Disk striping is essential in Video-On-Demand servers to allow concurr
ent access for a large number of users to the same copy of a video obj
ect. On the other hand, disk striping makes the system more vulnerable
to component failures. We propose a replication scheme for video cont
ent and an associated I/O scheduling algorithm that enables the system
to tolerate a component failure. The scheme, called doubly-striped mi
rrors, compares favorably to other approaches for implementing redunda
ncy such as RAID in terms of cost, bandwidth efficiency, buffer usage
and system availability.