Research Of Server Selection Algorithm Based On Services Classification

The rapid development of the Internet has brought a great convenience to people’s work and life. However, the traffic of the Internet has shown an explosive growth by the continually abundant applications. This make congestion of the network especially the backbone network an austere we must face. In order to solve this problem, Tim berners-Lee, one of the founders of the Internet, launched a technical challenge aimed at solving network congestion in1995, which finally lead to the birth of the Content Delivery Network. Content Delivery Network changes Internet transmission mode fundamentally. By placing nodeservers throughout the network, users can get the needed resource from the closest edge nodeservers. In this way we remit the congestion of the backbone network, and at the same time improve the Quality of the Experience of the users.Server selection is one of the important questions we need to consider in content routing which is one of the key technologys in Content Delivery Network. It decides the merits of content routing, and affects the QoE of users. The traditional way to do server selection is using polling, random, or minimum load policies. However, when the burst hot events, it is easy to cause overloading on single server, and it can not use in the situation of multiple services. This exist problems will limit the further development and application of Content Delivery Network to the most extent. In order to solve this problem, this article proposes a probabilistic redirect algorithm using virtual load based on hybrid Content Delivery Network architecture. In each service domain, each node has a connection queue, which is used for store each connection. Load balancer uses a round-robin way to distribute requests to each CDN node, when the connection request queue reaches a predetermined threshold, the algorithm of this article will automatic enabled. The nodes which are overloaded redirect some of its requests to the lighter neighbor nodes. The algorithm can avoid a single server overloading, and will take advantages of the various server hardware resources to provide users with better QoE. Finally, we build a concrete experimental platform and use Apache Jmeter to simulate multi-user concurrency scenarios. This article respectively compare the average response time from the user and the virtual load of the request queue using the proposed algorithm with some the traditional algorithms. The experiment results show that the proposed algorithm improves average response time at least by10percent compared with round-robin and least load algorithms and the virtual load of the requests queues are more balanced.