Towards locality-aware DHT for fast mapping service in future Internet

The identifier/locator separation has been shown to be critical for the design of future Internet. A key aspect of the identifier/locator separation is to design an identifier-to-locator mapping service to map identifiers onto locators. Although several mapping services have been presented in previous works, they either are designed based on aggregable identifiers, or suffer from high resolution latency. That is, they hardly meet the demands of the future Internet, which is desired to support fast mapping and self-certifying flat identifiers. In this paper, we propose LMChord, a fast mapping service that is based on the idea of locality-aware and hierarchical Distributed Hash Table (DHT). To address the mismatch problem between overlay and physical network, we present the LMChord construction model, which models the LMChord construction process as a Markov decision process (MDP). Moreover, we present a Markov decision construction algorithm, which improves reinforcement learning to get the global optimal or near-optimal construction strategy. To further improve routing efficiency, we also modify the finger table to optimize the LMChord’s routing hops. We show that, besides the capability to support incremental deployment and flat identifiers, the mapping scheme is more scalable and has lower resolution latency. The evaluation also demonstrates the performance of our approach.