Reinforcement learning inspired forwarding strategy for information centric networks using Q‐learning algorithm

Abstract

SummaryContent interest forwarding is a prominent research area in Information Centric Network (ICN). An efficient forwarding strategy can significantly improve data retrieval latency, origin server load, network congestion, and overhead. The state‐of‐the‐art work is either driven by flooding approach trying to minimize the adverse effect of Interest flooding or path‐driven approach trying to minimize the additional cost of maintaining routing information. These approaches are less efficient due to storm issues and excessive overhead. Proposed protocol aims to forward interest to the nearest cache without worrying about FIB construction and with significant improvement in latency and overhead. This paper presents the feasibility of integrating reinforcement learning based Q‐learning strategy for forwarding in ICN. By revising Q‐learning to address the inherent challenges, we introduce Q‐learning based interest packets and data packets forwarding mechanisms, namely, IPQ‐learning and DPQ‐learning. It aims to gain self‐learning through historical events and selects best next node to forward interest. Each node in the network acts as an agent with aim of forwarding packet to best next hop according to the Q value such that content can be fetched within fastest possible route and every action returns to be a learning process, which improves the accuracy of the Q value. The performance investigation of protocol in ndnSIM‐2.0 shows the improvement in a range of 10%–35% for metrics such as data retrieval delay, server hit rate, network overhead, network throughput, and network load. Outcomes are compared by integrating proposed protocol with state‐of‐the‐art caching protocols and also against recent forwarding mechanisms.