Dynamic Grouping: an Eecient Buuer Management Scheme for Video-on-demand Servers

We observe the misbehavior of traditional LRU-like buuer management schemes in video-on-demand (VOD) applications, which fail to utilize the buuer space eeciently. By taking advantage of the characteristics of these applications, we design a new buuer management technique called Dynamic Grouping. This new approach dynamically groups video streams into groups to allow streams in each group to share a single I/O stream. In other words, all video streams in a group, except one, will experience only buuer hits when retrieving their data. Signiicant reduction in the demand on the I/O bandwidth, therefore , can be achieved. Our technique optimizes the length of each group to maximize the overall performance of the media server. Through simulation study and mathematical analyses, we demonstrate that Dynamic Grouping ooers substantial improvement over conventional schemes.