A selective replication-based approach to load rebalance for VOD cluster

Video-on-Demand (VOD) is a popular data access pattern in network storage systems. One of an effective load balancing solutions used in VOD systems is cooperative cache partitioning, it divides each video file into chunks and stores them throughout the cluster, therefore, the loads are distributed across multiple servers to decrease or eliminate hotspots in the system. In a VOD cluster, the number of hottest videos is usually small but those videos account for a large percentage of requests, in this paper, we propose a selective replication-based approach, which is based on cooperative cache partitioning, to replicate N number of hottest videos across VOD servers to further rebalance the loads on the cluster and reduce request rejection rate. Results show that our approach can achieve about 60% request rejection rate decrease when the number of N ranges from 10 to 50.