Research On Content Caching System Mechanism For Smart Integration Identifier Network

With the rapid development of the Internet and the continuous growth of network equipment and application scale,the "static and rigid" original design of the traditional Internet is difficult to provide efficient,reliable,ubiquitous and other user needs.In this context,Smart Integration Identifier Network(SINET)innovatively puts forward the architecture of "three layers and three domains",which provides a new idea for realizing the intelligent perception and dynamic adaptation of network and service resources in a large-scale network environment.In SINET,the combination of caching technology and Network Function Virtualization(NFV)technology can effectively reduce network redundant traffic and improve user experience.However,the limitation of network node cache resources and the complexity of user requirements bring new challenges to the dynamic deployment of content.In order to solve this problem,this paper makes an in-depth study on the dynamic cache deployment based on the national key R & D plan "Research and Verification of Smart Integrated New Routing Switching Equipment"(2018YFE0206800).This paper mainly includes the following contents:Firstly,this paper proposes a centralized cache deployment scheme based on Deep Reinforcement Learning(DRL),which aims to minimize user delay and solve the problems of complex and changeable network environments and the dynamic heterogeneity of user requests.The scheme models the cache decision-making process and defines its state space,action space,and reward function.At the same time,a cache deployment algorithm based on long-term system performance benefits is developed,which optimizes the content placement by using Deep Q-Network(DQN)and designing the corresponding neural network structure.Then,this paper designs a content caching system based on SINET and completes the construction of resource adaptation layer knowledge domain.It includes a cache decision module,cache deployment module,information collection module,information storage module,and request analysis module.The information collection module collects node resource information and network status information through the NFV controller and Programming Protocol-independent Packet Processors(P4)language,inputs them to the cache decision module for training and calculates the cache strategy.The cache deployment module calls the NFV controller and starts the virtual cache server at the corresponding node according to the cache policy.The request parsing module parses the content requested by the user and queries whether it is hit in the information storage module.Finally,this paper tests the performance and functionality of the proposed centralized cache deployment scheme and content cache system in the simulation environment and SINET prototype system respectively.The test results show that each module can achieve its functions.Compared with the distributed DQN,LFU,and LRU algorithms,the centralized cache deployment scheme proposed in this paper improves the hit rate by 3.1%,11.7%,and 18.6%,respectively.In terms of request delay,it is reduced by 43.1%,45.4%,and 47.6%,respectively.The above results verify the feasibility and superiority of the designed content caching system.