We present ARK (Adaptive Replication of Keywords) which is a replication algorithm working on DHT overlay networks. We proved in this paper that ARK is near optimal in load balance and is competitive in replica management overhead. Simulations show that ARK has great improvements in load balance and fault tolerance, comparing with existing replication algorithms in CAN, Pastry and Bamboo. ARK is achieved without explicit metadata or the need for a replica directory service, works as an independent building block on top of any overlay system. Nodes unaware of ARK can work well with nodes equipped with ARK. Although our research is part of a content indexing system, the algorithm is suitable for any overlay based data item storage and lookup system.
展开▼
