Long-term movie popularity models in video-on-demand systems: or the life of an on-demand movie

Large scale video-on-demand systems require that the serv ers offering the video retrieval and playback services are arranged as a distributed system in order to support a lar ge number of concurrent streams. If such a system is hierarchical, an end-node serv er handles the requests from a particular area, the ne xt server in the hierarchy takes the request over for several end-node servers if those can not answer the request and so on. This architecture pro vides for cost efficiency, reliability and scalability of serv ers. The end-node servers store only a limited set of the o verall available information which changes over time due to user interests. If a video is requested which is not available, this server contacts the next server in the hierarchy. To decide the size and location of the video serv ers and the location of videos in the hierarch y, the access behaviour of users must be considered. Various models for the simulation of user behavior (and thus, of the load induced on the video serv ers) have been presented in the literature. Only a fe w of these models are designed to take long-term effects into account because the basis for most of the models are short-term influences on a single video server and the load on this single machine. In this paper we describe a new user behavior model and show that various assumptions made within other models are unrealistic.