Skip to Main Content
A significant portion of the HTTP multimedia traffic on the Internet comes from sites like Youtube which serve short videos. Caching of Youtube-like multimedia content, when possible, can reduce traffic on the backbone while providing faster access. The performance of such a caching system will depend on identifying the videos which should be cached and the appropriate duration. In this paper, we look at both of these questions from a social network perspective. We propose that the decision to cache a video should be based on the combined popularity of the individual as well as related videos rather than simply based on individual popularity of a video. We identify timescales at which the inter-relationships between the videos can change through a longitudinal data set. Using the concepts of centrality of nodes, we rank the set of videos in the data set according to their perceived importance. In doing so, we compare three centrality techniques - degree, closeness and betweenness. We evaluate how these centralities affect the performance of a cache. We show that ``Closeness" centrality always performs at least as well as the other two in all cases. Finally, we show that a distributed cache mechanism employing the centrality method to rank videos can reduce the load on the network significantly for even moderate content cache sizes.