Name
Abstract | ||
|---|---|---|
We present a possibility how to add proactive behaviour to Video-on-Demand systems. To do so we propose categorizing videos and using external information as well as observing the behaviour of our clients. We examined 23 predictor functions on artificial and real datasets using different similarity measures to compare them. Our model is quite simple; therefore some extensions are proposed at the end. On-demand streaming can be regarded as a solved problem, at least inside a good local area networking environment, as often used by enterprises. The same techniques may, however, lead to unsatisfactory results if used in a general Internet setting. Therefore, it seems to be a good idea to create a video server that is able to make local copies of its code and the required parts of its videos at the place where these are needed. Thus, the clients have the impression of being served by a local enterprise server for the price of a few, more or less idle nodes which temporarily host the downloaded code and videos. Such a video server implements an offensive adaptation strategy (1). As videos are usually large in size, replication itself may take a considerable time. Therefore, the videos should be replicated in advance - in other words proactively. This was our basic motivation in creating a proactive and offensively adaptive video server. Every proactive server needs to know something about the future behaviour of its users. It can obtain this information from outside the system or make predictions with the help of observations about the past. Our server uses a mixture of these two approaches. It can predict on its own which clients are interested in a given video topic and is able to use external hints about the popularity of videos. In this paper we examine different predictor functions (simple moving averages, autoregressive moving averages, a neural network approach) which help the server to choose which clients will be covered by proactive adaptation. We investigate the predictors in two different test scenarios: in an artificial one and in a real one. In order to compare the results of the predictors we have three different similarity measures at our disposal. |
Year | Venue | Keywords |
|---|---|---|
2005 | EuroIMSA | adms,proactive,client behaviour prediction,video-on-demand,ranking,neural network,local area network,moving average |
Field | DocType | Citations |
Video server,Computer science,Computer network,Video tracking,Multimedia | Conference | 2 |
PageRank | References | Authors |
0.40 | 5 | 3 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Peter Karpati | 1 | 99 | 10.46 |
| András Kocsor | 2 | 242 | 32.83 |
| László Böszörményi | 3 | 485 | 66.44 |