Research On Schemes Of Latency Optimized Content Delivery

Content Delivery Network (CDN) is an efficient technology to mitigate the con-gestion caused by the concurrent access. As a result, it has gained wide attention from both the industry and academic communities. In a generic CDN, contents are replicated to multiple mirrored servers, which leads to fast access for users. Thus, the average latency of demands is greatly reduced. Since more mirrored servers bring much more expenses of deployment and maintenance, it is significant to deploy the mirrored servers reasonably. Moreover, as the capacity limits of the mirrored servers, the selection of content mirrors delivered to each servers directly determines the performance of CDNs.On the basis of the existing research results, I explore the techniques of reducing the average latency of demands from two aspects:the deployment of mirrored server-s and content delivery. This paper studies both static and dynamic content delivery mechanisms. Note that I design dynamic content delivery mechanisms, which utilize the features of wireless mesh network (WMN) and wireless converged network (WCN), respectively. The main contents and contributions include four aspects listed as below.1. Based on the statistic model of demands and service model of servers, it inves-tigates the techniques of the deployment of mirrored servers in CDNs. The motivation of our work is to reduce the average latency of demands, and then improve the perfor-mance of networks. The demands are modeled as independent Poisson streams, while the servers are modeled as M/M/I queuing systems. With the constraints of demand rate, service rate, maximum number of mirrored servers and upper bound of waiting time in a queue, the objective is to select a subset of the potential locations that minimizes the expected total number of demands traveling to and waiting at their closest servers such that the average latency in the CDN could be minimized. This paper formally defines the problem and proves its NP-hardness. Two heuristic algorithms includes tabu search and Lagrangian relaxation are designed for the problem. During the two algorithms, tabu search introduced a tabu list to avoid repetitive search, while Lagrangian relax-ation could reach excellent approximation ratio. According to the simulation results, both of the two algorithms are efficient in finding a solution with low average latency and could be applied to practical applications of CDNs.2. The contents are divided into several objects in CDNs. The popularity of each object is defined as the percentage of demands for this object among all demands. Based on the popularity, content objects could be delivered to the mirrored servers in advance. Thus, the demands from users are directed to the closest server who holds the required object. It takes both the popularity of objects and service capacity of servers into ac- count. With appropriate models, I formally define the problem of popularity based static content delivery. As NP-hardness, I propose two heuristic algorithms:greedy dropping and tabu search. The simulation results show that both of the two algorithms are efficient in finding a feasible solution with high probability. Especially, tabu search significantly decreases the average latency in CDNs.3. Aimed at dynamic content delivery in wireless mesh network, I propose a nov-el distributed four phase based content delivery algorithm, which utilizes the relation-ship between density and popularity of objects in WMN. First, based on the feature of WMN, the topology is partitioned so that the distributed algorithm could be employed. Then, I calculate the required number of each object according to the collected popu-larity information. The content delivery policy with low average latency is generated by a distributed algorithm based on the partitions in the first step. Finally, the repli-cas are delivered to mesh routers in a collaborative way. The simulation results show that the proposed algorithm has good performance on average latency without obvious increasement of network load.4. I study the push based dynamic content delivery in converged broadcasting and cellular network. In this network model, popular contents are pushed to all users over the broadcasting network. The others are delivered to users on demand via 4G/LTE cellular network. Upon such procedure of content delivery, I design efficient BCSA based content delivery scheme that abandons the method of broadcasting the most pop-ular objects in existing works. Instead, I apply approximate algorithm to the selection of broadcast objects, with the aim of maximizing the weighted bandwidth of demands for the broadcast objects. In the part of cellular network, I investigate the relationship between queue length and performances in average latency and packet loss rate. Final-ly, I conduct experiments for BCSA based content delivery scheme. According to the experimental results, the proposed content delivery scheme could reduce the bandwidth consumption and decrease the average latency.