【作者】 邢建英； 李舟军； 张艳；

【Author】 Xing Jian-Ying1, Li Zhou-Jun2 and Zhang Yan 3 (1School of Computer, National University of Defence Technology, Changsha 410073) ( 2 School of Computer Science & Engineering, Beihang University, Beijing 100083) ( 3 School of Mathematics, Sichuan University, Chengdu 610064)

【机构】 国防科技大学计算机学院； 北京航空航天大学计算机学院； 四川大学数学学院；

【摘要】 多媒体服务器要能够提供实时的不间断服务,并保证服务的高可用性。但服务器总不可避免地会出现失效状况,这就要求服务器具有一个合适的恢复子系统。恢复过程要能够保证服务器迅速从失效中恢复并且不丢失数据,还要尽可能地少影响服务器的正常运行。本文介绍了一种利用视频流内部特征的磁盘失效恢复算法,利用视频播放的顺序性来降低在普通RAID阵列上实现的联机失效恢复算法的开销。更多还原

【Abstract】 Multimedia servers must provide real-time and uninterrupted services, and must guarantee high availability of services. However, failures may occur at times, server must have an efficient recovery subsystem. A recovery subsystem must not only provide mechanisms to rapidly recover from a disk failure without losing data, but also ensure that the recovery process operates without taking the system offline and has minimal impact on system performance. This paper presents a disk failure recovery algorithm that utilizes the inherent characteristics of video streams. It exploits the sequentiality of video playback to reduce the overhead of online failure recovery in conventional RAID arrays. Dr. Prashant J. Shenoy invented this algorithm in 2000. Observe that the server can recover blocks stored on the failed disk either in the round in which they are accessed, or at least one round prior to their access. These recovery policies are referred to as lazy and eager failure recovery, respectively. The fundamental difference between the lazy and the eager recovery algorithms is that the latter trades buffer space to replace a sequence of load fluctuations possible in the former by a constant increase in the load. Both the lazy and the eager failure recovery algorithms require that all media blocks contained within a parity group belong to the same video stream. Consequently, to implement these failure recovery algorithms, conventional array controllers must be suitably extended. Specifically, if the controller can implement a function that takes a logical block number as its input and returns a list of logical block numbers of all blocks which belong to its parity group, then a multimedia server can exercise precise control over membership of blocks within a parity group. Compared with standard recovery techniques, these recovery algorithms can reduce the overhead on disk. And, in case of a disk failure, the server either discards the block to be stored on the failed disk, or writes it to a replacement disk. Since the array operation is unchanged in the presence of failures, the full-stripe writes that are used in these algorithms impose no extra overhead on the server as compared to the fault-free state. In multimedia server recovery, the node failure recovery techniques are also important. Specially, the node failure recovery algorithms have much similarity in various architectures, such as Client/Server systems, distributed systems etc.更多还原