Novel spectral method for server placement in CDNs

摘要

In this paper, we present a novel method for replica server placement in CDNs. As we all known, replica server should be placed closer to clients in order to reduce the latency and the bandwidth consumption. To minimize the maximum distance between a node and the nearest center, the minimum K-center problem and the k-HST are presented by researchers. However, these theoretical methods are either computationally expensive or leave the characteristics of the network out of account. We improved the k-HST by applying the spectral clustering method into the decision problem about the number of surrogate server centers. The deterministic method we present is suitable for ISP or companies who already know the topology of its network to distribute server centers.