Intelligent Forwarding Strategy Based on Online Machine Learning in Named Data Networking

Abstract

The content-oriented model of Named Data Networking (NDN) allows consumers to pay more attention to the targeting data itself instead of the location of where the data is stored. Different from IP, NDN has a unique feature that forwarding plane enables each router to select the next forwarding hop independently without relying on routing. Therefore, forwarding strategies play a significant role for adaptive and efficient data transmission in NDN. Existing forwarding strategies are not smart enough to cope with the complexity of network and diversity of application demands. This paper presents an intelligent forwarding strategy, which integrates online machine learning method into the optimization of interface probabilities during forwarding process. Originally, a probabilistic binary tree structure is proposed to abstract the forwarding process as a path selection process traversing from the root node to the leaf node, which provides theoretical support for machine learning and reduces the complexity of forwarding process. In addition, we improved our strategy to prevent the convergence into limited local optimal solution by adopting the idea of simulated an nealing. Experimental results show that the proposed strategy can reduce time complexity, as well as achieve higher throughput, better load balance and lower packet drop rates in comparison with other existing forwarding strategies. The drop rates are reduced by 60% and 34% respectively in different scenarios compared with BestRoute, a strategy widely used in NDN.