Abstract

A number of techniques have been developed for reducing disk latency in a video server, including disk arm scheduling and data placement ones. In this paper we carefully study memory utilization in video servers, leading to enhanced techniques that significantly reduce memory use. This is achieved by introducing artificial delays between IOs. Surprisingly, the delays lead to better overall performance as compared to "traditional" disk latency reduction techniques. We also propose a novel disk scheduling policy that reduces initial latency (i.e., the time between the arrival of a request and the start of the presentation) to minimum. In addition, we study the memory requirements and costs of a video delivery system, showing how data placement on disk and multiple disks impact memory use. In doing so we show that it may not be advisable to achieve the maximum throughput that a disk can support.

Citations