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 servers offering the video retrieval and playback services are arranged as a distributed system in order to support a large number of concurrent streams. If such a system is hierarchical, an end-node server handles the requests from a particular area, the next 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 provides for cost efficiency, reliability and scalability of servers. The end-node servers store only a limited set of the overall 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 servers and the location of videos in the hierarchy, 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 servers) have been presented in the literature. Only a few 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.