Placement of replicated items in distributed databases

Abstract

We address the problem of determining an allocation scheme for replicated data in a distributed database, with the purpose of minimizing required communication. An allocation scheme establishes the number of copies of each data-item, and at which processors of a given computer network the copies should be located. The problem for general networks is shown NP-Complete, but we provide efficient algorithms to obtain an optimal allocation scheme for three common types of network topologies. They are completely-connected, tree, and ring networks. We also propose a new method by which a processor in a computer network should write all the copies of a replicated data item. This method leads to surprising results concerning the allocation schemes.