Research On ICN Cache Management Technology Based On Content Popularity

With the proliferation of massive amounts of data content and users and connected devices in the network,TCP/IP-based networks are revealing their design flaws.Information-Centric Networks(ICNs)are designed with the goal of increasing network throughput and reducing latency.In-network caching is an inherent feature shared by all proposed ICNs.However,due to the limited resources of the network and the secure delivery of cached data,there are still some challenges in the study of caching scheme.Storing the most popular content in cache nodes to improve cache hit rates,reduce the number of data forwarding hops and reduce user request latency is one of the effective solutions that are widely adopted today.The complex and highly dynamic state of ICN networks makes them difficult to model,predict and control and,moreover,the requested content to be responded varies with different characteristics,distribution and requirements.An ideal ICN caching scheme should make the best possible use of relevant information such as content information,network state and user requirements to achieve optimal selection,and be able to adaptively adjust the decision cache content for dynamic scenarios.Therefore,in this thesis,the following research is conducted with the goal of improving caching performance,using popularity as the key point.Firstly,reinforcement learning is considered as a good tool in the design of caching algorithms because of its interactive nature with the environment.To address the problem of modelling the need for a fixed data set in popularity-based caching approaches,a reinforcement learning-based caching scheme,RLPC is proposed.the scheme follows the basic ideas of reinforcement learning and models caching in an ICN network environment,the processing of packets,content popularity offsets as a state space,action space and rewards,based on feedback dynamically adjusting the actions.In this case,the intelligent body plays the role of exploring actions,taking corresponding caching actions for content that is expected to be popular,and will be rewarded with higher rewards when the content has higher popularity.Through the continuous adjustment of actions by the intelligences,the nodes will be filled with more popular content and the nodes maintain a high cache hit rate.Finally,this paper verifies the effectiveness of this caching scheme in improving cache hit rate and reducing of eviction operations through simulations on the ndn Sim platform using real network topologies.Since router nodes have limited cache space,it is then useless to pay any attention for the content which is unpopular,as this content has no chance of being cached.A more effective approach would be to focus on content with high popularity that influences caching decisions.As for different nodes,they have different sets of popular content,and using this property,this paper designs a caching policy based on popularity hierarchy with topological weights.The policy considers managing the cached content in nodes with a hierarchy of popularity and improving their distribution in terms of the importance of the nodes’ position in the network.Finally,simulation experiments are conducted by different parameter settings under several real topologies to confirm the feasibility of the scheme.