Research On Replica Consistency In Data Grids

In data grid, it is desirable to create replicas of files for wide area distributed users. Data replication can be used to reduce access latency, improve data locality and/or increase robustness, scalability and performance for distributed applications. The introduction of the data replication technology means that there will be a large number of replicas produced, and how to maintain the consistency between these replicas becomes a new problem.The synchronization of replicas in databases and distributed object systems only consider some small matters, while there are a great amount of data in data grid system, an operation may change the entire contents of the file; on the other hand, the data in data grid regular distribute in the WAN, which means that the traditional algorithms to maintain replica consistency are not suitable for data grid. Based on above characteristics of data grid, the traditional transaction control method, copy control method and message queues method have certain limitations, not suitable for data grid environment, therefore some new replica consistency algorithms that suitable for gird environment have also been studied by many people.This thesis first describes the features and researches of the data grid, and introduces the major data management techniques, and then it analyzed the key technologies of replica management, and researches the consistency algorithms. Through analyzing the advantages and shortages of two replication coherence protocols (aggressive-copy and lazy-copy) in data grids, this paper presents an improved replication coherence architecture and a novel lazy_agg-copy algorithm. Lazy_agg-copy algorithm updates replicas randomly in multilayer architecture. Three algorithms which are Lazy-copy algorithm, aggressive-copy algorithm and lazy_agg-copy algorithm are simulated by a simulator- OptorSim. Simulation results show that lazy_agg-copy algorithm can find better balance in real-time, network load and bandwidth.