Research On Dynamic Adaptive Forwarding Strategy Based On Deep Reinforcement Learning In Named Data Networking

In recent years,network technology has developed rapidly,and terminal devices have increased dramatically.With the rapid increase of traffic and terminal,users have more demanding on the network.For such problems,the Named Data Network(NDN)was proposed as a future network architecture.NDN is different from the current mainstream network architecture in that it separates content from terminal location and provides various services through subscribe/publish.This will change the user's focus from the terminal to the content,that is,the user only needs to know what the content he wants is,not where the content is.The NDN router is the core device for NDN and used for forwarding,congestion control,traffic control,bandwidth control,and cache control.The forwarding plane of the NDN enables the NDN router to independently select the next hop to forward interest packet.The forwarding policy of the NDN is the decision maker in the router--determining whether the received interest packet is forwarded,when it is forwarded,and which port to forward to.Therefore,the research of forwarding strategy plays an important role in adaptive and efficient data transmission in NDN.The main work of this dissertation is as follows:1.Most of the traditional NDN forwarding strategies are designed based on certain theoretical models.There are many assumptions and simplifications of the network model,which cannot fully simulate the real environment of the network.Based on that,this dissertation proposes a dynamic adaptive intelligent forwarding strategy based on deep reinforcement learning.The intelligent forwarding strategy combines the NDN forwarding platform with intelligent algorithms and divides the forwarding plane into training phase and forwarding phase to guide the forwarding of Interest.The strategy collects network status information and content information of Interest as features for training during the training phase and uses the results for real-time forwarding phase.2.In the NDN,the Interest is forwarded in real time,so the intelligent forwarding algorithm should also have the ability to process forwarding at high speed.If forwarding strategy simply introduces the d deep reinforcement learning algorithm,it will take a lot of time to get forwarding decisions,which will affect the efficiency of forwarding.Therefore,this dissertation combines the characteristics of Interest forwarding and introduces a temporary forwarding table to optimizes the basic intelligent forwarding strategy and accelerates the decision of intelligent forwarding strategy.The improved adaptive intelligent forwarding strategy can well sense the network status and content information,and improve the interest packet forwarding efficiency.3.This dissertation builds a practical platform of a certain scale under the existing experimental equipment,and successfully deploys the dynamic adaptive forwarding strategy based on deep reinforcement learning.In addition,in order to evaluate the performance of the adaptive intelligent forwarding strategy,this dissertation selects the four popular forwarding strategies(Best-Route,Multi-cast,NCC,ASF)in NDN as the control strategy designing and implementing the corresponding comparison experiment.The experimental results show that the dynamic adaptive intelligent forwarding strategy based on deep reinforcement learning proposed in this dissertation has better performance in reducing packet loss rate and delay and improving port throughput.