An accurate understanding on the content popularity distribution of media objects in streaming applications is required when designing efficient caching algorithms for the streaming media. However, there is no empirical model to describe this distribution. Our investigation shows that the content popularity of a hot streaming media file exhibits a Zipf-like distribution under K-transformation, after analyzing 250 hot clips in 2 long-term traces. A specific function is proposed to describe this phenomenon. This result builds a foundation for designing intelligent caching algorithms.
Previous studies have shown that an increasing fraction of the data on the WWW is fast-changed, such as news, sports scores, stock prices, and so on. An important issue to cache such dynamic content is the maintenance of consistency. We propose a self-organizing caching scheme to ensure the users get the up-to-date documents all along even if the documents are fast-changed. This scheme combines the mechanism of client polling and that of server pushing based on a mechanism of self-organizing multicast trees. The trees are created, optimized and destroyed automatically according to the rate of change and location of the dynamic document and the network conditions, such as the capacity, the traffic load. Meanwhile we derive a formula for cache nodes to determine which documents should be cached.
Caching video objects at the edge of the back-bone network has become a hot spot in the video delivery research. Some researchers proposed segmentation-based caching algorithms to improve the disk utility in the homogeneous environment. Other researchers proposed caching algorithms with video layering or transcoding to improve the perceptual quality in the heterogeneous environment. However, for a low cost proxy, these systems are suffering from either inflexible structures or high computation. In this paper, the overall quality hit ratio(QHR) is proposed to measure how well the clients' requests on the quality are satisfied. Aiming to achieve the highest QHR, an intelligent caching algorithm is designed to determine the appropriate perceptual quality of the segments cached in the proxy. This innovative approach takes not only the access bandwidth distribution into account, but also the segment popularity distribution. Simulation results demonstrate that our variable rate caching(VRC) algorithm not only achieves the highest QHR but also keeps its complexity at an acceptable level.