| As a kind of data and resource management infrastructure,grid has become an important research area. Data grid is also a grid environment which is a large-scale,scalable structure to share and to manage storage resources and distributed data resources.Data designed in order to slove how to use distributed heterogeneous data resources conveniently and efficiently in data-intensive computing applications.Replication technology can improve data access efficiency,data relability and fault tolerance of system through making a copy of the data and storing it in a different location.Replica consistency derives from replication technology which leads to multiple replicas of the same data existing in the system.It is the key issues to ensure the users can get the correct data,and affect the accuracy and performance of the data grid system directly.Two respects are included in the present replica consistency models:first, most of them are based on the hierarchical mechanism of master replica and secondary replica;second, master replica is the update source and secondary replica is updated from master replica passively.Undoubtedly,master replica as the only update source must become the bottleneck of system performance in present replica consistency models.To slove this problem,this paper proposes a new decentralized replica consistency service model called DRCSM which doesn't make the master replica as the update source any more but make every secondary node undertake update duty equally, and using the timestamp and replica identification to accomplish replica consistency. DRCSM can solve system performance bottleneck in present models effectively. This paper explians DRCSM from modeling, analysis,messages-transmit mechanism and other aspects in detail,and simulates three models which are Aggressive copy, Lazy copy and DRCSM by the simulator OptorSim. The results show that DRCSM can reduce the mean job time, network load and replicating times effectively, having better performance. |
PDF Full Text Request