Research On Optimization Methods For Content Placement In Cloud Content Distribution Networks

With the development of cloud computing,cloud based content delivery networks(CCDNs)have emerged.CCDNs provide users with services in a more flexible manner by deploying low-cost cloud servers.Compared with traditional content delivery networks(CDNs),CCDNs are the main measures for content delivery which are able to save a lot of operating costs.The issues of content placement are important for CCDNs,which determine how to delivery data to edge servers in a better way to reduce content delivery costs.However,with the development and popularity of the Internet,the number of Internet users continues to grow,and the demand for network content continues to increase as well.Thus,the cost of content delivery is increasing day by dayand the content placement optimization method of CCDNs is facing enormous challenges.There are many problems in the existing content placement optimization methods for CCDNs.First of all,for the content placement optimization method,the existing work has the disadvantages of large energy consumption and poor scalability.Secondly,the traditional content placement method only provides a fixed delivery path,but it cannot adapt to the dynamic characteristics of frequent changes of cloud proxy servers in CCDNs.In addition,it does not consider the dynamic changes of network congestion in CCDNs.Therefore,the traditional content placement method is no longer applicable to the dynamic CCDNs.To solve the above problems,we first propose an energy efficiency delivery model based on multicast tree,which can reduce the path length and overall energy consumption of replica delivery.Then,we propose a Q-learning based content placement method for dynamic CCDNs,which can adapt to dynamic changes of CCDNs and reduce the overall congestion cost of the network effectively.The main work and innovation of this paper are as follows:(1)To reduce the length of content placement path and the energy consumption,an energy efficiency delivery model(EEDM)based on multicast tree is constructed.Firstly,the K-Canopy algorithm is used to determine the number of key nodes;Next,the K-Means clustering algorithm is used to divide the whole network into k regions;Then,the key nodes in the k regions are identified as members of the multicast group through the key node selection algorithm;Finally,a delivery tree is constructed to connect all multicast members based on a replica placement multicast routing algorithm so as to minimize the cost of distribution.(2)In order to reduce the cost of content placement in CCDNs and adapt to the dynamic characteristics of CCDNs,we propose a Q-learning based content placement model for dynamicCCDNs,called Q-content placement model(Q-CPM).Then,on the basis of Q-CPM model,the Q-adaptive delivery tree(Q-ADT)is constructed by using adaptive delivery tree construction algorithm.By learning the data packet to propagate congestion information,the algorithm chooses paths with low congestion cost to distribute data.The proposed algorithm can adapt to the dynamic cloud content delivery network well.The experimental results show that the delivery tree based on the proposed EEDM model has lower energy consumption.It proves that the proposed energy efficiency delivery model can save energy costs effectively.Furthermore,our proposed Q-CPM model is adapted to dynamic characteristics of CCDNs and can make better routing decisions while reducing congestion costs.