Client Behaviour Prediction in a Proactive Video Server: New Considerations and Results

P. Kárpáti (Norway) and T. Szkaliczki (Hungary)


Client behaviour predictor, similarity measure, proactive, VoD server, simulation


A self-organizing Video-on-Demand server is able to dynamically adapt to the changing circumstances by replicating and removing the server components and the stored videos. The video replication takes considerable time; therefore, it should be performed in advance, in other words proactively. This paper is a follow up article of [1] in which we examined 23 predictor functions on artificial and real datasets in order to supply VoD systems with proactive behaviour. We investigated the applied similarity measures further and also developed new variants in order to incorporate task specific weight distributions. Using a new set of similarity measures, the ranking of the predictors were repeated. The two best predictors were put on trial in a large-scale simulation of proactive and reactive video servers. The results of the performance studies show how the type of the predictor or its absence affects the overall performance of the server.

Important Links:

Go Back