Deterministic load-balancing schemes for disk-based video-on-demand storage servers

Abstract

A video-on-demand (VOD) storage server is a parallel, storage-centric system used for playing a large number of relatively slow streams of compressed digitized video and audio concurrently. Data is read from disks in relatively large chunks, and is then "streamed" out onto a distribution network. The primary design goal is to maximize the ratio of the number of concurrent streams to system cost while guaranteeing glitch-free operation. This paper focuses on load-balancing for the purpose of providing throughput that is independent of viewing choices. At the interdisk level, data striping is the obvious solution, but may lead to a quadratic growth of RAM buffer requirements with system size. At the intradisk level multizone recording results in variable disk throughput. Deterministic schemes for solving each problem are discussed, as well as their joint operation. Finally, efficient staging of data from tertiary storage devices to disk is shown to be possible.