Performance modeling of video-on-demand systems in broadband networks

A video-on-demand (VoD) system allows a viewer to choose a video of his choice such as movies, electronic encyclopedia, or educational videos, which is delivered to him real time via a network. This paper is a performance study of a distributed video server system in a fully connected backbone network. Under the assumptions of uniform loading and symmetrical network, analytical models are proposed for two server selection strategies (Random and Least Loaded) and for two reservation schemes (the strict reservation and the residual reservation). The models can be used to determine the call blocking probability and the requirements of network bandwidth given the capacity of video servers. In addition, our models provide a useful tool for VoD operators to design and dimension their systems.