Maximizing Availability of Replicated Disk Array for Video-on-Demand Systems

碩士 === 國立成功大學 === 工程科學系 === 87 === Abstract During the past decade, advances in high-speed network technologies, video compression techniques, and storage devices, have made Video-on-Demand (VoD) feasible. A large-scale VoD system imposes a great demand on bandwidth and storage...

Full description

Bibliographic Details
Main Authors: Hao-Li Wang, 王皓立
Other Authors: Yueh-Min Huang
Format: Others
Language:en_US
Published: 1999
Online Access:http://ndltd.ncl.edu.tw/handle/34126545091448123225
Description
Summary:碩士 === 國立成功大學 === 工程科學系 === 87 === Abstract During the past decade, advances in high-speed network technologies, video compression techniques, and storage devices, have made Video-on-Demand (VoD) feasible. A large-scale VoD system imposes a great demand on bandwidth and storage resources, and therefore, parallel disks are typically used for providing VoD service. Replicated disk array is regarded as an efficient solution to achieve both reliability and availability of a video server. To balance the load between disks on a single disk failure, the placement and permutation of data blocks are considered as the most critical issue. In the thesis, we explore the approach of duplicating all movies to provide high data bandwidth and fault tolerance required in a disk-array-based video server. Then, we propose a dynamic load-balancing scheme on a replicated disk array for a VOD system to alleviate the skew access problem on a single disk failure. Based on the dynamic load-balancing scheme, we further elaborate the basic principle to place the data blocks on a disk array to maximize the availability. Our data placement principle with the dynamic load-balancing scheme is shown to maximize availability of a replicated disk array for video-on-demand systems. Additionally, we propose a replicated data placement scheme, namely shift chained declustering (SCD), which supports high data availability for disk arrays in a VOD environment under normal and failure situations. In essence, SCD is similar to the conventional chained declustering where primary data copy in disk i has a backup copy in disk (i+1). However, it is different from the latter in that primary data copy in disk i has a backup copy in disk (i+s), where s is determined according to our proposed rule. Our results show that SCD consistently outperforms the conventional methods in terms of initial delay, load balance, and fault-tolerance capability after disk failure, and is regarded as a feasible approach to supporting replica placement in a disk-array-based video server.