Dynamic Algorithms with Worst-Case Performance for Packet Classification

Packet classi cation involves | given a set of rules | nding the highest priority rule matching an incoming packet. When designing packet classi cation algorithms, three metrics need to be considered: query time, update time and storage requirements. The algorithms proposed to-date have been heuristics that exploit structure inherent in the classi cation rules, and/or trade o one or more metrics for others. In this paper, we describe two new simple dynamic classi cation algorithms, Heap-on-Trie or HoT and Binarysearchtree-on-Trie or BoT for general classi ers. The performance of these algorithms is considered in the worst-case, i.e., without assumptions about structure in the classi cation rules. They are also designed to perform well (though not necessarily the \best") in each of the metrics simultaneously.