LA-MDPF: A forwarding strategy based on learning automata and Markov decision process in named data networking

Abstract

Recently, Named Data Networking (NDN) has emerged as a leading paradigm among several Informa- tion-Centric Networking (ICN) schemes promising to address the shortcomings of TCP/IP architecture. Despite the great importance of the forwarding strategy in NDN, researches in this area are still somewhat limited. This paper presents a new strategy to improve the forwarding of request packets to the best interfaces, which improves the overall performance in NDN. Markov Decision Process (MDP) is employed to model this problem, where a router is presented as a decision-maker. To improve the performance of forwarding strategy, a Learning Automata (LA)-based algorithm is proposed, in which the experiences are considered in forwarding decisions. The proposed strategy (LA-MDPF) evaluates the previous decisions and responds by a penalty or a reward, affecting the probabilities of making the same decisions in the future. LA-MDPF can trace the best interface up to the original content or its cached replicas and overcome link failures and resource constraints. The LA-MDPF performance is evaluated by computer simulations using ndnSIM. Simulation results show that, the proposed algorithm improves content delivery performance in terms of network load balancing, average throughput, average packet drop, average Interest satisfaction ratio (ISR), and average content retrieval time at a reasonable computational cost compared to existing works.