The Study Of Disk I/O Scheduling Algorithm For High Definition Video Servers

It is one of the keys to the feasibility for the SDFC (servers are deployed along the frontier of cilents) to make good use of the hardware capacity so as to promote the cost performance as much as possible for a single video server. At present, the bandwidth of resource hard disks is the main bottleneck for video servers, because, especially in a concurrent environment, simultaneous access to a hard disk will lead the disk bandwidth to a serious decline to severely restrict the capacity of video servers. The followings are the major works of this paper. We investigate the I/O block devices subsystem in Linux and give in-depth analysis and evaluation for all of the four I/O scheduling algorithms provided in Linux. On the basis of the above, we design a dedicated I/O scheduling algorithm, called HVOD, for a video server in which the video resources are striped and stored in every resource hard disk, and implement the algorithm in Linux2.6.32kernel. The HVOD algorithm makes precise read prediction to improve the performance for sequential disk reads and employs timeout queues to prevent I/O starvation, so as to make sure as many videos are replayed smoothly as possible. The tests, which are made in a real VOD system and in a VOD simulator, show that, in a video server that video resources are striped, the performance of HVOD is40-50%higher than the present I/O scheduling algorithms provided in Linux. Our research confirms that the SDFC scheme have realistic feasibility even for the large-scale VOD services that provide4K high definition videos.