A data-driven and the deep learning based CDN recommendation framework for ICPs

Abstract

It is a significant trend that the Internet Content Providers (ICPs) improve the quality of service and reduce the cost of content distribution by the Content Delivery Networks (CDNs). In order to spend the less money to get better services, ICPs need to find a lot of information about CDNs, such as server deployment, performance, price and so on, to determine whether CDN services satisfy their requirements. Unfortunately, these information can’t be obtained by third party due to business secret. ICPs still choose CDNs on the basis of one-sided viewpoint. For this reason, we have proposed a data-driven and the deep learning based CDN recommendation framework for ICPs. The contributions lie in: 1) A three-tier CDN recommendation framework is presented to achieve data-driven and the deep learning based recommendation service. 2) A CDN recommendation model is built based on the deep neural network, which improves the efficiency of the recommendation service and satisfies the personalized demand. 3) A prototype system is developed and deployed on the real-world large-scale Internet in China. Experimental results demonstrate that the correctness of the recommendation results is up to 91%, and degree of satisfaction reached 80%.