Operation and cost optimization of a distributed servers architecture for on-demand video services

Distributed servers architecture offers storage and streaming scalabilities for video services. In this letter, we propose and analyze an on-demand scheme in which the local servers store the beginning portion (i.e., the "prefix") of videos and deliver it by means of unicast streams to the clients. The clients are able to merge onto on-going multicast streams delivered from the repository by means of their set-top buffers. Given a certain limited repository (and thereof multicast) bandwidth, we investigate how the total cost of the local servers can be minimized. We show that if the local storage is the main cost, the size of the prefixes is likely to increase with the video popularity. On the other hand, if the server cost mainly comes from streaming capacity, the size of the prefixes is likely to decrease asymptotically with the video popularity.