Scalable storage architectures allow for the addition or removal of storage devices to increase storage capacity and bandwidth or retire older devices. Assuming random placement of data objects across multiple storage devices of a storage pool, our optimization objective is to redistribute a minimum number of objects after scaling the pool. In addition, a uniform distribution, and hence a balanced load, should be ensured after redistribution. Moreover, the redistributed objects should be retrieved efficiently during the normal mode of operation: in one I/O access and with low complexity computation. To achieve this, we propose an algorithm called random disk labeling (RDL), based on double hashing, where storage can be added or removed without any increase in complexity. We compare RDL with other proposed techniques and demonstrate its effectiveness through experimentation.
