Modelling the Best-client Data Replication Method based on Stochastic Reward Nets

Abstract

In order to improve the data access performance in data grid, multiple replica of each data items are created in several data sites. However, due to the storage capacity constraints of data sites, replication costs and access pattern dynamicities it is hard to decide which data item from which data site should be replicated on which data site and when. Improving performance metrics such as response time and throughout are the main concern in many research works in the area of data grids. Numerous data replication policies are proposed for data replication in data grid and in order to evaluate these policies a stochastic simulation approach is employed. In this paper, a modelling framework, based on stochastic reward nets, is designed for modelling and evaluation of data replication policies in data grid. In this framework, separate sub-models are designed for workload generation, file replacement modules and the synchronization timers. Using the proposed framework, a wide family of replication policies can be specified and evaluated. Each replication policy can be simply specified by putting together the designed sub-models. We also employ the proposed framework to specify the best-client data replication method