Performance Improvements of Large-Scale Video Servers by Video Segment Allocation

摘要

As large-scale commercial video-on-demand services develop, the creation of large-scale video servers that can provide video to over 10,000 terminals is needed. A large-scale video server distributes and stores video in segments using striping technology on many hard disks. Segment duplication is an effective way to improve the performance of video servers, but unless the allocation of segments to several hard disks is appropriate, retrieval requests become concentrated on a particular hard disk, and retrieval delays result. Thus, in this paper the authors describe a method to find a segment allocation that improves video server performance, including the number of retrieval streams and response time, and without creating retrieval delays, by taking into consideration the allocation to hard disk of video segments stored in a video server. In the authors' method, segments to be duplicated are determined and an allocation found by solving a limited satisfaction problem using the access probability for segments. Based on simulation experiments, the authors confirmed that their method was able to create a segment allocation which allowed for better retrieval performance than allocation methods used in the past.