Scalable IP lookups using shape graphs

Haoyu Song, Murali Kodialam, Fang Hao, T.V. Lakshman
2009 2009 17th IEEE International Conference on Network Protocols  
Recently, there has been much renewed interest in developing compact data structures for packet processing functions such as longest prefix-match for IP lookups. This has been motivated by several factors: (1) The advent of 100Gbps interfaces necessitating correspondingly fast packet processing algorithms with a compact memory footprint; (2) network virtualization leading to virtualization of physical router platforms making it critical to reduce high-speed memory needs per virtual router; (3)
more » ... irtual router; (3) software routers built on multi-core processors requiring the use of compact data-structures that fit in on-chip caches for good performance. In this paper, we revisit this issue of developing compact data structures for key packet-processing functions. We develop a new data structure, called the shape graph, that significantly compacts the trie data-structure used for IP lookups. We accomplish this by identifying considerable structural similarities in IP lookup tries that have not previously been used in the literature for scalable IP lookups. We use these similarities to store lookup tries in a new graph data structure that has a significantly lower memory-footprint. Using real IP forwarding tables, we compare the memory usage of this new data structure to that of multi-bit tries and of Bloom filters used for IP lookups. The shape graph requires significantly less memory and allows the far more effective use of on-chip memory. This effective use of on-chip memory combined with multi-threading on a multi-core processor makes shape-graph-based IP lookups well suited for 100Gbps lookups. The small footprint also makes it well suited for use in router platforms that host a large number of virtual routers.
doi:10.1109/icnp.2009.5339697 dblp:conf/icnp/SongKHL09 fatcat:lv2dsmj2hzhkfhtemt65b2es3q