A Queuing Analysis of Server Sharing Collectives for Content Distribution

A content provider must choose how to provision its server resources to handle client demand for its content. When provisioning is based on typical demand, clients are turned away during peak demand periods. Provisioning based on peak demand or contracting a third party content distribution network service increases costs. An alternative approach is for the various content providers to form a collective in which their server resources are pooled together, servicing the demands of the entire collective. Here we compare the performance of server sharing collectives to content providers who operate their servers in isolation. We demonstrate using analysis and simulations upon fundamental queueing models that by hosting one another‘s content within a collective, content distributors can reduce the rate at which client requests are turned away. We also analyze a class of thresholding techniques that content providers within the collective can apply to limit the use of their own server resources by other collective members. We show that these thresholding techniques permit a content provider to assist other providers within the collective without sharing its resources to a point where its ability to service its own clients is unsatisfactory.