Low complexity content replication through clustering in Content-Delivery Networks

Contemporary Content Delivery Networks (CDN) handle a vast number of content items. At such a scale, the replication schemes require a significant amount of time to calculate and realize cache updates, and hence they are impractical in highly-dynamic environments. This paper introduces cluster-based replication, whereby content items are organized in clusters according to a set of features, given by the cache/network management entity. Each cluster is treated as a single item with certain attributes, e.g., size, popularity, etc. and it is then altogether replicated in network caches so as to minimize overall network traffic. Clustering items reduces replication complexity; hence it enables faster and more frequent caches updates, and it facilitates more accurate tracking of content popularity. However, clustering introduces some performance loss because replication of clusters is more coarse-grained compared to replication of individual items. This tradeoff can be addressed through proper selection of the number and composition of clusters. Due to the fact that the exact optimal number of clusters cannot be derived analytically, an efficient approximation method is proposed. Extensive numerical evaluations of time-varying content popularity scenarios allow to argue that the proposed approach reduces core network traffic, while being robust to errors in popularity estimation.