CLIENT BEHAVIOUR PREDICTION IN A PROACTIVE VIDEO SERVER: NEW CONSIDERATIONS AND RESULTS

摘要

A self-organizing Video-on-Demand server is able torndynamically adapt to the changing circumstances byrnreplicating and removing the server components and thernstored videos. The video replication takes considerablerntime therefore it should be performed in advance, in otherrnwords proactively. This paper is a follow up article of [1]rnin which we examined 23 predictor functions on artificialrnand real datasets in order to supply VoD systems withrnproactive behaviour. We investigated the appliedrnsimilarity measures further and also developed newrnvariants in order to incorporate task specific weightrndistributions. Using a new set of similarity measures, thernranking of the predictors were repeated. The two bestrnpredictors were put on trial in a large-scale simulation ofrnproactive and reactive video servers. The results of thernperformance studies show how the type of the predictor orrnits absence affects the overall performance of the server.