Performance analysis of data replication in data grid

Abstract

Data Grids are a fast emerging technology for managing huge amounts of distributed data. As Data Grids become more highly-anticipated by scientific communities, large data sets are being distributed and replicated to several sites, causing the issue of determining how replication can be implemented most effectively and efficiently. This can depend on many factors, such as physical characteristics of the resources, networks and storage devices. Replication can be utilized to reduce access latency, improve data locality, increase scalability, performance and robustness for Data Grid applications. Nowadays, Data replication is one of a major technique for lessen data access cost to Data Grid jobs. Two general data replication strategies or policies are full replication and partial replication. This research evaluates the effects of full replication strategy and partial replication strategy and compares their performances in a variety of Data Grid scenarios. A well-known Grid simulator, GridSim which provides essential building blocks for simulating Data Grid scenarios is chosen for the simulation of the replication strategies to compare and analyse the impact of data handling strategies before being built on the Data Grid. The results of the research and study show that the choice of data replication strategies can have a large effect on both job throughput and the overall performance of Data Grid.