An Effective Scheduling Policy in Large-Scale Video Servers

Abstract

Due to the advances in computer and multimedia techniques, the applications of video-on-demand are visible and influence people's life worldwide. However, with the growth of service demand, the single-server architecture is limited by its scalability and server -level fault -tolerance. Recently, several multi-server architectures are proposed to alleviate such problem. In this paper, we will propose a scalable architecture for video server management system. Our architecture employs not only distributed storage servers but also distributed subscriber management. Wide striping technique is a cost-effective data placement in multi-server architecture. However, the unpredictable access skew for users results in performance degradation. Therefore we design a new scheduling policy to fully utilize the potential bandwidth of wide striping by controlling the startup time of each request. With different modes of our policy, we can benefit from minimum delay or load balance. According to our simulation results, our policy efficiently alleviates load imbalance and thus improves system performance. The detailed information about design principles and performance evaluations will be described in the literature.