Dynamic Replica Management in Distributed Hash Tables

Interest in distributed storage is fueled by demand for reliability and resilience combined with decreasing hardware costs. Peer-to-peer storage networks based on distributed hash tables are an attractive solution due to their efficient use of resources and resulting performance. The placement and subsequent efficient location of replicas in such systems remain open problems, especially (1) the requirement to update replicated content, (2) working in the absence of global information, and (3) determination of the locations in a dynamic system without introducing single points of failure. We present and evaluate a novel and versatile technique, replica enumeration, which allows for controlled replication and replica access. The possibility of enumerating and addressing individual replicas allows dynamic updates as well as superior performance without burdening the network with state information, yet taking advantage of locality information when available. We simulate, analyze, and prove properties of the system, and discuss some applications.