A Video Caching Policy for Video on Demand System

• Main Authors: Yuan-Tse Yu, 余遠澤
• Other Authors: Sheau-Ru Tong
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/86944409100863307092

碩士 === 國立屏東科技大學 === 資訊管理系 === 87 === The architecture of video servers, responsible for generating continuous video bit streams in response to playback requests from clients, imposes a significant impact on the performance of Video-On-Demand systems. Normally, we take a bi-level architecture: A back-end level, termed archive level, consists of a set of tertiary storage devices (e.g., towers of tape-drive or CD-ROMs) for archiving the entire video collection. Data access on this level normally takes a long response time and can only afford a very limited number of current accesses. To improve the performance, a front-end level, termed cache level, is introduced. It consists of a set of secondary storage devices (e.g., hard disk drives) for caching a small subset of videos online to improve the hit-ratio and provide a large number of current accesses with a short response time. Videos usually require different network bandwidth and storage space with time-variant popularity and pricing structure. Apparently, the decision of the online video set by taking these parameters into account greatly affects the total profit gain. In[1], we focused on the cache-level server design and proposed a Network-based Cooperative Server Cluster (NCSC) architecture where servers, connecting to a network, are logically clustered into several logical servers with different capabilities employing an adaptive coarse-grained data striping technique. This architecture shows excellence on both resource integration as well as system scalability. In this paper, we take the NCSC architecture as the basis and study how to build up logical servers (as cache) for different quality videos in the context of the bi-level architecture. Specifically, we are interested in finding a global rewarding model involved with various system parameters in the bi-level system. Based on the model, we propose a complete strategic framework, consisting of a set of heuristic algorithms, for determining the best logical server deployment so that the total reward is maximized.