View divergence control of replicated data using update delay estimation

We propose a method to control the view divergence of replicated data when copies of sites in a replicated database are asynchronously updated. The view divergence of the replicated data is the difference in the lateness of the updates reflected in the data acquired by clients. Our method accesses multiple sites and provides a client with data that reflects all the updates received by the sites. We first define the probabilistic lateness of updates reflected in acquired data as read data freshness (RDF). The degrees of RDF of data acquired by clients is the range of the view divergence. Second, we propose a way to select sites in a replicated database by using the probability distribution of the update delays so that the data acquired by a client satisfies its required RDF. This way calculates the minimum number of sites in order to reduce the overhead of read transactions. Our method continues to adaptively and reliably provide data that meet the client´s requirements in an environment where the delay of update propagation varies and applications´ requirements change depending on situations. Finally, we evaluated the view divergence we can feasibly control using our method. The evaluation is done by means of simulations. The evaluation shows that our method can feasibly control the view divergence to about 1/4 that of a normal read transaction