Replication algorithms to retrieve scalable streaming media over Content Delivery Networks

Abstract

CDN (Content Delivery Networks) improves end-user performance by replicating web contents on a group of geographically distributed content servers. Replication Algorithm plays an important role in helping users to retrieve Web objects from the content servers. If a user can directly get the requested objects from the content server, he need not to contact the remote origin server and the user delay can be reduced. However, current replica strategies in CDN are to simply and repeatedly keep the complete replica of the original object on many content servers. This method has some disadvantages, including too much consumed server space and a waste of the storage cost. It is more serious for replicating some large-sized objects such as streaming media, which are being distributed over the Internet more and more.In this paper, we discuss a replication strategy for scalable video streaming in CDN to reduce user response and storage cost. Based on theoretical analysis, assuming layered video coding, we propose a novel replication algorithm which deals with following three problems. (1) How many content servers should be selected to replicate a given video content? (2) For a single video content, how many layers should be kept in a given content server? (3) After selecting a group of content servers for each video content, how do we decide the replication priority for each content server? Simulation results show that the proposed algorithm can efficiently resolve the above problems, and provide much better performance than the conventional methods.