Bounded dynamic data allocation in distributed systems

Recently, an interesting dynamic data allocation algorithm for the management of replicated data in distributed systems has been presented by Wolfson and Jajodia. We analyze this approach from cost as well as from availability perspectives and suggest an improved algorithm that exhibits superior characteristics. Our approach establishes lower and upper bounds in the number of replicas for guaranteeing minimal availability properties and for restricting operation costs. It additionally exploits an improved replica management that abandons the read-one/write-all approach for all replicas as well as the central server concept used by the original approach, thereby presenting a truly distributed solution. Analyses show that the improved algorithm is superior in terms of costs and availability assuming any reasonable workload.