A dynamic migration algorithm for a distributed memory-based file management system

Conventional migration strategies attempt to evenly balance the load across all available server machines. The paper discusses why conventional migration approaches are not necessarily appropriate for distributed memory based file systems and presents an alternative approach that spreads data (possibly unevenly) across as few machines as possible and involves other available machines only as needed. The main advantage of our approach is that it keeps the system minimally distributed, thereby reducing the failure rate among servers, the communication overhead among servers, the time needed to compute data relocation, distributed addressing costs, and the probability of unanticipated migrations (e.g., caused by, and an inconvenience to, returning users).