DataFall: A Policy-Driven Algorithm for Decentralized Placement and Reorganization of Replicated Data

We present DataFall, a simple yet effective policy-driven algorithm for decentralized placement and reorganization of replicated data. Without relying on a centralized location for data mapping, DataFall efficiently distributes data objects across storage devices using a mechanism built on multiple hash functions. When producing the data placement, policies can be enforced at object level in a flexible manner. In addition, with minimum data movement, DataFall is capable of accommodating a wide variety of changes such as topology massive upgrade and reorganization. The advantages of DataFall over the state-of-the-art are demonstrated through an experimental evaluation in a set of selected scenarios.