Adaptive partial CDN replication

Content Distribution Networks are a very important part of today's Internet and the Web. They enable the scalable provision of content and yield significant saving for the information provides (the clients of the CDN). In this paper we propose a scheme that tries to optimize the use of disk space in the CDN. The basic idea involves the partial replication of the contents of a web site. Site objects are transferred to the CDN in bundles. The site partitioning technique that produces bundles can be based on the site structure or the popularity of the site contents. We propose two schemes for site partitioning: the static partitioning and a technique based on Markov graphs (mostly intended for Web prefetching). We simulated the discussed schemes using extensive traces taken from the site of our department. Our findings are very promising for the proposed schemes.