Replica Placement and Repair Strategies in Multinode Storage Systems

摘要

A multimode storage system uses a global-placement and local-repair strategy for replica placement and repair of data objects. The data objects are initially placed randomly throughout the storage system, while the subsequent replacement replicas of any lost data objects are placed non-globally, for example within the same cluster as the repair source. This approach aggregates the bandwidth of all leaf switches/routers for repair and thus improves data repair bandwidth and repair speed. The approach also reserves more root switch bandwidth for tasks such as data rebalance, a process migrating data to the new nodes that have replaced failed ones.