On Storage Performance for Multi-Steam Video-on-Demand Systems

Abstract

One of the most important criteria in evaluating a video-on-demand system is the maximum number of video streams that can be simultaneously supported. There are several bottlenecks which limit this capability, such as storage I/O, network bandwidth, and CPU processing overhead. In this paper, we focus on the performance modeling of disk drives in a multi-stream video-on-demand system. We investigate three types of storage subsystems, namely the single-disk single-bus, the multi-disk dingle-bus and the multi0disk multi-bus. To derive the maximal number of video streams, the data size per I/O request and the required buffer size for each video stream is formulated. The result shows that the number of video streams is limited sue to the unavoidable mechanical delay of the disk drives. The performance improvement by suing data stripping on an array of disk drives is also shown to be limited since unavoidable overhead still exist. To verify the correctness of the analysis, and inexpensive video-on-demand system based on personal computers and Ethernet is demonstrated. The experimental results is shown to be very close to the analytical result.