The procedures for file creation and replica addition both require a file's parent directory to be present on the same node. Pangaea, in fact, demands that for every file, all intermediate directories, up to the root, are always replicated on the same node. This name-space-containment requirement yields two benefits. First, it naturally offers the availability and autonomy benefits of island-based replication [14]. That is, it enables lookup and access to every replica even when the server is disconnected and allows each node to take a backup of the file system locally. We quantify these benefits in Section 7.8. Second, it simplifies the conflict resolution of directory operations, as we discuss in Section 5.2.
On the other hand, this requirement increases the system-wide storage overhead by 1.5% to 25%, compared to an idealized scheme in which directories are stored on only one node [28].2 We consider the overhead to be reasonable, as users already pay many times more storage cost by replicating files in the first place.