Adaptive dynamic packet routing on internet networks based on reinforcement learning approach

In this thesis, we concern the problem of packet routing on the large scale networks like Internet which is a complex optimization due to a fast-growing, increasingly complex network of connected devices whereas the network models are conceptual. First, three synthesis Internet network models are proposed which are a random network, a random network with preferential attachment (PA) and a heuristically optimal topology (HOT) models. While Internet network models are constructed based on simplistic connections between nodes and connections formed sequentially by preferential attachment, the HOT model enhances to be more reflective of the Internet’s router level topology. Since, all traffic on the network has to be transmitted by traveling through interconnected routers. In addition, the volume of traffic has an effect on traffic congestion on different network connectivity as a result of complex routing optimization problems. Hence, Reinforcement learning (RL) is applied in this thesis because it has been introduced to solve complex and adaptive optimization problems. In particular, Q-routing which is an application of RL, is interested in the routing problem, but it is successful in only small various distributed wireless networks. Hence, the size of network is extended to be more realistic, and connectivity properties as seen in the Internet is represented as the aim of Q-routing on these networks is to support massive number of users. In addition, the Q-routing in this thesis is also applied on realistic network; JANET. Therefore, the results of Q-routing on the large scale network like Internet are represented by dealing with adaptive packet routing is embedded on all nodes in these networks which aims to optimize routing problem. Furthermore, the effect of the different network connectivity is also represented in how much the Q-routing can improve the network performance when the networks are subject to increasing amounts of traffic