Storage rebuild for automatic failure recovery in video-on-demand servers

In a previous study, Wong and Lee (see ICC'97, Montreal, Canada, 1997) proposed a redundant array of inexpensive servers (RAIS) architecture for designing scalable and fault-tolerant video-on-demand systems. Video data are striped across an array of autonomous servers, resulting in a scalable architecture where more concurrent video sessions can be supported by simply adding more servers. Moreover, by adding data redundancy among the servers, client recovery algorithms can be implemented to sustain server failure and provide non-stop video services. We consider the failure recovery issue in RAIS. Specifically, we propose and analyze three algorithms for rebuilding data at a failed server to a spare server in order to restore the system back to normal operation. We derive the performance model and use numerical results to show that automatic rebuild can be done in reasonable time using the proposed rebuild algorithms.