small forwarding table for the local traffic of the router. This table can be reconfigured along the time when the local traffic gradually changes. The proposed method can fit a forwarding table of size 130000 routing prefixes in about 1.5 MB of memory with only four memory accesses for each lookup search.