Architectural design of on-demand video delivery systems: the spatio-temporal storage allocation problem

The authors consider the problem of designing an on-demand video delivery system having a large number of video entities stored for distribution to a large number of customers. A logically layered information distribution network is assumed to serve the requests made by the customers. The authors propose a stochastic model for video entities in the distribution system and pose a spatio-temporal allocation problem to minimize a general cost function that includes storage, transport, and processing costs. Though such problems are generally known to be hard in terms of finding an implementable solution, the authors exploit the specific characteristics of the present problem to provide an efficient polynomial time solution. They also show a simple recursive estimation procedure to track the time-varying demand probabilities assigned to the video entities in the system. Various approaches to a practical design are discussed.<<ETX>>