Supervised Machine Learning-based Routing for Named Data Networking

Named Data Networking (NDN) ambitions the rank of Future Internet Architecture in uniquely addressing content items by their name. In NDN, routers forward Interests for content after finding Longest-Prefix Matches (LPM) of content names in their Forwarding Information Base (FIB). However, the scalability of this structure is challenged by the huge global Internet namespace. In this paper, we propose a novel approach to interest forwarding that compresses the FIB data structure into Artificial Neural Networks (ANNs). A bitwise trie splits the namespace and indexes ANNs. ANNs are offline trained by the control plane from the Routing Information Base and matching Interests. Then, they are made available to the data plane for interrogation. We demonstrate that this approach accelerates packet forwarding by several orders of magnitude. Noteworthily, leveraging ANNs as memory and processor for directing packets towards next hops reminds of Asking For Directions to people in the street, incurring similar reliability regards.