Research On Some Key Technologies Related To The QoS Of Content Delivery Networks

With the rapid development of network technology, the increasing large number of Internet users and diverse web applications highlight the bottleneck of traditional network architecture. Then the CDN (Content Delivery Networks) technology is brought to deal with the problem. And owning to the increasing of the number of the CDN operators, the competetion between them is very intense. And how to ensure the QoS(Quality of Service) of CDN is the key problem for all the CDN operators.According to the different roles of service user and provider, the QoS of CDN contains two aspects. On one hand, QoS is aiming at improving the service experience of the users. In order to do that, such things need to be done like deliverying the Internet content in the proper way according to the limitations of the content providers, and improving the utilization efficiency of network bandwidth. On the other hand, QoS means to improve the quality of CDN service with controllable costs of the CDN operators. In order to do that, the CDN operators need to solve the problems like improving the accuracy of the IP location technology, reducing the cost of server selection and optimizing the content copy storage mechanism.In this paper, we choose the Content Delivery Network QoS guarantee as the research direction. In order to solve the problems in server selection and Internet content copy storage mechanism, based on the analysis of the the main factors that may effect the QoS of content delivery network, we design and implement the optimization methods for IP location database,network distance measurement method, server selection algorithm as well as the Internet content copy storage mechanism.The main contributions of this thesis are as follows:1. Optimizing the accuracy of the IP geolocation databese in CDN based on network measurement and the existing geolocation databases.The IP geolocation of Internet host in CDN is not only an important factor in the process of content delivery, but also a key condition of realizing the content delivery based on the geographic locations. In this paper, we introduce two new local characteristic metrics: location degree and location betweenness, with the combination of huge traceroute data and the IP geographical location information in the IP geolocation databases, we give a comprehensive analysis on the performance of the existing geolocation databases of Internet hosts based on the two characteristic metrics. We propose an optimized scheme for the geolocation databases of Internet hosts, which is called GeoCop(Geolocation Cop). In addition to network measurement, which is always used by the existing geolocation methods, our geolocation model for Internet hosts is also derived by both routing policy and machine learning.After optimization with the GeoCop method, the geolocation databases of Internet hosts are less prone to imperfect measurement and irregular routing. Experiment results show that the GeoCop method not only achieves improved performance in both accuracy and robustness, but also enjoys less measurements and calculation overheads.2. Distance proximity prediction between the servers and the hosts of the users in CDN based on network distance level vector.Network distance is a very important parameter in server selection. We propose a scalable network distance proximity algorithm, which focus on the closest neighbor servers, not the special network distance values, which is called NDPE (Network Distance Proximity Estimation).We introduce two network distance characteristic metrics: network distance vector and network distance level vector, with the combination of network delay between the landmarks and the nodes on the CDN. We propose a clustering standard and a similarity algorithm to obtain the network distance similarity between the hosts and the servers in the same clustering, which is defined as network distance reference value. Simulation results show that NDPE not only achieves improved performance in both scalability and the accuracy of network distance prediction, but also enjoys less measurement and computation overhead.3. Global optimization for server selection in CDN based on global optimized algorithm.To cope with increasing number of clients and servers, CDN desires a more efficient way to select servers from multiple data centers in different geographic locations. We propose a three-stage global optimized server selection scheme called OMSS (Optimizaiton Method for Server Selection)in CDNs, which jointly considers the service experience, load balancing,traffic control and overheads. We introduce a penalty coefficient to leverage the quality of service experience in terms of network delay and the traffic control in terms of inter-domain transit traffic. In order to comprehensively analyze the performance of OMSS, we use typical benchmarks to compare OMSS with traditional approaches on the one hand,and also perform statistical tests to display the improvement of OMSS on the other hand. Simulation results show that OMSS not only achieves improved performance in both service experience and traffic control, but also enjoys less measurement and computation overhead.4. Internet content copy storage mechanism in CDN-P2P based on a hybrid architecture of CDN-P2P.Both CDN and Peer-to-Peer have their own limitations, but they can complement each other well in the cost of deployment, scalability, et.al,which can reduce the backbone network traffic and improve the hit rate of the edge servers. Based on a hybrid architecture of CDN-P2P, We propose an optimized content replica replacement scheme called CBO (optimized scheme ot x) in CDN, which jointly considers content size, the number of content requests and network distance. We also introduce time factor forcontent requests and interest factor in the content replica replacement.Simulation results show that CBO not only reduce the transmission costs between the edge server and source server, but also improve the hit ratio of the edge servers.