Approximately-perfect hashing: Improving network throughput through efficient off-chip routing table lookup

IP lookup is one of the key functions in the design of core routers. Its efficiency determines how fast a router can forward packets. As new content is continuously brought to the Internet, novel routing technologies must be developed to meet the increasing throughput demand. Hash-based lookup schemes are promising because they have low lookup delays and can handle large routing tables. To achieve high throughput, we must choose the hash function to reduce the lookup bandwidth from the off-chip memory where the routing table is stored. The routing table updates also need to be handled to avoid costly re-setup. In this paper, we propose AP-Hash, an approximately perfect hashing approach that not only distributes routing-table entries evenly in the hash buckets but also handles routing table updates with low overhead. We also present an enhanced approach, called AP-Hash-E, which is able to process far more updates before a complete re-setup becomes necessary. Experimental results based on real routing tables show that our new hashing approaches achieve a throughput of 250M packets per second and perform re-setup as few as just once per month.