On providing highly available fault-tolerant video-on-demand services

Due to the large number of magnetic disks involved in a video server, the probability of disk failure is fairly high. Traditional parity-based and duplex-based (mirroring) approaches provide full data recovery, but do not guarantee full availability of the service in the presence of even a single disk failure. If a video server would like to provide non-disruptable services to a large number of viewers, the issue of high availability needs to be addressed. In this paper, we proposed a general scheme for accomplishing both fault tolerance and high availability in a video server. The tradeoffs between memory buffer space and redundant disk storage space are discussed. A video allocation problem based on the proposed scheme is formulated and heuristic algorithms to solve this problem is presented. Our scheme can be easily implemented in a highly available fault-tolerant video server.