A unified data structure of name lookup for NDN data plane

Abstract

NDN data plane relays name-based packets by maintaining three tables: Content Store, Pending Interest Table and Forwarding Information Base. The three tables require similar but different schemes to be matched and updated in a nearly per-packet fashion, thus individual data structure is required for each table. In this work, we propose a unified data structure of name lookup for all three tables, namely CTrie, aiming at reducing the computational cost from three pipelined lookup rounds down to one unified round. CTrie extends the original Patricia trie to a combinational trie structure built from both component-based and byte-based hierarchical names. We compared CTrie with other approaches in speed and memory. The results show that CTrie runs 3.2 times faster and consumes about 38% memory than the current ones in terms of the whole data plane. CTrie fits for all application scenarios of NDN and especially well for IoT like lightweight-deployed scenarios.