Content Delivery Networks (CDNs) was introduced to serve content to the globally distributed clients with high availability, high performance and low latency. CDN replicates the content closer to the clients by replica servers, also known as surrogates. The primary concern in CDN is where to deploy these surrogates so that the content to client latency will get optimized. The existing replica server placement algorithms model this problem as NP hard facility location problem where there is no optimal solution. In this paper, an unsupervised clustering approach, Population-based clustering is proposed. The proposed algorithm decides where to deploy surrogates based on population threshold, a minimum number of clients a cluster should have. The experiments show a better result than traditional clustering when surrogate utilization is considered.
|