Level compressed DAGs for lookup tables

Trie-based data structures for implementing IP lookups have attracted considerable research aUention. Techniques such as path compression, level compression, generalized level compression, and controlled prefix expansion are commonly used to implement lookup tables. [0 this paper, we present a fundamentally new technique that relies on directed acyclic graphs (DAGs). which, when coupled with generalized level compression, yield significantly better performance than existing techniques. Current im· plementations of trie-hased lookup tables utilize a route validation table in addition to a trie to enable fixed-length sub-prefix resolution to support path compression. This path validation enables us to merge different, partially filled subtrees to form full subtrees. The resulting DAGs introduce spurious routes that are eliminated in the validation phase. When combined with level compression (and generalized level compression), this slructure yields considerably shoder paths than existing approaches. In this paper, we describe transformation of tries to DAGs, algorithms for packing complementary subtrees, and profile performance of these algorithms and resulting improvements in lookup time. Specifically, we demonstrate, on actual lookup tables, performance gains cfup to 34% compared to LC-tries with minimal memory overhead (a little over 1%). Considering the fad that an LC trie is already a highly optimized structure, these gains are remarkable.