Scalable Load Distribution for View Divergence Control of Data Freshness in Replicated Database Systems

Abstract

A scalable load distribution method for view divergence control of statistically defined data freshness in replicated database systems is proposed. This method enables a number of mobile and fixed client nodes to continue their processes even when they lose connectivity to a network or do not have sufficient bandwidth to meet application requirements, which is very likely to occur to mobile client nodes. This can be achieved by renewing copies of data in client nodes while they maintain connectivity to a network so that their copies of data are sufficiently fresh to meet application requirements while they lose network connectivity. The load distribution for view divergence control is achieved by determining multiple sets of replicas from which client nodes retrieve the values of data through read transactions. Client nodes calculate the value of data that reflect updates which have already reached one or more elements in the determined set of replicas. We show that our method reduces the load of processing read transactions to less than about 1/40 of that in the original method in order to improve data freshness to about 2/5 of the maximum update delay in a large-scale network.