Efficient Trie Braiding in Scalable Virtual Routers

Many popular algorithms for fast packet forwarding and filtering rely on the tree data structure. Examples are the trie-based IP lookup and packet classification algorithms. With the recent interest in network virtualization, the ability to run multiple virtual router instances on a common physical router platform is essential. An important scaling issue is the number of virtual router instances that can run on the platform. One limiting factor is the amount of high-speed memory and caches available for storing the packet forwarding and filtering data structures. An ideal goal is to achieve good scaling while maintaining total isolation among the virtual routers. However, total isolation requires maintaining separate data structures in high-speed memory for each virtual router. In this paper, we study the case where some sharing of the forwarding and filtering data structures is permissible and develop algorithms for combining tries used for IP lookup and packet classification. Specifically, we develop a mechanism called trie braiding that allows us to combine tries from the data structures of different virtual routers into just one compact trie. Two optimal braiding algorithms and a faster heuristic algorithm are presented, and the effectiveness is demonstrated using the real-world data sets.