Subtree partitioning and hashing are two common techniques used for current distributed metadata management schemes. Subtree partitioning schemes do not evenly distribute workload among metadata server cluster, and a number of metadata need to be migrated to achieve coarse load balancing. Though current hashing schemes evenly distribute workload among metadata server cluster, they eliminate the storage locality of files metadata in the same directory. They also can not effectively deal with some operations such as renaming a directory or accessing a file in the path of a linked directory. A new approach called parent Directory Object IDentifier and Filename Hashing (DOIDFH) metadata management is proposed, which could distribute evenly files metadata among metadata server cluster, as well as partially hold hierarchical locality. It solves effectively the existent problems of current hashing schemes and also takes advantages of them.
展开▼
