EFFICIENT VIDEO SERVER

Abstract

The large-size and continuous display requirement of video data makes the design of a video-on-demand (VOD) server to support a large number of simultaneous video streams a challenging task. One main challenge is to devise methods for striping video data onto multiple disks in order to effectively utilize the server's resources. A popular method for striping video data is the sequential method, where consecutive video blocks are striped sequentially onto multiple disks in a round-robin fashion. In this thesis, we propose a phase-based striping method which has the desirable characteristics that it is able to guarantee the maximum waiting lime for its users. We also develop a data replication scheme to further reduce the average waiting time of the phase method. We integrate the methods into a combined scheme, where the phase-based striping method with possible replication is used to store popular videos, while the sequential striping method is used for less popular videos. We comlucl simulations on a video server containing 36 videos under different loading and hardware configurations. The performance metric used in the simulation is the average waiting time. The simulation results demonstrate that the use of hybrid methods can lead to substantial improvement in the performance of video servers, as compared with using only the normal sequential striping method. The use of replication methods could further improve the server performance by up to 15%. The general results also indicate that with limited resources, the use of phase-based methods with replication is preferable under heavy loads.