Scalable pipelined IP lookup with prefix tries

While the throughput demand for a backbone router keeps increasing constantly, both routing and switching of packets are facing tough challenges for running faster. Currently, the prefix trie-based routing algorithms for IP lookup are playing a key role in building high performance routing systems. We propose in this article a scalable routing architecture for pipelining any prefix trie-based routing algorithm that links nodes by pointers. All the prefixes in a routing table are classified into the short and the long categories to organize as a quick table and a set of prefix tries, respectively, where the tries' nodes are evenly distributed into multiple memory blocks using a randomization technique. Each arriving packet finds its route information by looking up the quick table at first and then traverses a corresponding prefix trie if necessary. A routing buffer is employed to cache packet arrivals that suffer memory access contentions when visiting nodes of the prefix tries. The scheduling of queuing packets to access the memory blocks is modeled as a bipartite matching problem. The performance of the proposed architecture is evaluated by investigating the memory requirements and the maximum throughput with comparison to the linear and the circular pipelines. The results for both the analytical queuing model and the simulation experiments indicate that this architecture can be a potential candidate for building high bandwidth routing engines of backbone routers.