Research On Cooperative Optimization Techniques Of Video Delivery And Cache

In the triple-play network environment, there are more and more different types of services and business, in which the display service of video contents is one of the most important services. With the rapid development of the video service industry, video contents grow explosively.Video contents are delivered on the terminals such as TV、PC、mobile phone and PAD by the content delivery network. Video contents and terminal users increase rapidly; meanwhile, the network infrastructure develops slowly. To ensure the quality of video services, the distribution and cache of video contents are completed by the content delivery network (CDN) currently, which is constituted by content delivery and cache system. Much more requirements for video distribution and cache techniques are needed for large capacity, highly concurrent video services. Those accurate prediction of the video popularity, efficient cache replacement algorithm and replica placement algorithm influence the CDN performance greatly.In this thesis, the key techniques of video contents’ distribution and cache in the triple-play network environment are studied, which include the prediction of the video popularity in CDN, the cache replacement algorithm in distributed cache system and the replica placement algorithm in CDN based on cloud storage.The main contributions and innovations are as follows:1) We propose a prediction strategy for the video popularity based on the Bayesian network, to solve the problem that the video popularity has to be estimated artificially in CDN based-’Push’. Under the precondition that the users’ QoS should be satisfied, this strategy can reduce the adjustment of the video replicas and the burden of the system, and the storage resource of the CDN edge nodes can be used efficiently.This strategy use the Bayesian network to mining the data of video on demand to predict the popularity of video contents which should be distributed on the CDN edge nodes. The experiments demonstrate that the accuracy rate of the prediction is more than80%.2) We present a content replacement algorithm based on global information for distributed cooperative cache system. Theoretical derivation proves that the cooperative cache is better than the independent cache.The existing cache replica algorithm in cooperative cache system assumes that the size of file is uniform, and only considers local content information. To solve these two disadvantages, we build the distribution model of the video length and popularity, by analyzing the distribution of the video length and popularity in mobile environment. Then we propose a novel cache replacement algorithm named Value-Based Global algorithm (VBG), for the video services of the distributed cache system. The VBG algorithm both uses global content information and considers the files with different lengths. Numerical experiments show that the VBG algorithm has outstanding advantage in the bandwidth cost reduction compared with the local greedy algorithm, In the three experiments, the transfer cost of the local greedy algorithm is up to about2.5times of the VBG algorithm.3) We provide two offline replica placement algorithms:the GUCP (Greedy User Core Preallocation) algorithm and the PBP (Popularity Based Placement) algorithm, to satisfy the requirements of the video pre-deployment in CDN based on cloud storage.The GUCP algorithm uses the users’request information, places the replicas on the cloud storage nodes orderly, redirects the users to the nodes having the requested replicas, hence can solve the load imbalance problem caused by the GS (Greedy Sites) algorithm. Numerical experiments show that the local balance performance of the GUCP algorithm is much better than that of the GS algorithm. The load weight value of the GUCP algorithm is one eighth that of the GS algorithm, when the users’ number is1000. The PBP algorithm firstly calculates the deployment number of the replicas using the content popularity, and then redirects the users’requests to the cloud nodes which have content replicas. If the node does not have the replicas, it will copy replicas from other nodes which store the replicas. Compare the average delay of replace adjustment and the adjustment cost, we find the PBP algorithm is much better than the random pre-deployment algorithm. When the number of user increases, the adjustment cost of the PBP algorithm tends to be1/2that of the random algorithm, and the average delay is tends to be about3/4that of the random algorithm.This thesis relies on the project "The Development of Collaborative Supporting Environment for Operation, Management and Control of New Generation Services" of the national863major project "new generation of high-trusted network", and Huawei fund "Distributed hierarchical cache technology cooperative project". The research results of this thesis have been applied to these two projects.