Comparative study of scalable batching policies in disk-array-based deterministic video-on-demand servers

Presented is a comparative study of various batching policies in deterministic video-on-demand (DVoD) servers under realistic conditions including customers' behavior storage organization, and time variations in request arrival rates. DVoD servers provide video playback without any interaction support and customers' QoS is simply expressed in terms of average waiting time to receive service and defection rate caused by long waits. We focus on disk-array-based DVoD servers, whose movie selection and channel capacity are determined by cost and the underlying striping scheme. We analyze the well-known lack of scalability of on-demand batching, and introduce a methodology to measure scalability. We then compare various scalable batching policies and show that even the simplest batching policy may vastly improve on-demand channel allocation. Finally we study the feasibility of different storage organizations in which disks are partitioned into several clusters to reduce service disruptions during their reconfiguration. We also show that high service availability and scalable batching can be achieved in a cost-effective way.