Design and performance evaluation of fail-over and recovery strategies for large-scale multimedia storage systems

Failure-over and recovery is a crucial issue for large-scale multimedia storage system (MSS). In the literature, there are two fail-over policies: active-standby fan-over and active-active fail-over. In this paper, we adopt an active-active policy and design two heuristic algorithms to determine the number and the placement of the object replicas among the multimedia servers. One is named as "Minimum, Average Waiting Time" (MAWT) strategy, in which we will choose the server to store an object replica according to an optimal solution, and the other algorithm is named as "One of the Best Two Choices" (OBTC) strategy, in which we will randomly choose one out of the first two best choices according to an optimal solution. We analyze their performance in the cases when the system runs normally and when some failure occurs. Via rigorous simulations, we find that when the system is running normally (without failures), MAWT performs better than OBTC and when some server fails, OBTC performs much better than MAWT.