Name lookup is an essential function but a performance bottleneck in both today's and future network architectures. Trie is an excellent candidate data structure and has been widely used for looking up and updating names. However, the granularity of trie-at bit, byte (character), or component level-can dramatically affect the network performance in terms of memory usage and packet-processing speed, which has not yet been studied adequately. To fill this gap, we first show that the choice of trie's granularity for name lookups and updates (i.e., insertions and removals) is not a trivial problem due to the complex performance tradeoffs involved. We also introduce a new tool, called NameGen, which uses a Markov-based name learning model and generates pseudo-real datasets with different tunable name characteristics. We compare different trie granularities based on a collection of datasets and performance metrics, highlight the strengths and weaknesses of each granularity, and draw a conclusion on the choice of granularity. Surprisingly, our experimental evaluation finds that there are only two key rules to choose the proper trie's granularity for any kind of dataset: 1) bit-level trie is the choice when the memory requirement is a real concern and 2) character- and component-level tries are preferred for faster lookups and updates when dealing with names composed of short and long components, respectively.

Name lookup is an essential function, but a performance bottleneck in both today's and future network architectures. Variable-length and unbounded names rather than fixed-length addresses, as well as much larger and more dynamic forwarding tables call for a careful re-engineering of lookup structures for fast, memory-efficient, and scalable packet forwarding. We propose a novel data structure, called NameTrie, to store and index forwarding table entries efficiently and to support fast name lookups and updates. Its novelty lies in the optimized design and implementation of a character-trie structure. The nodes of NameTrie are stored compactly, improving cache efficiency and speeding up packet processing. Its edges are implemented using a hash table, facilitating fast name lookups and updates. A new scheme is used to encode control information without consuming additional memory. Running on conventional commodity hardware and using large-scale real-world name datasets, our implementation of NameTrie in software achieves 2.82~3.56, 3.48~3.72, and 2.73~3.25 million name insertions, lookups, and removals per second, respectively, for various datasets while requiring a small memory footprint. We have conducted a comprehensive performance evaluation against the state-of-the-art of named data networking (NDN) as a typical use-case. It is shown to require at least 35% less memory and runs at least 3x faster for name table lookups and updates than two well-known trie-based schemes in NDN.